Driver core: more fallout from class_device changes for pcmcia
[linux-2.6.git] / mm / shmem.c
1 /*
2  * Resizable virtual memory filesystem for Linux.
3  *
4  * Copyright (C) 2000 Linus Torvalds.
5  *               2000 Transmeta Corp.
6  *               2000-2001 Christoph Rohland
7  *               2000-2001 SAP AG
8  *               2002 Red Hat Inc.
9  * Copyright (C) 2002-2005 Hugh Dickins.
10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
12  *
13  * Extended attribute support for tmpfs:
14  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
16  *
17  * This file is released under the GPL.
18  */
19
20 /*
21  * This virtual memory filesystem is heavily based on the ramfs. It
22  * extends ramfs by the ability to use swap and honor resource limits
23  * which makes it a completely usable filesystem.
24  */
25
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/fs.h>
29 #include <linux/xattr.h>
30 #include <linux/generic_acl.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/ctype.h>
49 #include <linux/migrate.h>
50 #include <linux/highmem.h>
51 #include <linux/backing-dev.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/div64.h>
55 #include <asm/pgtable.h>
56
57 /* This magic number is used in glibc for posix shared memory */
58 #define TMPFS_MAGIC     0x01021994
59
60 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
63
64 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
66
67 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
68
69 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70 #define SHMEM_PAGEIN     VM_READ
71 #define SHMEM_TRUNCATE   VM_WRITE
72
73 /* Definition to limit shmem_truncate's steps between cond_rescheds */
74 #define LATENCY_LIMIT    64
75
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
78
79 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
80 enum sgp_type {
81         SGP_QUICK,      /* don't try more than file page cache lookup */
82         SGP_READ,       /* don't exceed i_size, don't allocate page */
83         SGP_CACHE,      /* don't exceed i_size, may allocate page */
84         SGP_WRITE,      /* may exceed i_size, may allocate page */
85 };
86
87 static int shmem_getpage(struct inode *inode, unsigned long idx,
88                          struct page **pagep, enum sgp_type sgp, int *type);
89
90 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
91 {
92         /*
93          * The above definition of ENTRIES_PER_PAGE, and the use of
94          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
95          * might be reconsidered if it ever diverges from PAGE_SIZE.
96          */
97         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
98 }
99
100 static inline void shmem_dir_free(struct page *page)
101 {
102         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
103 }
104
105 static struct page **shmem_dir_map(struct page *page)
106 {
107         return (struct page **)kmap_atomic(page, KM_USER0);
108 }
109
110 static inline void shmem_dir_unmap(struct page **dir)
111 {
112         kunmap_atomic(dir, KM_USER0);
113 }
114
115 static swp_entry_t *shmem_swp_map(struct page *page)
116 {
117         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
118 }
119
120 static inline void shmem_swp_balance_unmap(void)
121 {
122         /*
123          * When passing a pointer to an i_direct entry, to code which
124          * also handles indirect entries and so will shmem_swp_unmap,
125          * we must arrange for the preempt count to remain in balance.
126          * What kmap_atomic of a lowmem page does depends on config
127          * and architecture, so pretend to kmap_atomic some lowmem page.
128          */
129         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
130 }
131
132 static inline void shmem_swp_unmap(swp_entry_t *entry)
133 {
134         kunmap_atomic(entry, KM_USER1);
135 }
136
137 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
138 {
139         return sb->s_fs_info;
140 }
141
142 /*
143  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
144  * for shared memory and for shared anonymous (/dev/zero) mappings
145  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
146  * consistent with the pre-accounting of private mappings ...
147  */
148 static inline int shmem_acct_size(unsigned long flags, loff_t size)
149 {
150         return (flags & VM_ACCOUNT)?
151                 security_vm_enough_memory(VM_ACCT(size)): 0;
152 }
153
154 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
155 {
156         if (flags & VM_ACCOUNT)
157                 vm_unacct_memory(VM_ACCT(size));
158 }
159
160 /*
161  * ... whereas tmpfs objects are accounted incrementally as
162  * pages are allocated, in order to allow huge sparse files.
163  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
164  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
165  */
166 static inline int shmem_acct_block(unsigned long flags)
167 {
168         return (flags & VM_ACCOUNT)?
169                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
170 }
171
172 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
173 {
174         if (!(flags & VM_ACCOUNT))
175                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
176 }
177
178 static struct super_operations shmem_ops;
179 static const struct address_space_operations shmem_aops;
180 static const struct file_operations shmem_file_operations;
181 static const struct inode_operations shmem_inode_operations;
182 static const struct inode_operations shmem_dir_inode_operations;
183 static const struct inode_operations shmem_special_inode_operations;
184 static struct vm_operations_struct shmem_vm_ops;
185
186 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
187         .ra_pages       = 0,    /* No readahead */
188         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
189         .unplug_io_fn   = default_unplug_io_fn,
190 };
191
192 static LIST_HEAD(shmem_swaplist);
193 static DEFINE_SPINLOCK(shmem_swaplist_lock);
194
195 static void shmem_free_blocks(struct inode *inode, long pages)
196 {
197         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
198         if (sbinfo->max_blocks) {
199                 spin_lock(&sbinfo->stat_lock);
200                 sbinfo->free_blocks += pages;
201                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
202                 spin_unlock(&sbinfo->stat_lock);
203         }
204 }
205
206 /*
207  * shmem_recalc_inode - recalculate the size of an inode
208  *
209  * @inode: inode to recalc
210  *
211  * We have to calculate the free blocks since the mm can drop
212  * undirtied hole pages behind our back.
213  *
214  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
215  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
216  *
217  * It has to be called with the spinlock held.
218  */
219 static void shmem_recalc_inode(struct inode *inode)
220 {
221         struct shmem_inode_info *info = SHMEM_I(inode);
222         long freed;
223
224         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
225         if (freed > 0) {
226                 info->alloced -= freed;
227                 shmem_unacct_blocks(info->flags, freed);
228                 shmem_free_blocks(inode, freed);
229         }
230 }
231
232 /*
233  * shmem_swp_entry - find the swap vector position in the info structure
234  *
235  * @info:  info structure for the inode
236  * @index: index of the page to find
237  * @page:  optional page to add to the structure. Has to be preset to
238  *         all zeros
239  *
240  * If there is no space allocated yet it will return NULL when
241  * page is NULL, else it will use the page for the needed block,
242  * setting it to NULL on return to indicate that it has been used.
243  *
244  * The swap vector is organized the following way:
245  *
246  * There are SHMEM_NR_DIRECT entries directly stored in the
247  * shmem_inode_info structure. So small files do not need an addional
248  * allocation.
249  *
250  * For pages with index > SHMEM_NR_DIRECT there is the pointer
251  * i_indirect which points to a page which holds in the first half
252  * doubly indirect blocks, in the second half triple indirect blocks:
253  *
254  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
255  * following layout (for SHMEM_NR_DIRECT == 16):
256  *
257  * i_indirect -> dir --> 16-19
258  *            |      +-> 20-23
259  *            |
260  *            +-->dir2 --> 24-27
261  *            |        +-> 28-31
262  *            |        +-> 32-35
263  *            |        +-> 36-39
264  *            |
265  *            +-->dir3 --> 40-43
266  *                     +-> 44-47
267  *                     +-> 48-51
268  *                     +-> 52-55
269  */
270 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
271 {
272         unsigned long offset;
273         struct page **dir;
274         struct page *subdir;
275
276         if (index < SHMEM_NR_DIRECT) {
277                 shmem_swp_balance_unmap();
278                 return info->i_direct+index;
279         }
280         if (!info->i_indirect) {
281                 if (page) {
282                         info->i_indirect = *page;
283                         *page = NULL;
284                 }
285                 return NULL;                    /* need another page */
286         }
287
288         index -= SHMEM_NR_DIRECT;
289         offset = index % ENTRIES_PER_PAGE;
290         index /= ENTRIES_PER_PAGE;
291         dir = shmem_dir_map(info->i_indirect);
292
293         if (index >= ENTRIES_PER_PAGE/2) {
294                 index -= ENTRIES_PER_PAGE/2;
295                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
296                 index %= ENTRIES_PER_PAGE;
297                 subdir = *dir;
298                 if (!subdir) {
299                         if (page) {
300                                 *dir = *page;
301                                 *page = NULL;
302                         }
303                         shmem_dir_unmap(dir);
304                         return NULL;            /* need another page */
305                 }
306                 shmem_dir_unmap(dir);
307                 dir = shmem_dir_map(subdir);
308         }
309
310         dir += index;
311         subdir = *dir;
312         if (!subdir) {
313                 if (!page || !(subdir = *page)) {
314                         shmem_dir_unmap(dir);
315                         return NULL;            /* need a page */
316                 }
317                 *dir = subdir;
318                 *page = NULL;
319         }
320         shmem_dir_unmap(dir);
321         return shmem_swp_map(subdir) + offset;
322 }
323
324 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
325 {
326         long incdec = value? 1: -1;
327
328         entry->val = value;
329         info->swapped += incdec;
330         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
331                 struct page *page = kmap_atomic_to_page(entry);
332                 set_page_private(page, page_private(page) + incdec);
333         }
334 }
335
336 /*
337  * shmem_swp_alloc - get the position of the swap entry for the page.
