0d627a37da954ea68663797d4d45bffaf5fd89b5
[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 <asm/uaccess.h>
49 #include <asm/div64.h>
50 #include <asm/pgtable.h>
51
52 /* This magic number is used in glibc for posix shared memory */
53 #define TMPFS_MAGIC     0x01021994
54
55 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
56 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
57 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
58
59 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
60 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
61
62 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
63
64 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
65 #define SHMEM_PAGEIN     VM_READ
66 #define SHMEM_TRUNCATE   VM_WRITE
67
68 /* Definition to limit shmem_truncate's steps between cond_rescheds */
69 #define LATENCY_LIMIT    64
70
71 /* Pretend that each entry is of this size in directory's i_size */
72 #define BOGO_DIRENT_SIZE 20
73
74 /* Keep swapped page count in private field of indirect struct page */
75 #define nr_swapped              private
76
77 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
78 enum sgp_type {
79         SGP_QUICK,      /* don't try more than file page cache lookup */
80         SGP_READ,       /* don't exceed i_size, don't allocate page */
81         SGP_CACHE,      /* don't exceed i_size, may allocate page */
82         SGP_WRITE,      /* may exceed i_size, may allocate page */
83 };
84
85 static int shmem_getpage(struct inode *inode, unsigned long idx,
86                          struct page **pagep, enum sgp_type sgp, int *type);
87
88 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
89 {
90         /*
91          * The above definition of ENTRIES_PER_PAGE, and the use of
92          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
93          * might be reconsidered if it ever diverges from PAGE_SIZE.
94          */
95         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
96 }
97
98 static inline void shmem_dir_free(struct page *page)
99 {
100         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
101 }
102
103 static struct page **shmem_dir_map(struct page *page)
104 {
105         return (struct page **)kmap_atomic(page, KM_USER0);
106 }
107
108 static inline void shmem_dir_unmap(struct page **dir)
109 {
110         kunmap_atomic(dir, KM_USER0);
111 }
112
113 static swp_entry_t *shmem_swp_map(struct page *page)
114 {
115         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
116 }
117
118 static inline void shmem_swp_balance_unmap(void)
119 {
120         /*
121          * When passing a pointer to an i_direct entry, to code which
122          * also handles indirect entries and so will shmem_swp_unmap,
123          * we must arrange for the preempt count to remain in balance.
124          * What kmap_atomic of a lowmem page does depends on config
125          * and architecture, so pretend to kmap_atomic some lowmem page.
126          */
127         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
128 }
129
130 static inline void shmem_swp_unmap(swp_entry_t *entry)
131 {
132         kunmap_atomic(entry, KM_USER1);
133 }
134
135 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
136 {
137         return sb->s_fs_info;
138 }
139
140 /*
141  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
142  * for shared memory and for shared anonymous (/dev/zero) mappings
143  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
144  * consistent with the pre-accounting of private mappings ...
145  */
146 static inline int shmem_acct_size(unsigned long flags, loff_t size)
147 {
148         return (flags & VM_ACCOUNT)?
149                 security_vm_enough_memory(VM_ACCT(size)): 0;
150 }
151
152 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
153 {
154         if (flags & VM_ACCOUNT)
155                 vm_unacct_memory(VM_ACCT(size));
156 }
157
158 /*
159  * ... whereas tmpfs objects are accounted incrementally as
160  * pages are allocated, in order to allow huge sparse files.
161  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
162  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
163  */
164 static inline int shmem_acct_block(unsigned long flags)
165 {
166         return (flags & VM_ACCOUNT)?
167                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
168 }
169
170 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
171 {
172         if (!(flags & VM_ACCOUNT))
173                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
174 }
175
176 static struct super_operations shmem_ops;
177 static struct address_space_operations shmem_aops;
178 static struct file_operations shmem_file_operations;
179 static struct inode_operations shmem_inode_operations;
180 static struct inode_operations shmem_dir_inode_operations;
181 static struct vm_operations_struct shmem_vm_ops;
182
183 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
184         .ra_pages       = 0,    /* No readahead */
185         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
186         .unplug_io_fn   = default_unplug_io_fn,
187 };
188
189 static LIST_HEAD(shmem_swaplist);
190 static DEFINE_SPINLOCK(shmem_swaplist_lock);
191
192 static void shmem_free_blocks(struct inode *inode, long pages)
193 {
194         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
195         if (sbinfo->max_blocks) {
196                 spin_lock(&sbinfo->stat_lock);
197                 sbinfo->free_blocks += pages;
198                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
199                 spin_unlock(&sbinfo->stat_lock);
200         }
201 }
202
203 /*
204  * shmem_recalc_inode - recalculate the size of an inode
205  *
206  * @inode: inode to recalc
207  *
208  * We have to calculate the free blocks since the mm can drop
209  * undirtied hole pages behind our back.
210  *
211  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
212  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
213  *
214  * It has to be called with the spinlock held.
215  */
216 static void shmem_recalc_inode(struct inode *inode)
217 {
218         struct shmem_inode_info *info = SHMEM_I(inode);
219         long freed;
220
221         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
222         if (freed > 0) {
223                 info->alloced -= freed;
224                 shmem_unacct_blocks(info->flags, freed);
225                 shmem_free_blocks(inode, freed);
226         }
227 }
228
229 /*
230  * shmem_swp_entry - find the swap vector position in the info structure
231  *
232  * @info:  info structure for the inode
233  * @index: index of the page to find
234  * @page:  optional page to add to the structure. Has to be preset to
235  *         all zeros
236  *
237  * If there is no space allocated yet it will return NULL when
238  * page is NULL, else it will use the page for the needed block,
239  * setting it to NULL on return to indicate that it has been used.
240  *
241  * The swap vector is organized the following way:
242  *
243  * There are SHMEM_NR_DIRECT entries directly stored in the
244  * shmem_inode_info structure. So small files do not need an addional
245  * allocation.
246  *
247  * For pages with index > SHMEM_NR_DIRECT there is the pointer
248  * i_indirect which points to a page which holds in the first half
249  * doubly indirect blocks, in the second half triple indirect blocks:
250  *
251  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
252  * following layout (for SHMEM_NR_DIRECT == 16):
253  *
254  * i_indirect -> dir --> 16-19
255  *            |      +-> 20-23
256  *            |
257  *            +-->dir2 --> 24-27
258  *            |        +-> 28-31
259  *            |        +-> 32-35
260  *            |        +-> 36-39
261  *            |
262  *            +-->dir3 --> 40-43
263  *                     +-> 44-47
264  *                     +-> 48-51
265  *                     +-> 52-55
266  */
267 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
268 {
269         unsigned long offset;
270         struct page **dir;
271         struct page *subdir;
272
273         if (index < SHMEM_NR_DIRECT) {
274                 shmem_swp_balance_unmap();
275                 return info->i_direct+index;
276         }
277         if (!info->i_indirect) {
278                 if (page) {
279                         info->i_indirect = *page;
280                         *page = NULL;
281                 }
282                 return NULL;                    /* need another page */
283         }
284
285         index -= SHMEM_NR_DIRECT;
286         offset = index % ENTRIES_PER_PAGE;
287         index /= ENTRIES_PER_PAGE;
288         dir = shmem_dir_map(info->i_indirect);
289
290         if (index >= ENTRIES_PER_PAGE/2) {
291                 index -= ENTRIES_PER_PAGE/2;
292                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
293                 index %= ENTRIES_PER_PAGE;
294                 subdir = *dir;
295                 if (!subdir) {
296                         if (page) {
297                                 *dir = *page;
298                                 *page = NULL;
299                         }
300                         shmem_dir_unmap(dir);
301                         return NULL;            /* need another page */
302                 }
303                 shmem_dir_unmap(dir);
304                 dir = shmem_dir_map(subdir);
305         }
306
307         dir += index;
308         subdir = *dir;
309         if (!subdir) {
310                 if (!page || !(subdir = *page)) {
311                         shmem_dir_unmap(dir);
312                         return NULL;            /* need a page */
313                 }
314                 *dir = subdir;
315                 *page = NULL;
316         }
317         shmem_dir_unmap(dir);
318         return shmem_swp_map(subdir) + offset;
319 }
320
321 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
322 {
323         long incdec = value? 1: -1;
324
325         entry->val = value;
326         info->swapped += incdec;
327         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
328                 kmap_atomic_to_page(entry)->nr_swapped += incdec;
329 }
330
331 /*
332  * shmem_swp_alloc - get the position of the swap entry for the page.
333  *                   If it does not exist allocate the entry.
334  *
335  * @info:       info structure for the inode
336  * @index:      index of the page to find
337  * @sgp:        check and recheck i_size? skip allocation?