338  *                   If it does not exist allocate the entry.
339  *
340  * @info:       info structure for the inode
341  * @index:      index of the page to find
342  * @sgp:        check and recheck i_size? skip allocation?
343  */
344 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
345 {
346         struct inode *inode = &info->vfs_inode;
347         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
348         struct page *page = NULL;
349         swp_entry_t *entry;
350
351         if (sgp != SGP_WRITE &&
352             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
353                 return ERR_PTR(-EINVAL);
354
355         while (!(entry = shmem_swp_entry(info, index, &page))) {
356                 if (sgp == SGP_READ)
357                         return shmem_swp_map(ZERO_PAGE(0));
358                 /*
359                  * Test free_blocks against 1 not 0, since we have 1 data
360                  * page (and perhaps indirect index pages) yet to allocate:
361                  * a waste to allocate index if we cannot allocate data.
362                  */
363                 if (sbinfo->max_blocks) {
364                         spin_lock(&sbinfo->stat_lock);
365                         if (sbinfo->free_blocks <= 1) {
366                                 spin_unlock(&sbinfo->stat_lock);
367                                 return ERR_PTR(-ENOSPC);
368                         }
369                         sbinfo->free_blocks--;
370                         inode->i_blocks += BLOCKS_PER_PAGE;
371                         spin_unlock(&sbinfo->stat_lock);
372                 }
373
374                 spin_unlock(&info->lock);
375                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
376                 if (page)
377                         set_page_private(page, 0);
378                 spin_lock(&info->lock);
379
380                 if (!page) {
381                         shmem_free_blocks(inode, 1);
382                         return ERR_PTR(-ENOMEM);
383                 }
384                 if (sgp != SGP_WRITE &&
385                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
386                         entry = ERR_PTR(-EINVAL);
387                         break;
388                 }
389                 if (info->next_index <= index)
390                         info->next_index = index + 1;
391         }
392         if (page) {
393                 /* another task gave its page, or truncated the file */
394                 shmem_free_blocks(inode, 1);
395                 shmem_dir_free(page);
396         }
397         if (info->next_index <= index && !IS_ERR(entry))
398                 info->next_index = index + 1;
399         return entry;
400 }
401
402 /*
403  * shmem_free_swp - free some swap entries in a directory
404  *
405  * @dir:   pointer to the directory
406  * @edir:  pointer after last entry of the directory
407  */
408 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
409 {
410         swp_entry_t *ptr;
411         int freed = 0;
412
413         for (ptr = dir; ptr < edir; ptr++) {
414                 if (ptr->val) {
415                         free_swap_and_cache(*ptr);
416                         *ptr = (swp_entry_t){0};
417                         freed++;
418                 }
419         }
420         return freed;
421 }
422
423 static int shmem_map_and_free_swp(struct page *subdir,
424                 int offset, int limit, struct page ***dir)
425 {
426         swp_entry_t *ptr;
427         int freed = 0;
428
429         ptr = shmem_swp_map(subdir);
430         for (; offset < limit; offset += LATENCY_LIMIT) {
431                 int size = limit - offset;
432                 if (size > LATENCY_LIMIT)
433                         size = LATENCY_LIMIT;
434                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
435                 if (need_resched()) {
436                         shmem_swp_unmap(ptr);
437                         if (*dir) {
438                                 shmem_dir_unmap(*dir);
439                                 *dir = NULL;
440                         }
441                         cond_resched();
442                         ptr = shmem_swp_map(subdir);
443                 }
444         }
445         shmem_swp_unmap(ptr);
446         return freed;
447 }
448
449 static void shmem_free_pages(struct list_head *next)
450 {
451         struct page *page;
452         int freed = 0;
453
454         do {
455                 page = container_of(next, struct page, lru);
456                 next = next->next;
457                 shmem_dir_free(page);
458                 freed++;
459                 if (freed >= LATENCY_LIMIT) {
460                         cond_resched();
461                         freed = 0;
462                 }
463         } while (next);
464 }
465
466 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
467 {
468         struct shmem_inode_info *info = SHMEM_I(inode);
469         unsigned long idx;
470         unsigned long size;
471         unsigned long limit;
472         unsigned long stage;
473         unsigned long diroff;
474         struct page **dir;
475         struct page *topdir;
476         struct page *middir;
477         struct page *subdir;
478         swp_entry_t *ptr;
479         LIST_HEAD(pages_to_free);
480         long nr_pages_to_free = 0;
481         long nr_swaps_freed = 0;
482         int offset;
483         int freed;
484         int punch_hole = 0;
485
486         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
487         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
488         if (idx >= info->next_index)
489                 return;
490
491         spin_lock(&info->lock);
492         info->flags |= SHMEM_TRUNCATE;
493         if (likely(end == (loff_t) -1)) {
494                 limit = info->next_index;
495                 info->next_index = idx;
496         } else {
497                 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
498                 if (limit > info->next_index)
499                         limit = info->next_index;
500                 punch_hole = 1;
501         }
502
503         topdir = info->i_indirect;
504         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
505                 info->i_indirect = NULL;
506                 nr_pages_to_free++;
507                 list_add(&topdir->lru, &pages_to_free);
508         }
509         spin_unlock(&info->lock);
510
511         if (info->swapped && idx < SHMEM_NR_DIRECT) {
512                 ptr = info->i_direct;
513                 size = limit;
514                 if (size > SHMEM_NR_DIRECT)
515                         size = SHMEM_NR_DIRECT;
516                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
517         }
518
519         /*
520          * If there are no indirect blocks or we are punching a hole
521          * below indirect blocks, nothing to be done.
522          */
523         if (!topdir || (punch_hole && (limit <= SHMEM_NR_DIRECT)))
524                 goto done2;
525
526         BUG_ON(limit <= SHMEM_NR_DIRECT);
527         limit -= SHMEM_NR_DIRECT;
528         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
529         offset = idx % ENTRIES_PER_PAGE;
530         idx -= offset;
531
532         dir = shmem_dir_map(topdir);
533         stage = ENTRIES_PER_PAGEPAGE/2;
534         if (idx < ENTRIES_PER_PAGEPAGE/2) {
535                 middir = topdir;
536                 diroff = idx/ENTRIES_PER_PAGE;
537         } else {
538                 dir += ENTRIES_PER_PAGE/2;
539                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
540                 while (stage <= idx)
541                         stage += ENTRIES_PER_PAGEPAGE;
542                 middir = *dir;
543                 if (*dir) {
544                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
545                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
546                         if (!diroff && !offset) {
547                                 *dir = NULL;
548                                 nr_pages_to_free++;
549                                 list_add(&middir->lru, &pages_to_free);
550                         }
551                         shmem_dir_unmap(dir);
552                         dir = shmem_dir_map(middir);
553                 } else {
554                         diroff = 0;
555                         offset = 0;
556                         idx = stage;
557                 }
558         }
559
560         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
561                 if (unlikely(idx == stage)) {
562                         shmem_dir_unmap(dir);
563                         dir = shmem_dir_map(topdir) +
564                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
565                         while (!*dir) {
566                                 dir++;
567                                 idx += ENTRIES_PER_PAGEPAGE;
568                                 if (idx >= limit)
569                                         goto done1;
570                         }
571                         stage = idx + ENTRIES_PER_PAGEPAGE;
572                         middir = *dir;
573                         *dir = NULL;
574                         nr_pages_to_free++;
575                         list_add(&middir->lru, &pages_to_free);
576                         shmem_dir_unmap(dir);
577                         cond_resched();
578                         dir = shmem_dir_map(middir);
579                         diroff = 0;
580                 }
581                 subdir = dir[diroff];
582                 if (subdir && page_private(subdir)) {
583                         size = limit - idx;
584                         if (size > ENTRIES_PER_PAGE)
585                                 size = ENTRIES_PER_PAGE;
586                         freed = shmem_map_and_free_swp(subdir,
587                                                 offset, size, &dir);
588                         if (!dir)
589                                 dir = shmem_dir_map(middir);
590                         nr_swaps_freed += freed;
591                         if (offset)
592                                 spin_lock(&info->lock);
593                         set_page_private(subdir, page_private(subdir) - freed);
594                         if (offset)
595                                 spin_unlock(&info->lock);
596                         if (!punch_hole)
597                                 BUG_ON(page_private(subdir) > offset);
598                 }
599                 if (offset)
600                         offset = 0;
601                 else if (subdir && !page_private(subdir)) {
602                         dir[diroff] = NULL;
603                         nr_pages_to_free++;
604                         list_add(&subdir->lru, &pages_to_free);
605                 }
606         }
607 done1:
608         shmem_dir_unmap(dir);
609 done2:
610         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
611                 /*
612                  * Call truncate_inode_pages again: racing shmem_unuse_inode
613                  * may have swizzled a page in from swap since vmtruncate or
614                  * generic_delete_inode did it, before we lowered next_index.