338  */
339 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
340 {
341         struct inode *inode = &info->vfs_inode;
342         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
343         struct page *page = NULL;
344         swp_entry_t *entry;
345
346         if (sgp != SGP_WRITE &&
347             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
348                 return ERR_PTR(-EINVAL);
349
350         while (!(entry = shmem_swp_entry(info, index, &page))) {
351                 if (sgp == SGP_READ)
352                         return shmem_swp_map(ZERO_PAGE(0));
353                 /*
354                  * Test free_blocks against 1 not 0, since we have 1 data
355                  * page (and perhaps indirect index pages) yet to allocate:
356                  * a waste to allocate index if we cannot allocate data.
357                  */
358                 if (sbinfo->max_blocks) {
359                         spin_lock(&sbinfo->stat_lock);
360                         if (sbinfo->free_blocks <= 1) {
361                                 spin_unlock(&sbinfo->stat_lock);
362                                 return ERR_PTR(-ENOSPC);
363                         }
364                         sbinfo->free_blocks--;
365                         inode->i_blocks += BLOCKS_PER_PAGE;
366                         spin_unlock(&sbinfo->stat_lock);
367                 }
368
369                 spin_unlock(&info->lock);
370                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
371                 if (page) {
372                         page->nr_swapped = 0;
373                 }
374                 spin_lock(&info->lock);
375
376                 if (!page) {
377                         shmem_free_blocks(inode, 1);
378                         return ERR_PTR(-ENOMEM);
379                 }
380                 if (sgp != SGP_WRITE &&
381                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
382                         entry = ERR_PTR(-EINVAL);
383                         break;
384                 }
385                 if (info->next_index <= index)
386                         info->next_index = index + 1;
387         }
388         if (page) {
389                 /* another task gave its page, or truncated the file */
390                 shmem_free_blocks(inode, 1);
391                 shmem_dir_free(page);
392         }
393         if (info->next_index <= index && !IS_ERR(entry))
394                 info->next_index = index + 1;
395         return entry;
396 }
397
398 /*
399  * shmem_free_swp - free some swap entries in a directory
400  *
401  * @dir:   pointer to the directory
402  * @edir:  pointer after last entry of the directory
403  */
404 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
405 {
406         swp_entry_t *ptr;
407         int freed = 0;
408
409         for (ptr = dir; ptr < edir; ptr++) {
410                 if (ptr->val) {
411                         free_swap_and_cache(*ptr);
412                         *ptr = (swp_entry_t){0};
413                         freed++;
414                 }
415         }
416         return freed;
417 }
418
419 static int shmem_map_and_free_swp(struct page *subdir,
420                 int offset, int limit, struct page ***dir)
421 {
422         swp_entry_t *ptr;
423         int freed = 0;
424
425         ptr = shmem_swp_map(subdir);
426         for (; offset < limit; offset += LATENCY_LIMIT) {
427                 int size = limit - offset;
428                 if (size > LATENCY_LIMIT)
429                         size = LATENCY_LIMIT;
430                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
431                 if (need_resched()) {
432                         shmem_swp_unmap(ptr);
433                         if (*dir) {
434                                 shmem_dir_unmap(*dir);
435                                 *dir = NULL;
436                         }
437                         cond_resched();
438                         ptr = shmem_swp_map(subdir);
439                 }
440         }
441         shmem_swp_unmap(ptr);
442         return freed;
443 }
444
445 static void shmem_free_pages(struct list_head *next)
446 {
447         struct page *page;
448         int freed = 0;
449
450         do {
451                 page = container_of(next, struct page, lru);
452                 next = next->next;
453                 shmem_dir_free(page);
454                 freed++;
455                 if (freed >= LATENCY_LIMIT) {
456                         cond_resched();
457                         freed = 0;
458                 }
459         } while (next);
460 }
461
462 static void shmem_truncate(struct inode *inode)
463 {
464         struct shmem_inode_info *info = SHMEM_I(inode);
465         unsigned long idx;
466         unsigned long size;
467         unsigned long limit;
468         unsigned long stage;
469         unsigned long diroff;
470         struct page **dir;
471         struct page *topdir;
472         struct page *middir;
473         struct page *subdir;
474         swp_entry_t *ptr;
475         LIST_HEAD(pages_to_free);
476         long nr_pages_to_free = 0;
477         long nr_swaps_freed = 0;
478         int offset;
479         int freed;
480
481         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
482         idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
483         if (idx >= info->next_index)
484                 return;
485
486         spin_lock(&info->lock);
487         info->flags |= SHMEM_TRUNCATE;
488         limit = info->next_index;
489         info->next_index = idx;
490         topdir = info->i_indirect;
491         if (topdir && idx <= SHMEM_NR_DIRECT) {
492                 info->i_indirect = NULL;
493                 nr_pages_to_free++;
494                 list_add(&topdir->lru, &pages_to_free);
495         }
496         spin_unlock(&info->lock);
497
498         if (info->swapped && idx < SHMEM_NR_DIRECT) {
499                 ptr = info->i_direct;
500                 size = limit;
501                 if (size > SHMEM_NR_DIRECT)
502                         size = SHMEM_NR_DIRECT;
503                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
504         }
505         if (!topdir)
506                 goto done2;
507
508         BUG_ON(limit <= SHMEM_NR_DIRECT);
509         limit -= SHMEM_NR_DIRECT;
510         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
511         offset = idx % ENTRIES_PER_PAGE;
512         idx -= offset;
513
514         dir = shmem_dir_map(topdir);
515         stage = ENTRIES_PER_PAGEPAGE/2;
516         if (idx < ENTRIES_PER_PAGEPAGE/2) {
517                 middir = topdir;
518                 diroff = idx/ENTRIES_PER_PAGE;
519         } else {
520                 dir += ENTRIES_PER_PAGE/2;
521                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
522                 while (stage <= idx)
523                         stage += ENTRIES_PER_PAGEPAGE;
524                 middir = *dir;
525                 if (*dir) {
526                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
527                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
528                         if (!diroff && !offset) {
529                                 *dir = NULL;
530                                 nr_pages_to_free++;
531                                 list_add(&middir->lru, &pages_to_free);
532                         }
533                         shmem_dir_unmap(dir);
534                         dir = shmem_dir_map(middir);
535                 } else {
536                         diroff = 0;
537                         offset = 0;
538                         idx = stage;
539                 }
540         }
541
542         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
543                 if (unlikely(idx == stage)) {
544                         shmem_dir_unmap(dir);
545                         dir = shmem_dir_map(topdir) +
546                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
547                         while (!*dir) {
548                                 dir++;
549                                 idx += ENTRIES_PER_PAGEPAGE;
550                                 if (idx >= limit)
551                                         goto done1;
552                         }
553                         stage = idx + ENTRIES_PER_PAGEPAGE;
554                         middir = *dir;
555                         *dir = NULL;
556                         nr_pages_to_free++;
557                         list_add(&middir->lru, &pages_to_free);
558                         shmem_dir_unmap(dir);
559                         cond_resched();
560                         dir = shmem_dir_map(middir);
561                         diroff = 0;
562                 }
563                 subdir = dir[diroff];
564                 if (subdir && subdir->nr_swapped) {
565                         size = limit - idx;
566                         if (size > ENTRIES_PER_PAGE)
567                                 size = ENTRIES_PER_PAGE;
568                         freed = shmem_map_and_free_swp(subdir,
569                                                 offset, size, &dir);
570                         if (!dir)
571                                 dir = shmem_dir_map(middir);
572                         nr_swaps_freed += freed;
573                         if (offset)
574                                 spin_lock(&info->lock);
575                         subdir->nr_swapped -= freed;
576                         if (offset)
577                                 spin_unlock(&info->lock);
578                         BUG_ON(subdir->nr_swapped > offset);
579                 }
580                 if (offset)
581                         offset = 0;
582                 else if (subdir) {
583                         dir[diroff] = NULL;
584                         nr_pages_to_free++;
585                         list_add(&subdir->lru, &pages_to_free);
586                 }
587         }
588 done1:
589         shmem_dir_unmap(dir);
590 done2:
591         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
592                 /*
593                  * Call truncate_inode_pages again: racing shmem_unuse_inode
594                  * may have swizzled a page in from swap since vmtruncate or
595                  * generic_delete_inode did it, before we lowered next_index.
596                  * Also, though shmem_getpage checks i_size before adding to
597                  * cache, no recheck after: so fix the narrow window there too.
598                  */
599                 truncate_inode_pages(inode->i_mapping, inode->i_size);
600         }
601
602         spin_lock(&info->lock);
603         info->flags &= ~SHMEM_TRUNCATE;
604         info->swapped -= nr_swaps_freed;
605         if (nr_pages_to_free)
606                 shmem_free_blocks(inode, nr_pages_to_free);
607         shmem_recalc_inode(inode);
608         spin_unlock(&info->lock);
609
610         /*
611          * Empty swap vector directory pages to be freed?
612          */
613         if (!list_empty(&pages_to_free)) {
614                 pages_to_free.prev->next = NULL;
615                 shmem_free_pages(pages_to_free.next);
616         }
617 }
618
619 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
620 {
621         struct inode *inode = dentry->d_inode;
622         struct page *page = NULL;
623         int error;
624
625         if (attr->ia_valid & ATTR_SIZE) {
626                 if (attr->ia_size < inode->i_size) {
627                         /*
628                          * If truncating down to a partial page, then
629                          * if that page is already allocated, hold it
630                          * in memory until the truncation is over, so
631                          * truncate_partial_page cannnot miss it were
632                          * it assigned to swap.
633                          */
634                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
635                                 (void) shmem_getpage(inode,
636                                         attr->ia_size>>PAGE_CACHE_SHIFT,
637                                                 &page, SGP_READ, NULL);
638                         }
639                         /*
640                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
641                          * detect if any pages might have been added to cache
642                          * after truncate_inode_pages.  But we needn't bother
643                          * if it's being fully truncated to zero-length: the
644                          * nrpages check is efficient enough in that case.
645                          */
646                         if (attr->ia_size) {
647                                 struct shmem_inode_info *info = SHMEM_I(inode);
648                                 spin_lock(&info->lock);
649                                 info->flags &= ~SHMEM_PAGEIN;
650                                 spin_unlock(&info->lock);
651                         }
652                 }
653         }
654
655         error = inode_change_ok(inode, attr);
656         if (!error)
657                 error = inode_setattr(inode, attr);
658         if (page)
659                 page_cache_release(page);
660         return error;
661 }
662
663 static void shmem_delete_inode(struct inode *inode)
664 {
665         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
666         struct shmem_inode_info *info = SHMEM_I(inode);
667
668         if (inode->i_op->truncate == shmem_truncate) {
669                 truncate_inode_pages(inode->i_mapping, 0);
670                 shmem_unacct_size(info->flags, inode->i_size);
671                 inode->i_size = 0;
672                 shmem_truncate(inode);
673                 if (!list_empty(&info->swaplist)) {
674                         spin_lock(&shmem_swaplist_lock);
675                         list_del_init(&info->swaplist);
676                         spin_unlock(&shmem_swaplist_lock);
677                 }
678         }
679         BUG_ON(inode->i_blocks);
680         if (sbinfo->max_inodes) {
681                 spin_lock(&sbinfo->stat_lock);
682                 sbinfo->free_inodes++;
683                 spin_unlock(&sbinfo->stat_lock);
684         }
685         clear_inode(inode);
686 }
687
688 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
689 {
690         swp_entry_t *ptr;
691
692         for (ptr = dir; ptr < edir; ptr++) {
693                 if (ptr->val == entry.val)
694                         return ptr - dir;
695         }
696         return -1;
697 }
698
699 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
700 {
701         struct inode *inode;
702         unsigned long idx;
703         unsigned long size;
704         unsigned long limit;
705         unsigned long stage;
706         struct page **dir;
707         struct page *subdir;
708         swp_entry_t *ptr;
709         int offset;
710
711         idx = 0;
712         ptr = info->i_direct;
713         spin_lock(&info->lock);
714         limit = info->next_index;
715         size = limit;
716         if (size > SHMEM_NR_DIRECT)
717                 size = SHMEM_NR_DIRECT;
718         offset = shmem_find_swp(entry, ptr, ptr+size);
719         if (offset >= 0) {
720                 shmem_swp_balance_unmap();
721                 goto found;
722         }
723         if (!info->i_indirect)
724                 goto lost2;
725
726         dir = shmem_dir_map(info->i_indirect);
727         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
728
729         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
730                 if (unlikely(idx == stage)) {
731                         shmem_dir_unmap(dir-1);
732                         dir = shmem_dir_map(info->i_indirect) +
733                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
734                         while (!*dir) {
735                                 dir++;
736                                 idx += ENTRIES_PER_PAGEPAGE;
737                                 if (idx >= limit)
738                                         goto lost1;
739                         }
740                         stage = idx + ENTRIES_PER_PAGEPAGE;
741                         subdir = *dir;
742                         shmem_dir_unmap(dir);
743                         dir = shmem_dir_map(subdir);
744                 }
745                 subdir = *dir;
746                 if (subdir && subdir->nr_swapped) {
747                         ptr = shmem_swp_map(subdir);
748                         size = limit - idx;
749                         if (size > ENTRIES_PER_PAGE)
750                                 size = ENTRIES_PER_PAGE;
751                         offset = shmem_find_swp(entry, ptr, ptr+size);
752                         if (offset >= 0) {
753                                 shmem_dir_unmap(dir);
754                                 goto found;
755                         }
756                         shmem_swp_unmap(ptr);
757                 }
758         }
759 lost1:
760         shmem_dir_unmap(dir-1);
761 lost2:
762         spin_unlock(&info->lock);
763         return 0;
764 found:
765         idx += offset;
766         inode = &info->vfs_inode;
767         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
768                 info->flags |= SHMEM_PAGEIN;
769                 shmem_swp_set(info, ptr + offset, 0);
770         }
771         shmem_swp_unmap(ptr);
772         spin_unlock(&info->lock);
773         /*
774          * Decrement swap count even when the entry is left behind:
775          * try_to_unuse will skip over mms, then reincrement count.
776          */
777         swap_free(entry);
778         return 1;
779 }
780
781 /*
782  * shmem_unuse() search for an eventually swapped out shmem page.
783  */
784 int shmem_unuse(swp_entry_t entry, struct page *page)
785 {
786         struct list_head *p, *next;
787         struct shmem_inode_info *info;
788         int found = 0;
789
790         spin_lock(&shmem_swaplist_lock);
791         list_for_each_safe(p, next, &shmem_swaplist) {
792                 info = list_entry(p, struct shmem_inode_info, swaplist);
793                 if (!info->swapped)
794                         list_del_init(&info->swaplist);
795                 else if (shmem_unuse_inode(info, entry, page)) {
796                         /* move head to start search for next from here */
797                         list_move_tail(&shmem_swaplist, &info->swaplist);
798                         found = 1;
799                         break;
800                 }
801         }
802         spin_unlock(&shmem_swaplist_lock);
803         return found;
804 }
805
806 /*
807  * Move the page from the page cache to the swap cache.