615                  * Also, though shmem_getpage checks i_size before adding to
616                  * cache, no recheck after: so fix the narrow window there too.
617                  */
618                 truncate_inode_pages_range(inode->i_mapping, start, end);
619         }
620
621         spin_lock(&info->lock);
622         info->flags &= ~SHMEM_TRUNCATE;
623         info->swapped -= nr_swaps_freed;
624         if (nr_pages_to_free)
625                 shmem_free_blocks(inode, nr_pages_to_free);
626         shmem_recalc_inode(inode);
627         spin_unlock(&info->lock);
628
629         /*
630          * Empty swap vector directory pages to be freed?
631          */
632         if (!list_empty(&pages_to_free)) {
633                 pages_to_free.prev->next = NULL;
634                 shmem_free_pages(pages_to_free.next);
635         }
636 }
637
638 static void shmem_truncate(struct inode *inode)
639 {
640         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
641 }
642
643 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
644 {
645         struct inode *inode = dentry->d_inode;
646         struct page *page = NULL;
647         int error;
648
649         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
650                 if (attr->ia_size < inode->i_size) {
651                         /*
652                          * If truncating down to a partial page, then
653                          * if that page is already allocated, hold it
654                          * in memory until the truncation is over, so
655                          * truncate_partial_page cannnot miss it were
656                          * it assigned to swap.
657                          */
658                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
659                                 (void) shmem_getpage(inode,
660                                         attr->ia_size>>PAGE_CACHE_SHIFT,
661                                                 &page, SGP_READ, NULL);
662                         }
663                         /*
664                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
665                          * detect if any pages might have been added to cache
666                          * after truncate_inode_pages.  But we needn't bother
667                          * if it's being fully truncated to zero-length: the
668                          * nrpages check is efficient enough in that case.
669                          */
670                         if (attr->ia_size) {
671                                 struct shmem_inode_info *info = SHMEM_I(inode);
672                                 spin_lock(&info->lock);
673                                 info->flags &= ~SHMEM_PAGEIN;
674                                 spin_unlock(&info->lock);
675                         }
676                 }
677         }
678
679         error = inode_change_ok(inode, attr);
680         if (!error)
681                 error = inode_setattr(inode, attr);
682 #ifdef CONFIG_TMPFS_POSIX_ACL
683         if (!error && (attr->ia_valid & ATTR_MODE))
684                 error = generic_acl_chmod(inode, &shmem_acl_ops);
685 #endif
686         if (page)
687                 page_cache_release(page);
688         return error;
689 }
690
691 static void shmem_delete_inode(struct inode *inode)
692 {
693         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
694         struct shmem_inode_info *info = SHMEM_I(inode);
695
696         if (inode->i_op->truncate == shmem_truncate) {
697                 truncate_inode_pages(inode->i_mapping, 0);
698                 shmem_unacct_size(info->flags, inode->i_size);
699                 inode->i_size = 0;
700                 shmem_truncate(inode);
701                 if (!list_empty(&info->swaplist)) {
702                         spin_lock(&shmem_swaplist_lock);
703                         list_del_init(&info->swaplist);
704                         spin_unlock(&shmem_swaplist_lock);
705                 }
706         }
707         BUG_ON(inode->i_blocks);
708         if (sbinfo->max_inodes) {
709                 spin_lock(&sbinfo->stat_lock);
710                 sbinfo->free_inodes++;
711                 spin_unlock(&sbinfo->stat_lock);
712         }
713         clear_inode(inode);
714 }
715
716 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
717 {
718         swp_entry_t *ptr;
719
720         for (ptr = dir; ptr < edir; ptr++) {
721                 if (ptr->val == entry.val)
722                         return ptr - dir;
723         }
724         return -1;
725 }
726
727 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
728 {
729         struct inode *inode;
730         unsigned long idx;
731         unsigned long size;
732         unsigned long limit;
733         unsigned long stage;
734         struct page **dir;
735         struct page *subdir;
736         swp_entry_t *ptr;
737         int offset;
738
739         idx = 0;
740         ptr = info->i_direct;
741         spin_lock(&info->lock);
742         limit = info->next_index;
743         size = limit;
744         if (size > SHMEM_NR_DIRECT)
745                 size = SHMEM_NR_DIRECT;
746         offset = shmem_find_swp(entry, ptr, ptr+size);
747         if (offset >= 0) {
748                 shmem_swp_balance_unmap();
749                 goto found;
750         }
751         if (!info->i_indirect)
752                 goto lost2;
753
754         dir = shmem_dir_map(info->i_indirect);
755         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
756
757         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
758                 if (unlikely(idx == stage)) {
759                         shmem_dir_unmap(dir-1);
760                         dir = shmem_dir_map(info->i_indirect) +
761                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
762                         while (!*dir) {
763                                 dir++;
764                                 idx += ENTRIES_PER_PAGEPAGE;
765                                 if (idx >= limit)
766                                         goto lost1;
767                         }
768                         stage = idx + ENTRIES_PER_PAGEPAGE;
769                         subdir = *dir;
770                         shmem_dir_unmap(dir);
771                         dir = shmem_dir_map(subdir);
772                 }
773                 subdir = *dir;
774                 if (subdir && page_private(subdir)) {
775                         ptr = shmem_swp_map(subdir);
776                         size = limit - idx;
777                         if (size > ENTRIES_PER_PAGE)
778                                 size = ENTRIES_PER_PAGE;
779                         offset = shmem_find_swp(entry, ptr, ptr+size);
780                         if (offset >= 0) {
781                                 shmem_dir_unmap(dir);
782                                 goto found;
783                         }
784                         shmem_swp_unmap(ptr);
785                 }
786         }
787 lost1:
788         shmem_dir_unmap(dir-1);
789 lost2:
790         spin_unlock(&info->lock);
791         return 0;
792 found:
793         idx += offset;
794         inode = &info->vfs_inode;
795         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
796                 info->flags |= SHMEM_PAGEIN;
797                 shmem_swp_set(info, ptr + offset, 0);
798         }
799         shmem_swp_unmap(ptr);
800         spin_unlock(&info->lock);
801         /*
802          * Decrement swap count even when the entry is left behind:
803          * try_to_unuse will skip over mms, then reincrement count.
804          */
805         swap_free(entry);
806         return 1;
807 }
808
809 /*
810  * shmem_unuse() search for an eventually swapped out shmem page.
811  */
812 int shmem_unuse(swp_entry_t entry, struct page *page)
813 {
814         struct list_head *p, *next;
815         struct shmem_inode_info *info;
816         int found = 0;
817
818         spin_lock(&shmem_swaplist_lock);
819         list_for_each_safe(p, next, &shmem_swaplist) {
820                 info = list_entry(p, struct shmem_inode_info, swaplist);
821                 if (!info->swapped)
822                         list_del_init(&info->swaplist);
823                 else if (shmem_unuse_inode(info, entry, page)) {
824                         /* move head to start search for next from here */
825                         list_move_tail(&shmem_swaplist, &info->swaplist);
826                         found = 1;
827                         break;
828                 }
829         }
830         spin_unlock(&shmem_swaplist_lock);
831         return found;
832 }
833
834 /*
835  * Move the page from the page cache to the swap cache.