808  */
809 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
810 {
811         struct shmem_inode_info *info;
812         swp_entry_t *entry, swap;
813         struct address_space *mapping;
814         unsigned long index;
815         struct inode *inode;
816
817         BUG_ON(!PageLocked(page));
818         BUG_ON(page_mapped(page));
819
820         mapping = page->mapping;
821         index = page->index;
822         inode = mapping->host;
823         info = SHMEM_I(inode);
824         if (info->flags & VM_LOCKED)
825                 goto redirty;
826         swap = get_swap_page();
827         if (!swap.val)
828                 goto redirty;
829
830         spin_lock(&info->lock);
831         shmem_recalc_inode(inode);
832         if (index >= info->next_index) {
833                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
834                 goto unlock;
835         }
836         entry = shmem_swp_entry(info, index, NULL);
837         BUG_ON(!entry);
838         BUG_ON(entry->val);
839
840         if (move_to_swap_cache(page, swap) == 0) {
841                 shmem_swp_set(info, entry, swap.val);
842                 shmem_swp_unmap(entry);
843                 spin_unlock(&info->lock);
844                 if (list_empty(&info->swaplist)) {
845                         spin_lock(&shmem_swaplist_lock);
846                         /* move instead of add in case we're racing */
847                         list_move_tail(&info->swaplist, &shmem_swaplist);
848                         spin_unlock(&shmem_swaplist_lock);
849                 }
850                 unlock_page(page);
851                 return 0;
852         }
853
854         shmem_swp_unmap(entry);
855 unlock:
856         spin_unlock(&info->lock);
857         swap_free(swap);
858 redirty:
859         set_page_dirty(page);
860         return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
861 }
862
863 #ifdef CONFIG_NUMA
864 static struct page *shmem_swapin_async(struct shared_policy *p,
865                                        swp_entry_t entry, unsigned long idx)
866 {
867         struct page *page;
868         struct vm_area_struct pvma;
869
870         /* Create a pseudo vma that just contains the policy */
871         memset(&pvma, 0, sizeof(struct vm_area_struct));
872         pvma.vm_end = PAGE_SIZE;
873         pvma.vm_pgoff = idx;
874         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
875         page = read_swap_cache_async(entry, &pvma, 0);
876         mpol_free(pvma.vm_policy);
877         return page;
878 }
879
880 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
881                           unsigned long idx)
882 {
883         struct shared_policy *p = &info->policy;
884         int i, num;
885         struct page *page;
886         unsigned long offset;
887
888         num = valid_swaphandles(entry, &offset);
889         for (i = 0; i < num; offset++, i++) {
890                 page = shmem_swapin_async(p,
891                                 swp_entry(swp_type(entry), offset), idx);
892                 if (!page)
893                         break;
894                 page_cache_release(page);
895         }
896         lru_add_drain();        /* Push any new pages onto the LRU now */
897         return shmem_swapin_async(p, entry, idx);
898 }
899
900 static struct page *
901 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
902                  unsigned long idx)
903 {
904         struct vm_area_struct pvma;
905         struct page *page;
906
907         memset(&pvma, 0, sizeof(struct vm_area_struct));
908         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
909         pvma.vm_pgoff = idx;
910         pvma.vm_end = PAGE_SIZE;
911         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
912         mpol_free(pvma.vm_policy);
913         return page;
914 }
915 #else
916 static inline struct page *
917 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
918 {
919         swapin_readahead(entry, 0, NULL);
920         return read_swap_cache_async(entry, NULL, 0);
921 }
922
923 static inline struct page *
924 shmem_alloc_page(unsigned int __nocast gfp,struct shmem_inode_info *info,
925                                  unsigned long idx)
926 {
927         return alloc_page(gfp | __GFP_ZERO);
928 }
929 #endif
930
931 /*
932  * shmem_getpage - either get the page from swap or allocate a new one
933  *
934  * If we allocate a new one we do not mark it dirty. That's up to the
935  * vm. If we swap it in we mark it dirty since we also free the swap
936  * entry since a page cannot live in both the swap and page cache
937  */
938 static int shmem_getpage(struct inode *inode, unsigned long idx,
939                         struct page **pagep, enum sgp_type sgp, int *type)
940 {
941         struct address_space *mapping = inode->i_mapping;
942         struct shmem_inode_info *info = SHMEM_I(inode);
943         struct shmem_sb_info *sbinfo;
944         struct page *filepage = *pagep;
945         struct page *swappage;
946         swp_entry_t *entry;
947         swp_entry_t swap;
948         int error;
949
950         if (idx >= SHMEM_MAX_INDEX)
951                 return -EFBIG;
952         /*
953          * Normally, filepage is NULL on entry, and either found
954          * uptodate immediately, or allocated and zeroed, or read
955          * in under swappage, which is then assigned to filepage.
956          * But shmem_prepare_write passes in a locked filepage,
957          * which may be found not uptodate by other callers too,
958          * and may need to be copied from the swappage read in.
959          */
960 repeat:
961         if (!filepage)
962                 filepage = find_lock_page(mapping, idx);
963         if (filepage && PageUptodate(filepage))
964                 goto done;
965         error = 0;
966         if (sgp == SGP_QUICK)
967                 goto failed;
968
969         spin_lock(&info->lock);
970         shmem_recalc_inode(inode);
971         entry = shmem_swp_alloc(info, idx, sgp);
972         if (IS_ERR(entry)) {
973                 spin_unlock(&info->lock);
974                 error = PTR_ERR(entry);
975                 goto failed;
976         }
977         swap = *entry;
978
979         if (swap.val) {
980                 /* Look it up and read it in.. */
981                 swappage = lookup_swap_cache(swap);
982                 if (!swappage) {
983                         shmem_swp_unmap(entry);
984                         spin_unlock(&info->lock);
985                         /* here we actually do the io */
986                         if (type && *type == VM_FAULT_MINOR) {
987                                 inc_page_state(pgmajfault);
988                                 *type = VM_FAULT_MAJOR;
989                         }
990                         swappage = shmem_swapin(info, swap, idx);
991                         if (!swappage) {
992                                 spin_lock(&info->lock);
993                                 entry = shmem_swp_alloc(info, idx, sgp);
994                                 if (IS_ERR(entry))
995                                         error = PTR_ERR(entry);
996                                 else {
997                                         if (entry->val == swap.val)
998                                                 error = -ENOMEM;
999                                         shmem_swp_unmap(entry);
1000                                 }
1001                                 spin_unlock(&info->lock);
1002                                 if (error)
1003                                         goto failed;
1004                                 goto repeat;
1005                         }
1006                         wait_on_page_locked(swappage);
1007                         page_cache_release(swappage);
1008                         goto repeat;
1009                 }
1010
1011                 /* We have to do this with page locked to prevent races */
1012                 if (TestSetPageLocked(swappage)) {
1013                         shmem_swp_unmap(entry);
1014                         spin_unlock(&info->lock);
1015                         wait_on_page_locked(swappage);
1016                         page_cache_release(swappage);
1017                         goto repeat;
1018                 }
1019                 if (PageWriteback(swappage)) {
1020                         shmem_swp_unmap(entry);
1021                         spin_unlock(&info->lock);
1022                         wait_on_page_writeback(swappage);
1023                         unlock_page(swappage);
1024                         page_cache_release(swappage);
1025                         goto repeat;
1026                 }
1027                 if (!PageUptodate(swappage)) {
1028                         shmem_swp_unmap(entry);
1029                         spin_unlock(&info->lock);
1030                         unlock_page(swappage);
1031                         page_cache_release(swappage);
1032                         error = -EIO;
1033                         goto failed;
1034                 }
1035
1036                 if (filepage) {
1037                         shmem_swp_set(info, entry, 0);
1038                         shmem_swp_unmap(entry);
1039                         delete_from_swap_cache(swappage);
1040                         spin_unlock(&info->lock);
1041                         copy_highpage(filepage, swappage);
1042                         unlock_page(swappage);
1043                         page_cache_release(swappage);
1044                         flush_dcache_page(filepage);
1045                         SetPageUptodate(filepage);
1046                         set_page_dirty(filepage);
1047                         swap_free(swap);
1048                 } else if (!(error = move_from_swap_cache(
1049                                 swappage, idx, mapping))) {
1050                         info->flags |= SHMEM_PAGEIN;
1051                         shmem_swp_set(info, entry, 0);
1052                         shmem_swp_unmap(entry);
1053                         spin_unlock(&info->lock);
1054                         filepage = swappage;
1055                         swap_free(swap);
1056                 } else {
1057                         shmem_swp_unmap(entry);
1058                         spin_unlock(&info->lock);
1059                         unlock_page(swappage);
1060                         page_cache_release(swappage);
1061                         if (error == -ENOMEM) {
1062                                 /* let kswapd refresh zone for GFP_ATOMICs */
1063                                 blk_congestion_wait(WRITE, HZ/50);
1064                         }
1065                         goto repeat;
1066                 }
1067         } else if (sgp == SGP_READ && !filepage) {
1068                 shmem_swp_unmap(entry);
1069                 filepage = find_get_page(mapping, idx);
1070                 if (filepage &&
1071                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1072                         spin_unlock(&info->lock);
1073                         wait_on_page_locked(filepage);
1074                         page_cache_release(filepage);
1075                         filepage = NULL;
1076                         goto repeat;
1077                 }
1078                 spin_unlock(&info->lock);
1079         } else {
1080                 shmem_swp_unmap(entry);
1081                 sbinfo = SHMEM_SB(inode->i_sb);
1082                 if (sbinfo->max_blocks) {
1083                         spin_lock(&sbinfo->stat_lock);
1084                         if (sbinfo->free_blocks == 0 ||
1085                             shmem_acct_block(info->flags)) {
1086                                 spin_unlock(&sbinfo->stat_lock);
1087                                 spin_unlock(&info->lock);
1088                                 error = -ENOSPC;
1089                                 goto failed;
1090                         }
1091                         sbinfo->free_blocks--;
1092                         inode->i_blocks += BLOCKS_PER_PAGE;
1093                         spin_unlock(&sbinfo->stat_lock);
1094                 } else if (shmem_acct_block(info->flags)) {
1095                         spin_unlock(&info->lock);
1096                         error = -ENOSPC;
1097                         goto failed;
1098                 }
1099
1100                 if (!filepage) {
1101                         spin_unlock(&info->lock);
1102                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1103                                                     info,
1104                                                     idx);
1105                         if (!filepage) {
1106                                 shmem_unacct_blocks(info->flags, 1);
1107                                 shmem_free_blocks(inode, 1);
1108                                 error = -ENOMEM;
1109                                 goto failed;
1110                         }
1111
1112                         spin_lock(&info->lock);
1113                         entry = shmem_swp_alloc(info, idx, sgp);
1114                         if (IS_ERR(entry))
1115                                 error = PTR_ERR(entry);
1116                         else {
1117                                 swap = *entry;
1118                                 shmem_swp_unmap(entry);
1119                         }
1120                         if (error || swap.val || 0 != add_to_page_cache_lru(
1121                                         filepage, mapping, idx, GFP_ATOMIC)) {
1122                                 spin_unlock(&info->lock);
1123                                 page_cache_release(filepage);
1124                                 shmem_unacct_blocks(info->flags, 1);
1125                                 shmem_free_blocks(inode, 1);
1126                                 filepage = NULL;
1127                                 if (error)
1128                                         goto failed;
1129                                 goto repeat;
1130                         }
1131                         info->flags |= SHMEM_PAGEIN;
1132                 }
1133
1134                 info->alloced++;
1135                 spin_unlock(&info->lock);
1136                 flush_dcache_page(filepage);
1137                 SetPageUptodate(filepage);
1138         }
1139 done:
1140         if (*pagep != filepage) {
1141                 unlock_page(filepage);
1142                 *pagep = filepage;
1143         }
1144         return 0;
1145
1146 failed:
1147         if (*pagep != filepage) {
1148                 unlock_page(filepage);
1149                 page_cache_release(filepage);
1150         }
1151         return error;
1152 }
1153
1154 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1155 {
1156         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1157         struct page *page = NULL;
1158         unsigned long idx;
1159         int error;
1160
1161         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1162         idx += vma->vm_pgoff;
1163         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1164         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1165                 return NOPAGE_SIGBUS;
1166
1167         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1168         if (error)
1169                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1170
1171         mark_page_accessed(page);
1172         return page;
1173 }
1174
1175 static int shmem_populate(struct vm_area_struct *vma,
1176         unsigned long addr, unsigned long len,
1177         pgprot_t prot, unsigned long pgoff, int nonblock)
1178 {
1179         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1180         struct mm_struct *mm = vma->vm_mm;
1181         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1182         unsigned long size;
1183
1184         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1185         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1186                 return -EINVAL;
1187
1188         while ((long) len > 0) {
1189                 struct page *page = NULL;
1190                 int err;
1191                 /*
1192                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1193                  */
1194                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1195                 if (err)
1196                         return err;
1197                 /* Page may still be null, but only if nonblock was set. */
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 {
1206                         /* No page was found just because we can't read it in
1207                          * now (being here implies nonblock != 0), but the page
1208                          * may exist, so set the PTE to fault it in later. */
1209                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1210                         if (err)
1211                                 return err;
1212                 }
1213
1214                 len -= PAGE_SIZE;
1215                 addr += PAGE_SIZE;
1216                 pgoff++;
1217         }
1218         return 0;
1219 }
1220
1221 #ifdef CONFIG_NUMA
1222 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1223 {
1224         struct inode *i = vma->vm_file->f_dentry->d_inode;
1225         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1226 }
1227
1228 struct mempolicy *
1229 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1230 {
1231         struct inode *i = vma->vm_file->f_dentry->d_inode;
1232         unsigned long idx;
1233
1234         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1235         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1236 }
1237 #endif
1238
1239 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1240 {
1241         struct inode *inode = file->f_dentry->d_inode;
1242         struct shmem_inode_info *info = SHMEM_I(inode);
1243         int retval = -ENOMEM;
1244
1245         spin_lock(&info->lock);
1246         if (lock && !(info->flags & VM_LOCKED)) {
1247                 if (!user_shm_lock(inode->i_size, user))
1248                         goto out_nomem;
1249                 info->flags |= VM_LOCKED;
1250         }
1251         if (!lock && (info->flags & VM_LOCKED) && user) {
1252                 user_shm_unlock(inode->i_size, user);
1253                 info->flags &= ~VM_LOCKED;
1254         }
1255         retval = 0;
1256 out_nomem:
1257         spin_unlock(&info->lock);
1258         return retval;
1259 }
1260
1261 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1262 {
1263         file_accessed(file);
1264         vma->vm_ops = &shmem_vm_ops;
1265         return 0;
1266 }
1267
1268 static struct inode *
1269 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1270 {
1271         struct inode *inode;
1272         struct shmem_inode_info *info;
1273         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1274
1275         if (sbinfo->max_inodes) {
1276                 spin_lock(&sbinfo->stat_lock);
1277                 if (!sbinfo->free_inodes) {
1278                         spin_unlock(&sbinfo->stat_lock);
1279                         return NULL;
1280                 }
1281                 sbinfo->free_inodes--;
1282                 spin_unlock(&sbinfo->stat_lock);
1283         }
1284
1285         inode = new_inode(sb);
1286         if (inode) {
1287                 inode->i_mode = mode;
1288                 inode->i_uid = current->fsuid;
1289                 inode->i_gid = current->fsgid;
1290                 inode->i_blksize = PAGE_CACHE_SIZE;
1291                 inode->i_blocks = 0;
1292                 inode->i_mapping->a_ops = &shmem_aops;
1293                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1294                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1295                 info = SHMEM_I(inode);
1296                 memset(info, 0, (char *)inode - (char *)info);
1297                 spin_lock_init(&info->lock);
1298                 INIT_LIST_HEAD(&info->swaplist);
1299
1300                 switch (mode & S_IFMT) {
1301                 default:
1302                         init_special_inode(inode, mode, dev);
1303                         break;
1304                 case S_IFREG:
1305                         inode->i_op = &shmem_inode_operations;
1306                         inode->i_fop = &shmem_file_operations;
1307                         mpol_shared_policy_init(&info->policy);
1308                         break;
1309                 case S_IFDIR:
1310                         inode->i_nlink++;
1311                         /* Some things misbehave if size == 0 on a directory */
1312                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1313                         inode->i_op = &shmem_dir_inode_operations;
1314                         inode->i_fop = &simple_dir_operations;
1315                         break;
1316                 case S_IFLNK:
1317                         /*
1318                          * Must not load anything in the rbtree,
1319                          * mpol_free_shared_policy will not be called.
1320                          */
1321                         mpol_shared_policy_init(&info->policy);
1322                         break;
1323                 }
1324         } else if (sbinfo->max_inodes) {
1325                 spin_lock(&sbinfo->stat_lock);
1326                 sbinfo->free_inodes++;
1327                 spin_unlock(&sbinfo->stat_lock);
1328         }
1329         return inode;
1330 }
1331
1332 #ifdef CONFIG_TMPFS
1333 static struct inode_operations shmem_symlink_inode_operations;
1334 static struct inode_operations shmem_symlink_inline_operations;
1335
1336 /*
1337  * Normally tmpfs makes no use of shmem_prepare_write, but it
1338  * lets a tmpfs file be used read-write below the loop driver.
1339  */
1340 static int
1341 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1342 {
1343         struct inode *inode = page->mapping->host;
1344         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1345 }
1346
1347 static ssize_t
1348 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1349 {
1350         struct inode    *inode = file->f_dentry->d_inode;
1351         loff_t          pos;
1352         unsigned long   written;
1353         ssize_t         err;
1354
1355         if ((ssize_t) count < 0)
1356                 return -EINVAL;
1357
1358         if (!access_ok(VERIFY_READ, buf, count))
1359                 return -EFAULT;
1360
1361         down(&inode->i_sem);
1362
1363         pos = *ppos;
1364         written = 0;
1365
1366         err = generic_write_checks(file, &pos, &count, 0);
1367         if (err || !count)
1368                 goto out;
1369
1370         err = remove_suid(file->f_dentry);
1371         if (err)
1372                 goto out;
1373
1374         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1375
1376         do {
1377                 struct page *page = NULL;
1378                 unsigned long bytes, index, offset;
1379                 char *kaddr;
1380                 int left;
1381
1382                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1383                 index = pos >> PAGE_CACHE_SHIFT;
1384                 bytes = PAGE_CACHE_SIZE - offset;
1385                 if (bytes > count)
1386                         bytes = count;
1387
1388                 /*
1389                  * We don't hold page lock across copy from user -
1390                  * what would it guard against? - so no deadlock here.
1391                  * But it still may be a good idea to prefault below.