836  */
837 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
838 {
839         struct shmem_inode_info *info;
840         swp_entry_t *entry, swap;
841         struct address_space *mapping;
842         unsigned long index;
843         struct inode *inode;
844
845         BUG_ON(!PageLocked(page));
846         BUG_ON(page_mapped(page));
847
848         mapping = page->mapping;
849         index = page->index;
850         inode = mapping->host;
851         info = SHMEM_I(inode);
852         if (info->flags & VM_LOCKED)
853                 goto redirty;
854         swap = get_swap_page();
855         if (!swap.val)
856                 goto redirty;
857
858         spin_lock(&info->lock);
859         shmem_recalc_inode(inode);
860         if (index >= info->next_index) {
861                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
862                 goto unlock;
863         }
864         entry = shmem_swp_entry(info, index, NULL);
865         BUG_ON(!entry);
866         BUG_ON(entry->val);
867
868         if (move_to_swap_cache(page, swap) == 0) {
869                 shmem_swp_set(info, entry, swap.val);
870                 shmem_swp_unmap(entry);
871                 spin_unlock(&info->lock);
872                 if (list_empty(&info->swaplist)) {
873                         spin_lock(&shmem_swaplist_lock);
874                         /* move instead of add in case we're racing */
875                         list_move_tail(&info->swaplist, &shmem_swaplist);
876                         spin_unlock(&shmem_swaplist_lock);
877                 }
878                 unlock_page(page);
879                 return 0;
880         }
881
882         shmem_swp_unmap(entry);
883 unlock:
884         spin_unlock(&info->lock);
885         swap_free(swap);
886 redirty:
887         set_page_dirty(page);
888         return AOP_WRITEPAGE_ACTIVATE;  /* Return with the page locked */
889 }
890
891 #ifdef CONFIG_NUMA
892 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
893 {
894         char *nodelist = strchr(value, ':');
895         int err = 1;
896
897         if (nodelist) {
898                 /* NUL-terminate policy string */
899                 *nodelist++ = '\0';
900                 if (nodelist_parse(nodelist, *policy_nodes))
901                         goto out;
902         }
903         if (!strcmp(value, "default")) {
904                 *policy = MPOL_DEFAULT;
905                 /* Don't allow a nodelist */
906                 if (!nodelist)
907                         err = 0;
908         } else if (!strcmp(value, "prefer")) {
909                 *policy = MPOL_PREFERRED;
910                 /* Insist on a nodelist of one node only */
911                 if (nodelist) {
912                         char *rest = nodelist;
913                         while (isdigit(*rest))
914                                 rest++;
915                         if (!*rest)
916                                 err = 0;
917                 }
918         } else if (!strcmp(value, "bind")) {
919                 *policy = MPOL_BIND;
920                 /* Insist on a nodelist */
921                 if (nodelist)
922                         err = 0;
923         } else if (!strcmp(value, "interleave")) {
924                 *policy = MPOL_INTERLEAVE;
925                 /* Default to nodes online if no nodelist */
926                 if (!nodelist)
927                         *policy_nodes = node_online_map;
928                 err = 0;
929         }
930 out:
931         /* Restore string for error message */
932         if (nodelist)
933                 *--nodelist = ':';
934         return err;
935 }
936
937 static struct page *shmem_swapin_async(struct shared_policy *p,
938                                        swp_entry_t entry, unsigned long idx)
939 {
940         struct page *page;
941         struct vm_area_struct pvma;
942
943         /* Create a pseudo vma that just contains the policy */
944         memset(&pvma, 0, sizeof(struct vm_area_struct));
945         pvma.vm_end = PAGE_SIZE;
946         pvma.vm_pgoff = idx;
947         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
948         page = read_swap_cache_async(entry, &pvma, 0);
949         mpol_free(pvma.vm_policy);
950         return page;
951 }
952
953 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
954                           unsigned long idx)
955 {
956         struct shared_policy *p = &info->policy;
957         int i, num;
958         struct page *page;
959         unsigned long offset;
960
961         num = valid_swaphandles(entry, &offset);
962         for (i = 0; i < num; offset++, i++) {
963                 page = shmem_swapin_async(p,
964                                 swp_entry(swp_type(entry), offset), idx);
965                 if (!page)
966                         break;
967                 page_cache_release(page);
968         }
969         lru_add_drain();        /* Push any new pages onto the LRU now */
970         return shmem_swapin_async(p, entry, idx);
971 }
972
973 static struct page *
974 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
975                  unsigned long idx)
976 {
977         struct vm_area_struct pvma;
978         struct page *page;
979
980         memset(&pvma, 0, sizeof(struct vm_area_struct));
981         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
982         pvma.vm_pgoff = idx;
983         pvma.vm_end = PAGE_SIZE;
984         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
985         mpol_free(pvma.vm_policy);
986         return page;
987 }
988 #else
989 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
990 {
991         return 1;
992 }
993
994 static inline struct page *
995 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
996 {
997         swapin_readahead(entry, 0, NULL);
998         return read_swap_cache_async(entry, NULL, 0);
999 }
1000
1001 static inline struct page *
1002 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
1003 {
1004         return alloc_page(gfp | __GFP_ZERO);
1005 }
1006 #endif
1007
1008 /*
1009  * shmem_getpage - either get the page from swap or allocate a new one
1010  *
1011  * If we allocate a new one we do not mark it dirty. That's up to the
1012  * vm. If we swap it in we mark it dirty since we also free the swap
1013  * entry since a page cannot live in both the swap and page cache
1014  */
1015 static int shmem_getpage(struct inode *inode, unsigned long idx,
1016                         struct page **pagep, enum sgp_type sgp, int *type)
1017 {
1018         struct address_space *mapping = inode->i_mapping;
1019         struct shmem_inode_info *info = SHMEM_I(inode);
1020         struct shmem_sb_info *sbinfo;
1021         struct page *filepage = *pagep;
1022         struct page *swappage;
1023         swp_entry_t *entry;
1024         swp_entry_t swap;
1025         int error;
1026
1027         if (idx >= SHMEM_MAX_INDEX)
1028                 return -EFBIG;
1029         /*
1030          * Normally, filepage is NULL on entry, and either found
1031          * uptodate immediately, or allocated and zeroed, or read
1032          * in under swappage, which is then assigned to filepage.
1033          * But shmem_prepare_write passes in a locked filepage,
1034          * which may be found not uptodate by other callers too,
1035          * and may need to be copied from the swappage read in.
1036          */
1037 repeat:
1038         if (!filepage)
1039                 filepage = find_lock_page(mapping, idx);
1040         if (filepage && PageUptodate(filepage))
1041                 goto done;
1042         error = 0;
1043         if (sgp == SGP_QUICK)
1044                 goto failed;
1045
1046         spin_lock(&info->lock);
1047         shmem_recalc_inode(inode);
1048         entry = shmem_swp_alloc(info, idx, sgp);
1049         if (IS_ERR(entry)) {
1050                 spin_unlock(&info->lock);
1051                 error = PTR_ERR(entry);
1052                 goto failed;
1053         }
1054         swap = *entry;
1055
1056         if (swap.val) {
1057                 /* Look it up and read it in.. */
1058                 swappage = lookup_swap_cache(swap);
1059                 if (!swappage) {
1060                         shmem_swp_unmap(entry);
1061                         /* here we actually do the io */
1062                         if (type && *type == VM_FAULT_MINOR) {
1063                                 __count_vm_event(PGMAJFAULT);
1064                                 *type = VM_FAULT_MAJOR;
1065                         }
1066                         spin_unlock(&info->lock);
1067                         swappage = shmem_swapin(info, swap, idx);
1068                         if (!swappage) {
1069                                 spin_lock(&info->lock);
1070                                 entry = shmem_swp_alloc(info, idx, sgp);
1071                                 if (IS_ERR(entry))
1072                                         error = PTR_ERR(entry);
1073                                 else {
1074                                         if (entry->val == swap.val)
1075                                                 error = -ENOMEM;
1076                                         shmem_swp_unmap(entry);
1077                                 }
1078                                 spin_unlock(&info->lock);
1079                                 if (error)
1080                                         goto failed;
1081                                 goto repeat;
1082                         }
1083                         wait_on_page_locked(swappage);
1084                         page_cache_release(swappage);
1085                         goto repeat;
1086                 }
1087
1088                 /* We have to do this with page locked to prevent races */
1089                 if (TestSetPageLocked(swappage)) {
1090                         shmem_swp_unmap(entry);
1091                         spin_unlock(&info->lock);
1092                         wait_on_page_locked(swappage);
1093                         page_cache_release(swappage);
1094                         goto repeat;
1095                 }
1096                 if (PageWriteback(swappage)) {
1097                         shmem_swp_unmap(entry);
1098                         spin_unlock(&info->lock);
1099                         wait_on_page_writeback(swappage);
1100                         unlock_page(swappage);
1101                         page_cache_release(swappage);
1102                         goto repeat;
1103                 }
1104                 if (!PageUptodate(swappage)) {
1105                         shmem_swp_unmap(entry);
1106                         spin_unlock(&info->lock);
1107                         unlock_page(swappage);
1108                         page_cache_release(swappage);
1109                         error = -EIO;
1110                         goto failed;
1111                 }
1112
1113                 if (filepage) {
1114                         shmem_swp_set(info, entry, 0);
1115                         shmem_swp_unmap(entry);
1116                         delete_from_swap_cache(swappage);
1117                         spin_unlock(&info->lock);
1118                         copy_highpage(filepage, swappage);
1119                         unlock_page(swappage);
1120                         page_cache_release(swappage);
1121                         flush_dcache_page(filepage);
1122                         SetPageUptodate(filepage);
1123                         set_page_dirty(filepage);
1124                         swap_free(swap);
1125                 } else if (!(error = move_from_swap_cache(
1126                                 swappage, idx, mapping))) {
1127                         info->flags |= SHMEM_PAGEIN;
1128                         shmem_swp_set(info, entry, 0);
1129                         shmem_swp_unmap(entry);
1130                         spin_unlock(&info->lock);
1131                         filepage = swappage;
1132                         swap_free(swap);
1133                 } else {
1134                         shmem_swp_unmap(entry);
1135                         spin_unlock(&info->lock);
1136                         unlock_page(swappage);
1137                         page_cache_release(swappage);
1138                         if (error == -ENOMEM) {
1139                                 /* let kswapd refresh zone for GFP_ATOMICs */
1140                                 congestion_wait(WRITE, HZ/50);
1141                         }
1142                         goto repeat;
1143                 }
1144         } else if (sgp == SGP_READ && !filepage) {
1145                 shmem_swp_unmap(entry);
1146                 filepage = find_get_page(mapping, idx);
1147                 if (filepage &&
1148                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1149                         spin_unlock(&info->lock);
1150                         wait_on_page_locked(filepage);
1151                         page_cache_release(filepage);
1152                         filepage = NULL;
1153                         goto repeat;
1154                 }
1155                 spin_unlock(&info->lock);
1156         } else {
1157                 shmem_swp_unmap(entry);
1158                 sbinfo = SHMEM_SB(inode->i_sb);
1159                 if (sbinfo->max_blocks) {
1160                         spin_lock(&sbinfo->stat_lock);
1161                         if (sbinfo->free_blocks == 0 ||
1162                             shmem_acct_block(info->flags)) {
1163                                 spin_unlock(&sbinfo->stat_lock);
1164                                 spin_unlock(&info->lock);
1165                                 error = -ENOSPC;
1166                                 goto failed;
1167                         }
1168                         sbinfo->free_blocks--;
1169                         inode->i_blocks += BLOCKS_PER_PAGE;
1170                         spin_unlock(&sbinfo->stat_lock);
1171                 } else if (shmem_acct_block(info->flags)) {
1172                         spin_unlock(&info->lock);
1173                         error = -ENOSPC;
1174                         goto failed;
1175                 }
1176
1177                 if (!filepage) {
1178                         spin_unlock(&info->lock);
1179                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1180                                                     info,
1181                                                     idx);
1182                         if (!filepage) {
1183                                 shmem_unacct_blocks(info->flags, 1);
1184                                 shmem_free_blocks(inode, 1);
1185                                 error = -ENOMEM;
1186                                 goto failed;
1187                         }
1188
1189                         spin_lock(&info->lock);
1190                         entry = shmem_swp_alloc(info, idx, sgp);
1191                         if (IS_ERR(entry))
1192                                 error = PTR_ERR(entry);
1193                         else {
1194                                 swap = *entry;
1195                                 shmem_swp_unmap(entry);
1196                         }
1197                         if (error || swap.val || 0 != add_to_page_cache_lru(
1198                                         filepage, mapping, idx, GFP_ATOMIC)) {
1199                                 spin_unlock(&info->lock);
1200                                 page_cache_release(filepage);
1201                                 shmem_unacct_blocks(info->flags, 1);
1202                                 shmem_free_blocks(inode, 1);
1203                                 filepage = NULL;
1204                                 if (error)
1205                                         goto failed;
1206                                 goto repeat;
1207                         }
1208                         info->flags |= SHMEM_PAGEIN;
1209                 }
1210
1211                 info->alloced++;
1212                 spin_unlock(&info->lock);
1213                 flush_dcache_page(filepage);
1214                 SetPageUptodate(filepage);
1215         }
1216 done:
1217         if (*pagep != filepage) {
1218                 unlock_page(filepage);
1219                 *pagep = filepage;
1220         }
1221         return 0;
1222
1223 failed:
1224         if (*pagep != filepage) {
1225                 unlock_page(filepage);
1226                 page_cache_release(filepage);
1227         }
1228         return error;
1229 }
1230
1231 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1232 {
1233         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1234         struct page *page = NULL;
1235         unsigned long idx;
1236         int error;
1237
1238         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1239         idx += vma->vm_pgoff;
1240         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1241         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1242                 return NOPAGE_SIGBUS;
1243
1244         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1245         if (error)
1246                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1247
1248         mark_page_accessed(page);
1249         return page;
1250 }
1251
1252 static int shmem_populate(struct vm_area_struct *vma,
1253         unsigned long addr, unsigned long len,
1254         pgprot_t prot, unsigned long pgoff, int nonblock)
1255 {
1256         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1257         struct mm_struct *mm = vma->vm_mm;
1258         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1259         unsigned long size;
1260
1261         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1262         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1263                 return -EINVAL;
1264
1265         while ((long) len > 0) {
1266                 struct page *page = NULL;
1267                 int err;
1268                 /*
1269                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1270                  */
1271                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1272                 if (err)
1273                         return err;
1274                 /* Page may still be null, but only if nonblock was set. */
1275                 if (page) {
1276                         mark_page_accessed(page);
1277                         err = install_page(mm, vma, addr, page, prot);
1278                         if (err) {
1279                                 page_cache_release(page);
1280                                 return err;
1281                         }
1282                 } else if (vma->vm_flags & VM_NONLINEAR) {
1283                         /* No page was found just because we can't read it in
1284                          * now (being here implies nonblock != 0), but the page
1285                          * may exist, so set the PTE to fault it in later. */
1286                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1287                         if (err)
1288                                 return err;
1289                 }
1290
1291                 len -= PAGE_SIZE;
1292                 addr += PAGE_SIZE;
1293                 pgoff++;
1294         }
1295         return 0;
1296 }
1297
1298 #ifdef CONFIG_NUMA
1299 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1300 {
1301         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1302         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1303 }
1304
1305 struct mempolicy *
1306 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1307 {
1308         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1309         unsigned long idx;
1310
1311         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1312         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1313 }
1314 #endif
1315
1316 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1317 {
1318         struct inode *inode = file->f_path.dentry->d_inode;
1319         struct shmem_inode_info *info = SHMEM_I(inode);
1320         int retval = -ENOMEM;
1321
1322         spin_lock(&info->lock);
1323         if (lock && !(info->flags & VM_LOCKED)) {
1324                 if (!user_shm_lock(inode->i_size, user))
1325                         goto out_nomem;
1326                 info->flags |= VM_LOCKED;
1327         }
1328         if (!lock && (info->flags & VM_LOCKED) && user) {
1329                 user_shm_unlock(inode->i_size, user);
1330                 info->flags &= ~VM_LOCKED;
1331         }
1332         retval = 0;
1333 out_nomem:
1334         spin_unlock(&info->lock);
1335         return retval;
1336 }
1337
1338 int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1339 {
1340         file_accessed(file);
1341         vma->vm_ops = &shmem_vm_ops;
1342         return 0;
1343 }
1344
1345 static struct inode *
1346 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1347 {
1348         struct inode *inode;
1349         struct shmem_inode_info *info;
1350         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1351
1352         if (sbinfo->max_inodes) {
1353                 spin_lock(&sbinfo->stat_lock);
1354                 if (!sbinfo->free_inodes) {
1355                         spin_unlock(&sbinfo->stat_lock);
1356                         return NULL;
1357                 }
1358                 sbinfo->free_inodes--;
1359                 spin_unlock(&sbinfo->stat_lock);
1360         }
1361
1362         inode = new_inode(sb);
1363         if (inode) {
1364                 inode->i_mode = mode;
1365                 inode->i_uid = current->fsuid;
1366                 inode->i_gid = current->fsgid;
1367                 inode->i_blocks = 0;
1368                 inode->i_mapping->a_ops = &shmem_aops;
1369                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1370                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1371                 inode->i_generation = get_seconds();
1372                 info = SHMEM_I(inode);
1373                 memset(info, 0, (char *)inode - (char *)info);
1374                 spin_lock_init(&info->lock);
1375                 INIT_LIST_HEAD(&info->swaplist);
1376
1377                 switch (mode & S_IFMT) {
1378                 default:
1379                         inode->i_op = &shmem_special_inode_operations;
1380                         init_special_inode(inode, mode, dev);
1381                         break;
1382                 case S_IFREG:
1383                         inode->i_op = &shmem_inode_operations;
1384                         inode->i_fop = &shmem_file_operations;
1385                         mpol_shared_policy_init(&info->policy, sbinfo->policy,
1386                                                         &sbinfo->policy_nodes);
1387                         break;
1388                 case S_IFDIR:
1389                         inc_nlink(inode);
1390                         /* Some things misbehave if size == 0 on a directory */
1391                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1392                         inode->i_op = &shmem_dir_inode_operations;
1393                         inode->i_fop = &simple_dir_operations;
1394                         break;
1395                 case S_IFLNK:
1396                         /*
1397                          * Must not load anything in the rbtree,
1398                          * mpol_free_shared_policy will not be called.
1399                          */
1400                         mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1401                                                 NULL);
1402                         break;
1403                 }
1404         } else if (sbinfo->max_inodes) {
1405                 spin_lock(&sbinfo->stat_lock);
1406                 sbinfo->free_inodes++;
1407                 spin_unlock(&sbinfo->stat_lock);
1408         }
1409         return inode;
1410 }
1411
1412 #ifdef CONFIG_TMPFS
1413 static const struct inode_operations shmem_symlink_inode_operations;
1414 static const struct inode_operations shmem_symlink_inline_operations;
1415
1416 /*
1417  * Normally tmpfs makes no use of shmem_prepare_write, but it
1418  * lets a tmpfs file be used read-write below the loop driver.