1392                  */
1393
1394                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1395                 if (err)
1396                         break;
1397
1398                 left = bytes;
1399                 if (PageHighMem(page)) {
1400                         volatile unsigned char dummy;
1401                         __get_user(dummy, buf);
1402                         __get_user(dummy, buf + bytes - 1);
1403
1404                         kaddr = kmap_atomic(page, KM_USER0);
1405                         left = __copy_from_user_inatomic(kaddr + offset,
1406                                                         buf, bytes);
1407                         kunmap_atomic(kaddr, KM_USER0);
1408                 }
1409                 if (left) {
1410                         kaddr = kmap(page);
1411                         left = __copy_from_user(kaddr + offset, buf, bytes);
1412                         kunmap(page);
1413                 }
1414
1415                 written += bytes;
1416                 count -= bytes;
1417                 pos += bytes;
1418                 buf += bytes;
1419                 if (pos > inode->i_size)
1420                         i_size_write(inode, pos);
1421
1422                 flush_dcache_page(page);
1423                 set_page_dirty(page);
1424                 mark_page_accessed(page);
1425                 page_cache_release(page);
1426
1427                 if (left) {
1428                         pos -= left;
1429                         written -= left;
1430                         err = -EFAULT;
1431                         break;
1432                 }
1433
1434                 /*
1435                  * Our dirty pages are not counted in nr_dirty,
1436                  * and we do not attempt to balance dirty pages.
1437                  */
1438
1439                 cond_resched();
1440         } while (count);
1441
1442         *ppos = pos;
1443         if (written)
1444                 err = written;
1445 out:
1446         up(&inode->i_sem);
1447         return err;
1448 }
1449
1450 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1451 {
1452         struct inode *inode = filp->f_dentry->d_inode;
1453         struct address_space *mapping = inode->i_mapping;
1454         unsigned long index, offset;
1455
1456         index = *ppos >> PAGE_CACHE_SHIFT;
1457         offset = *ppos & ~PAGE_CACHE_MASK;
1458
1459         for (;;) {
1460                 struct page *page = NULL;
1461                 unsigned long end_index, nr, ret;
1462                 loff_t i_size = i_size_read(inode);
1463
1464                 end_index = i_size >> PAGE_CACHE_SHIFT;
1465                 if (index > end_index)
1466                         break;
1467                 if (index == end_index) {
1468                         nr = i_size & ~PAGE_CACHE_MASK;
1469                         if (nr <= offset)
1470                                 break;
1471                 }
1472
1473                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1474                 if (desc->error) {
1475                         if (desc->error == -EINVAL)
1476                                 desc->error = 0;
1477                         break;
1478                 }
1479
1480                 /*
1481                  * We must evaluate after, since reads (unlike writes)
1482                  * are called without i_sem protection against truncate
1483                  */
1484                 nr = PAGE_CACHE_SIZE;
1485                 i_size = i_size_read(inode);
1486                 end_index = i_size >> PAGE_CACHE_SHIFT;
1487                 if (index == end_index) {
1488                         nr = i_size & ~PAGE_CACHE_MASK;
1489                         if (nr <= offset) {
1490                                 if (page)
1491                                         page_cache_release(page);
1492                                 break;
1493                         }
1494                 }
1495                 nr -= offset;
1496
1497                 if (page) {
1498                         /*
1499                          * If users can be writing to this page using arbitrary
1500                          * virtual addresses, take care about potential aliasing
1501                          * before reading the page on the kernel side.
1502                          */
1503                         if (mapping_writably_mapped(mapping))
1504                                 flush_dcache_page(page);
1505                         /*
1506                          * Mark the page accessed if we read the beginning.
1507                          */
1508                         if (!offset)
1509                                 mark_page_accessed(page);
1510                 } else
1511                         page = ZERO_PAGE(0);
1512
1513                 /*
1514                  * Ok, we have the page, and it's up-to-date, so
1515                  * now we can copy it to user space...
1516                  *
1517                  * The actor routine returns how many bytes were actually used..
1518                  * NOTE! This may not be the same as how much of a user buffer
1519                  * we filled up (we may be padding etc), so we can only update
1520                  * "pos" here (the actor routine has to update the user buffer
1521                  * pointers and the remaining count).
1522                  */
1523                 ret = actor(desc, page, offset, nr);
1524                 offset += ret;
1525                 index += offset >> PAGE_CACHE_SHIFT;
1526                 offset &= ~PAGE_CACHE_MASK;
1527
1528                 page_cache_release(page);
1529                 if (ret != nr || !desc->count)
1530                         break;
1531
1532                 cond_resched();
1533         }
1534
1535         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1536         file_accessed(filp);
1537 }
1538
1539 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1540 {
1541         read_descriptor_t desc;
1542
1543         if ((ssize_t) count < 0)
1544                 return -EINVAL;
1545         if (!access_ok(VERIFY_WRITE, buf, count))
1546                 return -EFAULT;
1547         if (!count)
1548                 return 0;
1549
1550         desc.written = 0;
1551         desc.count = count;
1552         desc.arg.buf = buf;
1553         desc.error = 0;
1554
1555         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1556         if (desc.written)
1557                 return desc.written;
1558         return desc.error;
1559 }
1560
1561 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1562                          size_t count, read_actor_t actor, void *target)
1563 {
1564         read_descriptor_t desc;
1565
1566         if (!count)
1567                 return 0;
1568
1569         desc.written = 0;
1570         desc.count = count;
1571         desc.arg.data = target;
1572         desc.error = 0;
1573
1574         do_shmem_file_read(in_file, ppos, &desc, actor);
1575         if (desc.written)
1576                 return desc.written;
1577         return desc.error;
1578 }
1579
1580 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1581 {
1582         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1583
1584         buf->f_type = TMPFS_MAGIC;
1585         buf->f_bsize = PAGE_CACHE_SIZE;
1586         buf->f_namelen = NAME_MAX;
1587         spin_lock(&sbinfo->stat_lock);
1588         if (sbinfo->max_blocks) {
1589                 buf->f_blocks = sbinfo->max_blocks;
1590                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1591         }
1592         if (sbinfo->max_inodes) {
1593                 buf->f_files = sbinfo->max_inodes;
1594                 buf->f_ffree = sbinfo->free_inodes;
1595         }
1596         /* else leave those fields 0 like simple_statfs */
1597         spin_unlock(&sbinfo->stat_lock);
1598         return 0;
1599 }
1600
1601 /*
1602  * File creation. Allocate an inode, and we're done..
1603  */
1604 static int
1605 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1606 {
1607         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1608         int error = -ENOSPC;
1609
1610         if (inode) {
1611                 if (dir->i_mode & S_ISGID) {
1612                         inode->i_gid = dir->i_gid;
1613                         if (S_ISDIR(mode))
1614                                 inode->i_mode |= S_ISGID;
1615                 }
1616                 dir->i_size += BOGO_DIRENT_SIZE;
1617                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1618                 d_instantiate(dentry, inode);
1619                 dget(dentry); /* Extra count - pin the dentry in core */
1620                 error = 0;
1621         }
1622         return error;
1623 }
1624
1625 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1626 {
1627         int error;
1628
1629         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1630                 return error;
1631         dir->i_nlink++;
1632         return 0;
1633 }
1634
1635 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1636                 struct nameidata *nd)
1637 {
1638         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1639 }
1640
1641 /*
1642  * Link a file..
1643  */
1644 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1645 {
1646         struct inode *inode = old_dentry->d_inode;
1647         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1648
1649         /*
1650          * No ordinary (disk based) filesystem counts links as inodes;
1651          * but each new link needs a new dentry, pinning lowmem, and
1652          * tmpfs dentries cannot be pruned until they are unlinked.
1653          */
1654         if (sbinfo->max_inodes) {
1655                 spin_lock(&sbinfo->stat_lock);
1656                 if (!sbinfo->free_inodes) {
1657                         spin_unlock(&sbinfo->stat_lock);
1658                         return -ENOSPC;
1659                 }
1660                 sbinfo->free_inodes--;
1661                 spin_unlock(&sbinfo->stat_lock);
1662         }
1663
1664         dir->i_size += BOGO_DIRENT_SIZE;
1665         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1666         inode->i_nlink++;
1667         atomic_inc(&inode->i_count);    /* New dentry reference */
1668         dget(dentry);           /* Extra pinning count for the created dentry */
1669         d_instantiate(dentry, inode);
1670         return 0;
1671 }
1672
1673 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1674 {
1675         struct inode *inode = dentry->d_inode;
1676
1677         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1678                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1679                 if (sbinfo->max_inodes) {
1680                         spin_lock(&sbinfo->stat_lock);
1681                         sbinfo->free_inodes++;
1682                         spin_unlock(&sbinfo->stat_lock);
1683                 }
1684         }
1685
1686         dir->i_size -= BOGO_DIRENT_SIZE;
1687         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1688         inode->i_nlink--;
1689         dput(dentry);   /* Undo the count from "create" - this does all the work */
1690         return 0;
1691 }
1692
1693 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1694 {
1695         if (!simple_empty(dentry))
1696                 return -ENOTEMPTY;
1697
1698         dir->i_nlink--;
1699         return shmem_unlink(dir, dentry);
1700 }
1701
1702 /*
1703  * The VFS layer already does all the dentry stuff for rename,
1704  * we just have to decrement the usage count for the target if
1705  * it exists so that the VFS layer correctly free's it when it
1706  * gets overwritten.