1419  */
1420 static int
1421 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1422 {
1423         struct inode *inode = page->mapping->host;
1424         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1425 }
1426
1427 static ssize_t
1428 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1429 {
1430         struct inode    *inode = file->f_path.dentry->d_inode;
1431         loff_t          pos;
1432         unsigned long   written;
1433         ssize_t         err;
1434
1435         if ((ssize_t) count < 0)
1436                 return -EINVAL;
1437
1438         if (!access_ok(VERIFY_READ, buf, count))
1439                 return -EFAULT;
1440
1441         mutex_lock(&inode->i_mutex);
1442
1443         pos = *ppos;
1444         written = 0;
1445
1446         err = generic_write_checks(file, &pos, &count, 0);
1447         if (err || !count)
1448                 goto out;
1449
1450         err = remove_suid(file->f_path.dentry);
1451         if (err)
1452                 goto out;
1453
1454         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1455
1456         do {
1457                 struct page *page = NULL;
1458                 unsigned long bytes, index, offset;
1459                 char *kaddr;
1460                 int left;
1461
1462                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1463                 index = pos >> PAGE_CACHE_SHIFT;
1464                 bytes = PAGE_CACHE_SIZE - offset;
1465                 if (bytes > count)
1466                         bytes = count;
1467
1468                 /*
1469                  * We don't hold page lock across copy from user -
1470                  * what would it guard against? - so no deadlock here.
1471                  * But it still may be a good idea to prefault below.
1472                  */
1473
1474                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1475                 if (err)
1476                         break;
1477
1478                 left = bytes;
1479                 if (PageHighMem(page)) {
1480                         volatile unsigned char dummy;
1481                         __get_user(dummy, buf);
1482                         __get_user(dummy, buf + bytes - 1);
1483
1484                         kaddr = kmap_atomic(page, KM_USER0);
1485                         left = __copy_from_user_inatomic(kaddr + offset,
1486                                                         buf, bytes);
1487                         kunmap_atomic(kaddr, KM_USER0);
1488                 }
1489                 if (left) {
1490                         kaddr = kmap(page);
1491                         left = __copy_from_user(kaddr + offset, buf, bytes);
1492                         kunmap(page);
1493                 }
1494
1495                 written += bytes;
1496                 count -= bytes;
1497                 pos += bytes;
1498                 buf += bytes;
1499                 if (pos > inode->i_size)
1500                         i_size_write(inode, pos);
1501
1502                 flush_dcache_page(page);
1503                 set_page_dirty(page);
1504                 mark_page_accessed(page);
1505                 page_cache_release(page);
1506
1507                 if (left) {
1508                         pos -= left;
1509                         written -= left;
1510                         err = -EFAULT;
1511                         break;
1512                 }
1513
1514                 /*
1515                  * Our dirty pages are not counted in nr_dirty,
1516                  * and we do not attempt to balance dirty pages.
1517                  */
1518
1519                 cond_resched();
1520         } while (count);
1521
1522         *ppos = pos;
1523         if (written)
1524                 err = written;
1525 out:
1526         mutex_unlock(&inode->i_mutex);
1527         return err;
1528 }
1529
1530 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1531 {
1532         struct inode *inode = filp->f_path.dentry->d_inode;
1533         struct address_space *mapping = inode->i_mapping;
1534         unsigned long index, offset;
1535
1536         index = *ppos >> PAGE_CACHE_SHIFT;
1537         offset = *ppos & ~PAGE_CACHE_MASK;
1538
1539         for (;;) {
1540                 struct page *page = NULL;
1541                 unsigned long end_index, nr, ret;
1542                 loff_t i_size = i_size_read(inode);
1543
1544                 end_index = i_size >> PAGE_CACHE_SHIFT;
1545                 if (index > end_index)
1546                         break;
1547                 if (index == end_index) {
1548                         nr = i_size & ~PAGE_CACHE_MASK;
1549                         if (nr <= offset)
1550                                 break;
1551                 }
1552
1553                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1554                 if (desc->error) {
1555                         if (desc->error == -EINVAL)
1556                                 desc->error = 0;
1557                         break;
1558                 }
1559
1560                 /*
1561                  * We must evaluate after, since reads (unlike writes)
1562                  * are called without i_mutex protection against truncate
1563                  */
1564                 nr = PAGE_CACHE_SIZE;
1565                 i_size = i_size_read(inode);
1566                 end_index = i_size >> PAGE_CACHE_SHIFT;
1567                 if (index == end_index) {
1568                         nr = i_size & ~PAGE_CACHE_MASK;
1569                         if (nr <= offset) {
1570                                 if (page)
1571                                         page_cache_release(page);
1572                                 break;
1573                         }
1574                 }
1575                 nr -= offset;
1576
1577                 if (page) {
1578                         /*
1579                          * If users can be writing to this page using arbitrary
1580                          * virtual addresses, take care about potential aliasing
1581                          * before reading the page on the kernel side.
1582                          */
1583                         if (mapping_writably_mapped(mapping))
1584                                 flush_dcache_page(page);
1585                         /*
1586                          * Mark the page accessed if we read the beginning.
1587                          */
1588                         if (!offset)
1589                                 mark_page_accessed(page);
1590                 } else {
1591                         page = ZERO_PAGE(0);
1592                         page_cache_get(page);
1593                 }
1594
1595                 /*
1596                  * Ok, we have the page, and it's up-to-date, so
1597                  * now we can copy it to user space...
1598                  *
1599                  * The actor routine returns how many bytes were actually used..
1600                  * NOTE! This may not be the same as how much of a user buffer
1601                  * we filled up (we may be padding etc), so we can only update
1602                  * "pos" here (the actor routine has to update the user buffer
1603                  * pointers and the remaining count).
1604                  */
1605                 ret = actor(desc, page, offset, nr);
1606                 offset += ret;
1607                 index += offset >> PAGE_CACHE_SHIFT;
1608                 offset &= ~PAGE_CACHE_MASK;
1609
1610                 page_cache_release(page);
1611                 if (ret != nr || !desc->count)
1612                         break;
1613
1614                 cond_resched();
1615         }
1616
1617         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1618         file_accessed(filp);
1619 }
1620
1621 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1622 {
1623         read_descriptor_t desc;
1624
1625         if ((ssize_t) count < 0)
1626                 return -EINVAL;
1627         if (!access_ok(VERIFY_WRITE, buf, count))
1628                 return -EFAULT;
1629         if (!count)
1630                 return 0;
1631
1632         desc.written = 0;
1633         desc.count = count;
1634         desc.arg.buf = buf;
1635         desc.error = 0;
1636
1637         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1638         if (desc.written)
1639                 return desc.written;
1640         return desc.error;
1641 }
1642
1643 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1644                          size_t count, read_actor_t actor, void *target)
1645 {
1646         read_descriptor_t desc;
1647
1648         if (!count)
1649                 return 0;
1650
1651         desc.written = 0;
1652         desc.count = count;
1653         desc.arg.data = target;
1654         desc.error = 0;
1655
1656         do_shmem_file_read(in_file, ppos, &desc, actor);
1657         if (desc.written)
1658                 return desc.written;
1659         return desc.error;
1660 }
1661
1662 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1663 {
1664         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1665
1666         buf->f_type = TMPFS_MAGIC;
1667         buf->f_bsize = PAGE_CACHE_SIZE;
1668         buf->f_namelen = NAME_MAX;
1669         spin_lock(&sbinfo->stat_lock);
1670         if (sbinfo->max_blocks) {
1671                 buf->f_blocks = sbinfo->max_blocks;
1672                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1673         }
1674         if (sbinfo->max_inodes) {
1675                 buf->f_files = sbinfo->max_inodes;
1676                 buf->f_ffree = sbinfo->free_inodes;
1677         }
1678         /* else leave those fields 0 like simple_statfs */
1679         spin_unlock(&sbinfo->stat_lock);
1680         return 0;
1681 }
1682
1683 /*
1684  * File creation. Allocate an inode, and we're done..
1685  */
1686 static int
1687 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1688 {
1689         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1690         int error = -ENOSPC;
1691
1692         if (inode) {
1693                 error = security_inode_init_security(inode, dir, NULL, NULL,
1694                                                      NULL);
1695                 if (error) {
1696                         if (error != -EOPNOTSUPP) {
1697                                 iput(inode);
1698                                 return error;
1699                         }
1700                 }
1701                 error = shmem_acl_init(inode, dir);
1702                 if (error) {
1703                         iput(inode);
1704                         return error;
1705                 }
1706                 if (dir->i_mode & S_ISGID) {
1707                         inode->i_gid = dir->i_gid;
1708                         if (S_ISDIR(mode))
1709                                 inode->i_mode |= S_ISGID;
1710                 }
1711                 dir->i_size += BOGO_DIRENT_SIZE;
1712                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1713                 d_instantiate(dentry, inode);
1714                 dget(dentry); /* Extra count - pin the dentry in core */
1715         }
1716         return error;
1717 }
1718
1719 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1720 {
1721         int error;
1722
1723         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1724                 return error;
1725         inc_nlink(dir);
1726         return 0;
1727 }
1728
1729 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1730                 struct nameidata *nd)
1731 {
1732         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1733 }
1734
1735 /*
1736  * Link a file..
1737  */
1738 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1739 {
1740         struct inode *inode = old_dentry->d_inode;
1741         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1742
1743         /*
1744          * No ordinary (disk based) filesystem counts links as inodes;
1745          * but each new link needs a new dentry, pinning lowmem, and
1746          * tmpfs dentries cannot be pruned until they are unlinked.
1747          */
1748         if (sbinfo->max_inodes) {
1749                 spin_lock(&sbinfo->stat_lock);
1750                 if (!sbinfo->free_inodes) {
1751                         spin_unlock(&sbinfo->stat_lock);
1752                         return -ENOSPC;
1753                 }
1754                 sbinfo->free_inodes--;
1755                 spin_unlock(&sbinfo->stat_lock);
1756         }
1757
1758         dir->i_size += BOGO_DIRENT_SIZE;
1759         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1760         inc_nlink(inode);
1761         atomic_inc(&inode->i_count);    /* New dentry reference */
1762         dget(dentry);           /* Extra pinning count for the created dentry */
1763         d_instantiate(dentry, inode);
1764         return 0;
1765 }
1766
1767 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1768 {
1769         struct inode *inode = dentry->d_inode;
1770
1771         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1772                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1773                 if (sbinfo->max_inodes) {
1774                         spin_lock(&sbinfo->stat_lock);
1775                         sbinfo->free_inodes++;
1776                         spin_unlock(&sbinfo->stat_lock);
1777                 }
1778         }
1779
1780         dir->i_size -= BOGO_DIRENT_SIZE;
1781         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1782         drop_nlink(inode);
1783         dput(dentry);   /* Undo the count from "create" - this does all the work */
1784         return 0;
1785 }
1786
1787 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1788 {
1789         if (!simple_empty(dentry))
1790                 return -ENOTEMPTY;
1791
1792         drop_nlink(dentry->d_inode);
1793         drop_nlink(dir);
1794         return shmem_unlink(dir, dentry);
1795 }
1796
1797 /*
1798  * The VFS layer already does all the dentry stuff for rename,
1799  * we just have to decrement the usage count for the target if
1800  * it exists so that the VFS layer correctly free's it when it
1801  * gets overwritten.
1802  */
1803 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1804 {
1805         struct inode *inode = old_dentry->d_inode;
1806         int they_are_dirs = S_ISDIR(inode->i_mode);
1807
1808         if (!simple_empty(new_dentry))
1809                 return -ENOTEMPTY;
1810
1811         if (new_dentry->d_inode) {
1812                 (void) shmem_unlink(new_dir, new_dentry);
1813                 if (they_are_dirs)
1814                         drop_nlink(old_dir);
1815         } else if (they_are_dirs) {
1816                 drop_nlink(old_dir);
1817                 inc_nlink(new_dir);
1818         }
1819
1820         old_dir->i_size -= BOGO_DIRENT_SIZE;
1821         new_dir->i_size += BOGO_DIRENT_SIZE;
1822         old_dir->i_ctime = old_dir->i_mtime =
1823         new_dir->i_ctime = new_dir->i_mtime =
1824         inode->i_ctime = CURRENT_TIME;
1825         return 0;
1826 }
1827
1828 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1829 {
1830         int error;
1831         int len;
1832         struct inode *inode;
1833         struct page *page = NULL;
1834         char *kaddr;
1835         struct shmem_inode_info *info;
1836
1837         len = strlen(symname) + 1;
1838         if (len > PAGE_CACHE_SIZE)
1839                 return -ENAMETOOLONG;
1840
1841         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1842         if (!inode)
1843                 return -ENOSPC;
1844
1845         error = security_inode_init_security(inode, dir, NULL, NULL,
1846                                              NULL);
1847         if (error) {
1848                 if (error != -EOPNOTSUPP) {
1849                         iput(inode);
1850                         return error;
1851                 }
1852                 error = 0;
1853         }
1854
1855         info = SHMEM_I(inode);
1856         inode->i_size = len-1;
1857         if (len <= (char *)inode - (char *)info) {
1858                 /* do it inline */
1859                 memcpy(info, symname, len);
1860                 inode->i_op = &shmem_symlink_inline_operations;
1861         } else {
1862                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1863                 if (error) {
1864                         iput(inode);
1865                         return error;
1866                 }
1867                 inode->i_op = &shmem_symlink_inode_operations;
1868                 kaddr = kmap_atomic(page, KM_USER0);
1869                 memcpy(kaddr, symname, len);
1870                 kunmap_atomic(kaddr, KM_USER0);
1871                 set_page_dirty(page);
1872                 page_cache_release(page);
1873         }
1874         if (dir->i_mode & S_ISGID)
1875                 inode->i_gid = dir->i_gid;
1876         dir->i_size += BOGO_DIRENT_SIZE;
1877         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1878         d_instantiate(dentry, inode);
1879         dget(dentry);
1880         return 0;
1881 }
1882
1883 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1884 {
1885         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1886         return NULL;
1887 }
1888
1889 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1890 {
1891         struct page *page = NULL;
1892         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1893         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1894         return page;
1895 }
1896
1897 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1898 {
1899         if (!IS_ERR(nd_get_link(nd))) {
1900                 struct page *page = cookie;
1901                 kunmap(page);
1902                 mark_page_accessed(page);
1903                 page_cache_release(page);
1904         }
1905 }
1906
1907 static const struct inode_operations shmem_symlink_inline_operations = {
1908         .readlink       = generic_readlink,
1909         .follow_link    = shmem_follow_link_inline,
1910 };
1911
1912 static const struct inode_operations shmem_symlink_inode_operations = {
1913         .truncate       = shmem_truncate,
1914         .readlink       = generic_readlink,
1915         .follow_link    = shmem_follow_link,
1916         .put_link       = shmem_put_link,
1917 };
1918
1919 #ifdef CONFIG_TMPFS_POSIX_ACL
1920 /**
1921  * Superblocks without xattr inode operations will get security.* xattr
1922  * support from the VFS "for free". As soon as we have any other xattrs
1923  * like ACLs, we also need to implement the security.* handlers at
1924  * filesystem level, though.
1925  */
1926
1927 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1928                                         size_t list_len, const char *name,
1929                                         size_t name_len)
1930 {
1931         return security_inode_listsecurity(inode, list, list_len);
1932 }
1933
1934 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1935                                     void *buffer, size_t size)
1936 {
1937         if (strcmp(name, "") == 0)
1938                 return -EINVAL;
1939         return security_inode_getsecurity(inode, name, buffer, size,
1940                                           -EOPNOTSUPP);
1941 }
1942
1943 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1944                                     const void *value, size_t size, int flags)
1945 {
1946         if (strcmp(name, "") == 0)
1947                 return -EINVAL;
1948         return security_inode_setsecurity(inode, name, value, size, flags);
1949 }
1950
1951 static struct xattr_handler shmem_xattr_security_handler = {
1952         .prefix = XATTR_SECURITY_PREFIX,
1953         .list   = shmem_xattr_security_list,
1954         .get    = shmem_xattr_security_get,
1955         .set    = shmem_xattr_security_set,
1956 };
1957
1958 static struct xattr_handler *shmem_xattr_handlers[] = {
1959         &shmem_xattr_acl_access_handler,
1960         &shmem_xattr_acl_default_handler,
1961         &shmem_xattr_security_handler,
1962         NULL
1963 };
1964 #endif
1965
1966 static struct dentry *shmem_get_parent(struct dentry *child)
1967 {
1968         return ERR_PTR(-ESTALE);
1969 }
1970
1971 static int shmem_match(struct inode *ino, void *vfh)
1972 {
1973         __u32 *fh = vfh;
1974         __u64 inum = fh[2];
1975         inum = (inum << 32) | fh[1];
1976         return ino->i_ino == inum && fh[0] == ino->i_generation;
1977 }
1978
1979 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
1980 {
1981         struct dentry *de = NULL;
1982         struct inode *inode;
1983         __u32 *fh = vfh;
1984         __u64 inum = fh[2];
1985         inum = (inum << 32) | fh[1];
1986
1987         inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
1988         if (inode) {
1989                 de = d_find_alias(inode);
1990                 iput(inode);
1991         }
1992
1993         return de? de: ERR_PTR(-ESTALE);
1994 }
1995
1996 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
1997                 int len, int type,
1998                 int (*acceptable)(void *context, struct dentry *de),
1999                 void *context)
2000 {
2001         if (len < 3)
2002                 return ERR_PTR(-ESTALE);
2003
2004         return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2005                                                         context);
2006 }
2007
2008 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2009                                 int connectable)
2010 {
2011         struct inode *inode = dentry->d_inode;
2012
2013         if (*len < 3)
2014                 return 255;
2015
2016         if (hlist_unhashed(&inode->i_hash)) {
2017                 /* Unfortunately insert_inode_hash is not idempotent,
2018                  * so as we hash inodes here rather than at creation
2019                  * time, we need a lock to ensure we only try
2020                  * to do it once
2021                  */
2022                 static DEFINE_SPINLOCK(lock);
2023                 spin_lock(&lock);
2024                 if (hlist_unhashed(&inode->i_hash))
2025                         __insert_inode_hash(inode,
2026                                             inode->i_ino + inode->i_generation);
2027                 spin_unlock(&lock);
2028         }
2029
2030         fh[0] = inode->i_generation;
2031         fh[1] = inode->i_ino;
2032         fh[2] = ((__u64)inode->i_ino) >> 32;
2033
2034         *len = 3;
2035         return 1;
2036 }
2037
2038 static struct export_operations shmem_export_ops = {
2039         .get_parent     = shmem_get_parent,
2040         .get_dentry     = shmem_get_dentry,
2041         .encode_fh      = shmem_encode_fh,
2042         .decode_fh      = shmem_decode_fh,
2043 };
2044
2045 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2046         gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2047         int *policy, nodemask_t *policy_nodes)
2048 {
2049         char *this_char, *value, *rest;
2050
2051         while (options != NULL) {
2052                 this_char = options;
2053                 for (;;) {
2054                         /*
2055                          * NUL-terminate this option: unfortunately,
2056                          * mount options form a comma-separated list,
2057                          * but mpol's nodelist may also contain commas.