1707  */
1708 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1709 {
1710         struct inode *inode = old_dentry->d_inode;
1711         int they_are_dirs = S_ISDIR(inode->i_mode);
1712
1713         if (!simple_empty(new_dentry))
1714                 return -ENOTEMPTY;
1715
1716         if (new_dentry->d_inode) {
1717                 (void) shmem_unlink(new_dir, new_dentry);
1718                 if (they_are_dirs)
1719                         old_dir->i_nlink--;
1720         } else if (they_are_dirs) {
1721                 old_dir->i_nlink--;
1722                 new_dir->i_nlink++;
1723         }
1724
1725         old_dir->i_size -= BOGO_DIRENT_SIZE;
1726         new_dir->i_size += BOGO_DIRENT_SIZE;
1727         old_dir->i_ctime = old_dir->i_mtime =
1728         new_dir->i_ctime = new_dir->i_mtime =
1729         inode->i_ctime = CURRENT_TIME;
1730         return 0;
1731 }
1732
1733 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1734 {
1735         int error;
1736         int len;
1737         struct inode *inode;
1738         struct page *page = NULL;
1739         char *kaddr;
1740         struct shmem_inode_info *info;
1741
1742         len = strlen(symname) + 1;
1743         if (len > PAGE_CACHE_SIZE)
1744                 return -ENAMETOOLONG;
1745
1746         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1747         if (!inode)
1748                 return -ENOSPC;
1749
1750         info = SHMEM_I(inode);
1751         inode->i_size = len-1;
1752         if (len <= (char *)inode - (char *)info) {
1753                 /* do it inline */
1754                 memcpy(info, symname, len);
1755                 inode->i_op = &shmem_symlink_inline_operations;
1756         } else {
1757                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1758                 if (error) {
1759                         iput(inode);
1760                         return error;
1761                 }
1762                 inode->i_op = &shmem_symlink_inode_operations;
1763                 kaddr = kmap_atomic(page, KM_USER0);
1764                 memcpy(kaddr, symname, len);
1765                 kunmap_atomic(kaddr, KM_USER0);
1766                 set_page_dirty(page);
1767                 page_cache_release(page);
1768         }
1769         if (dir->i_mode & S_ISGID)
1770                 inode->i_gid = dir->i_gid;
1771         dir->i_size += BOGO_DIRENT_SIZE;
1772         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1773         d_instantiate(dentry, inode);
1774         dget(dentry);
1775         return 0;
1776 }
1777
1778 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1779 {
1780         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1781         return NULL;
1782 }
1783
1784 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1785 {
1786         struct page *page = NULL;
1787         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1788         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1789         return page;
1790 }
1791
1792 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1793 {
1794         if (!IS_ERR(nd_get_link(nd))) {
1795                 struct page *page = cookie;
1796                 kunmap(page);
1797                 mark_page_accessed(page);
1798                 page_cache_release(page);
1799         }
1800 }
1801
1802 static struct inode_operations shmem_symlink_inline_operations = {
1803         .readlink       = generic_readlink,
1804         .follow_link    = shmem_follow_link_inline,
1805 };
1806
1807 static struct inode_operations shmem_symlink_inode_operations = {
1808         .truncate       = shmem_truncate,
1809         .readlink       = generic_readlink,
1810         .follow_link    = shmem_follow_link,
1811         .put_link       = shmem_put_link,
1812 };
1813
1814 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1815 {
1816         char *this_char, *value, *rest;
1817
1818         while ((this_char = strsep(&options, ",")) != NULL) {
1819                 if (!*this_char)
1820                         continue;
1821                 if ((value = strchr(this_char,'=')) != NULL) {
1822                         *value++ = 0;
1823                 } else {
1824                         printk(KERN_ERR
1825                             "tmpfs: No value for mount option '%s'\n",
1826                             this_char);
1827                         return 1;
1828                 }
1829
1830                 if (!strcmp(this_char,"size")) {
1831                         unsigned long long size;
1832                         size = memparse(value,&rest);
1833                         if (*rest == '%') {
1834                                 size <<= PAGE_SHIFT;
1835                                 size *= totalram_pages;
1836                                 do_div(size, 100);
1837                                 rest++;
1838                         }
1839                         if (*rest)
1840                                 goto bad_val;
1841                         *blocks = size >> PAGE_CACHE_SHIFT;
1842                 } else if (!strcmp(this_char,"nr_blocks")) {
1843                         *blocks = memparse(value,&rest);
1844                         if (*rest)
1845                                 goto bad_val;
1846                 } else if (!strcmp(this_char,"nr_inodes")) {
1847                         *inodes = memparse(value,&rest);
1848                         if (*rest)
1849                                 goto bad_val;
1850                 } else if (!strcmp(this_char,"mode")) {
1851                         if (!mode)
1852                                 continue;
1853                         *mode = simple_strtoul(value,&rest,8);
1854                         if (*rest)
1855                                 goto bad_val;
1856                 } else if (!strcmp(this_char,"uid")) {
1857                         if (!uid)
1858                                 continue;
1859                         *uid = simple_strtoul(value,&rest,0);
1860                         if (*rest)
1861                                 goto bad_val;
1862                 } else if (!strcmp(this_char,"gid")) {
1863                         if (!gid)
1864                                 continue;
1865                         *gid = simple_strtoul(value,&rest,0);
1866                         if (*rest)
1867                                 goto bad_val;
1868                 } else {
1869                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1870                                this_char);
1871                         return 1;
1872                 }
1873         }
1874         return 0;
1875
1876 bad_val:
1877         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1878                value, this_char);
1879         return 1;
1880
1881 }
1882
1883 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1884 {
1885         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1886         unsigned long max_blocks = sbinfo->max_blocks;
1887         unsigned long max_inodes = sbinfo->max_inodes;
1888         unsigned long blocks;
1889         unsigned long inodes;
1890         int error = -EINVAL;
1891
1892         if (shmem_parse_options(data, NULL, NULL, NULL,
1893                                 &max_blocks, &max_inodes))
1894                 return error;
1895
1896         spin_lock(&sbinfo->stat_lock);
1897         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1898         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1899         if (max_blocks < blocks)
1900                 goto out;
1901         if (max_inodes < inodes)
1902                 goto out;
1903         /*
1904          * Those tests also disallow limited->unlimited while any are in
1905          * use, so i_blocks will always be zero when max_blocks is zero;
1906          * but we must separately disallow unlimited->limited, because
1907          * in that case we have no record of how much is already in use.
1908          */
1909         if (max_blocks && !sbinfo->max_blocks)
1910                 goto out;
1911         if (max_inodes && !sbinfo->max_inodes)
1912                 goto out;
1913
1914         error = 0;
1915         sbinfo->max_blocks  = max_blocks;
1916         sbinfo->free_blocks = max_blocks - blocks;
1917         sbinfo->max_inodes  = max_inodes;
1918         sbinfo->free_inodes = max_inodes - inodes;
1919 out:
1920         spin_unlock(&sbinfo->stat_lock);
1921         return error;
1922 }
1923 #endif
1924
1925 static void shmem_put_super(struct super_block *sb)
1926 {
1927         kfree(sb->s_fs_info);
1928         sb->s_fs_info = NULL;
1929 }
1930
1931 static int shmem_fill_super(struct super_block *sb,
1932                             void *data, int silent)
1933 {
1934         struct inode *inode;
1935         struct dentry *root;
1936         int mode   = S_IRWXUGO | S_ISVTX;
1937         uid_t uid = current->fsuid;
1938         gid_t gid = current->fsgid;
1939         int err = -ENOMEM;
1940         struct shmem_sb_info *sbinfo;
1941         unsigned long blocks = 0;
1942         unsigned long inodes = 0;
1943
1944 #ifdef CONFIG_TMPFS
1945         /*
1946          * Per default we only allow half of the physical ram per
1947          * tmpfs instance, limiting inodes to one per page of lowmem;
1948          * but the internal instance is left unlimited.
1949          */
1950         if (!(sb->s_flags & MS_NOUSER)) {
1951                 blocks = totalram_pages / 2;
1952                 inodes = totalram_pages - totalhigh_pages;
1953                 if (inodes > blocks)
1954                         inodes = blocks;
1955                 if (shmem_parse_options(data, &mode, &uid, &gid,
1956                                         &blocks, &inodes))
1957                         return -EINVAL;
1958         }
1959 #else
1960         sb->s_flags |= MS_NOUSER;
1961 #endif
1962
1963         /* Round up to L1_CACHE_BYTES to resist false sharing */
1964         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
1965                                 L1_CACHE_BYTES), GFP_KERNEL);
1966         if (!sbinfo)
1967                 return -ENOMEM;
1968
1969         spin_lock_init(&sbinfo->stat_lock);
1970         sbinfo->max_blocks = blocks;
1971         sbinfo->free_blocks = blocks;
1972         sbinfo->max_inodes = inodes;
1973         sbinfo->free_inodes = inodes;
1974
1975         sb->s_fs_info = sbinfo;
1976         sb->s_maxbytes = SHMEM_MAX_BYTES;
1977         sb->s_blocksize = PAGE_CACHE_SIZE;
1978         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1979         sb->s_magic = TMPFS_MAGIC;
1980         sb->s_op = &shmem_ops;
1981
1982         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1983         if (!inode)
1984                 goto failed;
1985         inode->i_uid = uid;
1986         inode->i_gid = gid;
1987         root = d_alloc_root(inode);
1988         if (!root)
1989                 goto failed_iput;
1990         sb->s_root = root;
1991         return 0;
1992
1993 failed_iput:
1994         iput(inode);
1995 failed:
1996         shmem_put_super(sb);
1997         return err;
1998 }
1999
2000 static kmem_cache_t *shmem_inode_cachep;
2001
2002 static struct inode *shmem_alloc_inode(struct super_block *sb)
2003 {
2004         struct shmem_inode_info *p;
2005         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2006         if (!p)
2007                 return NULL;
2008         return &p->vfs_inode;
2009 }
2010
2011 static void shmem_destroy_inode(struct inode *inode)
2012 {
2013         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2014                 /* only struct inode is valid if it's an inline symlink */
2015                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2016         }
2017         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2018 }
2019
2020 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
2021 {
2022         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2023
2024         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2025             SLAB_CTOR_CONSTRUCTOR) {
2026                 inode_init_once(&p->vfs_inode);
2027         }
2028 }
2029
2030 static int init_inodecache(void)
2031 {
2032         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2033                                 sizeof(struct shmem_inode_info),
2034                                 0, 0, init_once, NULL);
2035         if (shmem_inode_cachep == NULL)
2036                 return -ENOMEM;
2037         return 0;
2038 }
2039
2040 static void destroy_inodecache(void)
2041 {
2042         if (kmem_cache_destroy(shmem_inode_cachep))
2043                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2044 }
2045
2046 static struct address_space_operations shmem_aops = {
2047         .writepage      = shmem_writepage,
2048         .set_page_dirty = __set_page_dirty_nobuffers,
2049 #ifdef CONFIG_TMPFS
2050         .prepare_write  = shmem_prepare_write,
2051         .commit_write   = simple_commit_write,
2052 #endif
2053 };
2054
2055 static struct file_operations shmem_file_operations = {
2056         .mmap           = shmem_mmap,
2057 #ifdef CONFIG_TMPFS
2058         .llseek         = generic_file_llseek,
2059         .read           = shmem_file_read,
2060         .write          = shmem_file_write,
2061         .fsync          = simple_sync_file,
2062         .sendfile       = shmem_file_sendfile,
2063 #endif
2064 };
2065
2066 static struct inode_operations shmem_inode_operations = {
2067         .truncate       = shmem_truncate,
2068         .setattr        = shmem_notify_change,
2069 };
2070
2071 static struct inode_operations shmem_dir_inode_operations = {
2072 #ifdef CONFIG_TMPFS
2073         .create         = shmem_create,
2074         .lookup         = simple_lookup,
2075         .link           = shmem_link,
2076         .unlink         = shmem_unlink,
2077         .symlink        = shmem_symlink,
2078         .mkdir          = shmem_mkdir,
2079         .rmdir          = shmem_rmdir,
2080         .mknod          = shmem_mknod,
2081         .rename         = shmem_rename,
2082 #endif
2083 };
2084
2085 static struct super_operations shmem_ops = {
2086         .alloc_inode    = shmem_alloc_inode,
2087         .destroy_inode  = shmem_destroy_inode,
2088 #ifdef CONFIG_TMPFS
2089         .statfs         = shmem_statfs,
2090         .remount_fs     = shmem_remount_fs,
2091 #endif
2092         .delete_inode   = shmem_delete_inode,
2093         .drop_inode     = generic_delete_inode,
2094         .put_super      = shmem_put_super,
2095 };
2096
2097 static struct vm_operations_struct shmem_vm_ops = {
2098         .nopage         = shmem_nopage,
2099         .populate       = shmem_populate,
2100 #ifdef CONFIG_NUMA
2101         .set_policy     = shmem_set_policy,
2102         .get_policy     = shmem_get_policy,
2103 #endif
2104 };
2105
2106
2107 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2108         int flags, const char *dev_name, void *data)
2109 {
2110         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2111 }
2112
2113 static struct file_system_type tmpfs_fs_type = {
2114         .owner          = THIS_MODULE,
2115         .name           = "tmpfs",
2116         .get_sb         = shmem_get_sb,
2117         .kill_sb        = kill_litter_super,
2118 };
2119 static struct vfsmount *shm_mnt;
2120
2121 static int __init init_tmpfs(void)
2122 {
2123         int error;
2124
2125         error = init_inodecache();
2126         if (error)
2127                 goto out3;
2128
2129         error = register_filesystem(&tmpfs_fs_type);
2130         if (error) {
2131                 printk(KERN_ERR "Could not register tmpfs\n");
2132                 goto out2;
2133         }
2134 #ifdef CONFIG_TMPFS
2135         devfs_mk_dir("shm");
2136 #endif
2137         shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2138                                 tmpfs_fs_type.name, NULL);
2139         if (IS_ERR(shm_mnt)) {
2140                 error = PTR_ERR(shm_mnt);
2141                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2142                 goto out1;
2143         }
2144         return 0;
2145
2146 out1:
2147         unregister_filesystem(&tmpfs_fs_type);
2148 out2:
2149         destroy_inodecache();
2150 out3:
2151         shm_mnt = ERR_PTR(error);
2152         return error;
2153 }
2154 module_init(init_tmpfs)
2155
2156 /*
2157  * shmem_file_setup - get an unlinked file living in tmpfs
2158  *
2159  * @name: name for dentry (to be seen in /proc/<pid>/maps
2160  * @size: size to be set for the file
2161  *
2162  */
2163 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2164 {
2165         int error;
2166         struct file *file;
2167         struct inode *inode;
2168         struct dentry *dentry, *root;
2169         struct qstr this;
2170
2171         if (IS_ERR(shm_mnt))
2172                 return (void *)shm_mnt;
2173
2174         if (size < 0 || size > SHMEM_MAX_BYTES)
2175                 return ERR_PTR(-EINVAL);
2176
2177         if (shmem_acct_size(flags, size))
2178                 return ERR_PTR(-ENOMEM);
2179
2180         error = -ENOMEM;
2181         this.name = name;
2182         this.len = strlen(name);
2183         this.hash = 0; /* will go */
2184         root = shm_mnt->mnt_root;
2185         dentry = d_alloc(root, &this);
2186         if (!dentry)
2187                 goto put_memory;
2188
2189         error = -ENFILE;
2190         file = get_empty_filp();
2191         if (!file)
2192                 goto put_dentry;
2193
2194         error = -ENOSPC;
2195         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2196         if (!inode)
2197                 goto close_file;
2198
2199         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2200         d_instantiate(dentry, inode);
2201         inode->i_size = size;
2202         inode->i_nlink = 0;     /* It is unlinked */
2203         file->f_vfsmnt = mntget(shm_mnt);
2204         file->f_dentry = dentry;
2205         file->f_mapping = inode->i_mapping;
2206         file->f_op = &shmem_file_operations;
2207         file->f_mode = FMODE_WRITE | FMODE_READ;
2208         return file;
2209
2210 close_file:
2211         put_filp(file);
2212 put_dentry:
2213         dput(dentry);
2214 put_memory:
2215         shmem_unacct_size(flags, size);
2216         return ERR_PTR(error);
2217 }
2218
2219 /*
2220  * shmem_zero_setup - setup a shared anonymous mapping
2221  *
2222  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2223  */
2224 int shmem_zero_setup(struct vm_area_struct *vma)
2225 {
2226         struct file *file;
2227         loff_t size = vma->vm_end - vma->vm_start;
2228
2229         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2230         if (IS_ERR(file))
2231                 return PTR_ERR(file);
2232
2233         if (vma->vm_file)
2234                 fput(vma->vm_file);
2235         vma->vm_file = file;
2236         vma->vm_ops = &shmem_vm_ops;
2237         return 0;
2238 }