2058                          */
2059                         options = strchr(options, ',');
2060                         if (options == NULL)
2061                                 break;
2062                         options++;
2063                         if (!isdigit(*options)) {
2064                                 options[-1] = '\0';
2065                                 break;
2066                         }
2067                 }
2068                 if (!*this_char)
2069                         continue;
2070                 if ((value = strchr(this_char,'=')) != NULL) {
2071                         *value++ = 0;
2072                 } else {
2073                         printk(KERN_ERR
2074                             "tmpfs: No value for mount option '%s'\n",
2075                             this_char);
2076                         return 1;
2077                 }
2078
2079                 if (!strcmp(this_char,"size")) {
2080                         unsigned long long size;
2081                         size = memparse(value,&rest);
2082                         if (*rest == '%') {
2083                                 size <<= PAGE_SHIFT;
2084                                 size *= totalram_pages;
2085                                 do_div(size, 100);
2086                                 rest++;
2087                         }
2088                         if (*rest)
2089                                 goto bad_val;
2090                         *blocks = size >> PAGE_CACHE_SHIFT;
2091                 } else if (!strcmp(this_char,"nr_blocks")) {
2092                         *blocks = memparse(value,&rest);
2093                         if (*rest)
2094                                 goto bad_val;
2095                 } else if (!strcmp(this_char,"nr_inodes")) {
2096                         *inodes = memparse(value,&rest);
2097                         if (*rest)
2098                                 goto bad_val;
2099                 } else if (!strcmp(this_char,"mode")) {
2100                         if (!mode)
2101                                 continue;
2102                         *mode = simple_strtoul(value,&rest,8);
2103                         if (*rest)
2104                                 goto bad_val;
2105                 } else if (!strcmp(this_char,"uid")) {
2106                         if (!uid)
2107                                 continue;
2108                         *uid = simple_strtoul(value,&rest,0);
2109                         if (*rest)
2110                                 goto bad_val;
2111                 } else if (!strcmp(this_char,"gid")) {
2112                         if (!gid)
2113                                 continue;
2114                         *gid = simple_strtoul(value,&rest,0);
2115                         if (*rest)
2116                                 goto bad_val;
2117                 } else if (!strcmp(this_char,"mpol")) {
2118                         if (shmem_parse_mpol(value,policy,policy_nodes))
2119                                 goto bad_val;
2120                 } else {
2121                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2122                                this_char);
2123                         return 1;
2124                 }
2125         }
2126         return 0;
2127
2128 bad_val:
2129         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2130                value, this_char);
2131         return 1;
2132
2133 }
2134
2135 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2136 {
2137         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2138         unsigned long max_blocks = sbinfo->max_blocks;
2139         unsigned long max_inodes = sbinfo->max_inodes;
2140         int policy = sbinfo->policy;
2141         nodemask_t policy_nodes = sbinfo->policy_nodes;
2142         unsigned long blocks;
2143         unsigned long inodes;
2144         int error = -EINVAL;
2145
2146         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2147                                 &max_inodes, &policy, &policy_nodes))
2148                 return error;
2149
2150         spin_lock(&sbinfo->stat_lock);
2151         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2152         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2153         if (max_blocks < blocks)
2154                 goto out;
2155         if (max_inodes < inodes)
2156                 goto out;
2157         /*
2158          * Those tests also disallow limited->unlimited while any are in
2159          * use, so i_blocks will always be zero when max_blocks is zero;
2160          * but we must separately disallow unlimited->limited, because
2161          * in that case we have no record of how much is already in use.
2162          */
2163         if (max_blocks && !sbinfo->max_blocks)
2164                 goto out;
2165         if (max_inodes && !sbinfo->max_inodes)
2166                 goto out;
2167
2168         error = 0;
2169         sbinfo->max_blocks  = max_blocks;
2170         sbinfo->free_blocks = max_blocks - blocks;
2171         sbinfo->max_inodes  = max_inodes;
2172         sbinfo->free_inodes = max_inodes - inodes;
2173         sbinfo->policy = policy;
2174         sbinfo->policy_nodes = policy_nodes;
2175 out:
2176         spin_unlock(&sbinfo->stat_lock);
2177         return error;
2178 }
2179 #endif
2180
2181 static void shmem_put_super(struct super_block *sb)
2182 {
2183         kfree(sb->s_fs_info);
2184         sb->s_fs_info = NULL;
2185 }
2186
2187 static int shmem_fill_super(struct super_block *sb,
2188                             void *data, int silent)
2189 {
2190         struct inode *inode;
2191         struct dentry *root;
2192         int mode   = S_IRWXUGO | S_ISVTX;
2193         uid_t uid = current->fsuid;
2194         gid_t gid = current->fsgid;
2195         int err = -ENOMEM;
2196         struct shmem_sb_info *sbinfo;
2197         unsigned long blocks = 0;
2198         unsigned long inodes = 0;
2199         int policy = MPOL_DEFAULT;
2200         nodemask_t policy_nodes = node_online_map;
2201
2202 #ifdef CONFIG_TMPFS
2203         /*
2204          * Per default we only allow half of the physical ram per
2205          * tmpfs instance, limiting inodes to one per page of lowmem;
2206          * but the internal instance is left unlimited.
2207          */
2208         if (!(sb->s_flags & MS_NOUSER)) {
2209                 blocks = totalram_pages / 2;
2210                 inodes = totalram_pages - totalhigh_pages;
2211                 if (inodes > blocks)
2212                         inodes = blocks;
2213                 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2214                                         &inodes, &policy, &policy_nodes))
2215                         return -EINVAL;
2216         }
2217         sb->s_export_op = &shmem_export_ops;
2218 #else
2219         sb->s_flags |= MS_NOUSER;
2220 #endif
2221
2222         /* Round up to L1_CACHE_BYTES to resist false sharing */
2223         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2224                                 L1_CACHE_BYTES), GFP_KERNEL);
2225         if (!sbinfo)
2226                 return -ENOMEM;
2227
2228         spin_lock_init(&sbinfo->stat_lock);
2229         sbinfo->max_blocks = blocks;
2230         sbinfo->free_blocks = blocks;
2231         sbinfo->max_inodes = inodes;
2232         sbinfo->free_inodes = inodes;
2233         sbinfo->policy = policy;
2234         sbinfo->policy_nodes = policy_nodes;
2235
2236         sb->s_fs_info = sbinfo;
2237         sb->s_maxbytes = SHMEM_MAX_BYTES;
2238         sb->s_blocksize = PAGE_CACHE_SIZE;
2239         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2240         sb->s_magic = TMPFS_MAGIC;
2241         sb->s_op = &shmem_ops;
2242         sb->s_time_gran = 1;
2243 #ifdef CONFIG_TMPFS_POSIX_ACL
2244         sb->s_xattr = shmem_xattr_handlers;
2245         sb->s_flags |= MS_POSIXACL;
2246 #endif
2247
2248         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2249         if (!inode)
2250                 goto failed;
2251         inode->i_uid = uid;
2252         inode->i_gid = gid;
2253         root = d_alloc_root(inode);
2254         if (!root)
2255                 goto failed_iput;
2256         sb->s_root = root;
2257         return 0;
2258
2259 failed_iput:
2260         iput(inode);
2261 failed:
2262         shmem_put_super(sb);
2263         return err;
2264 }
2265
2266 static struct kmem_cache *shmem_inode_cachep;
2267
2268 static struct inode *shmem_alloc_inode(struct super_block *sb)
2269 {
2270         struct shmem_inode_info *p;
2271         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2272         if (!p)
2273                 return NULL;
2274         return &p->vfs_inode;
2275 }
2276
2277 static void shmem_destroy_inode(struct inode *inode)
2278 {
2279         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2280                 /* only struct inode is valid if it's an inline symlink */
2281                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2282         }
2283         shmem_acl_destroy_inode(inode);
2284         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2285 }
2286
2287 static void init_once(void *foo, struct kmem_cache *cachep,
2288                       unsigned long flags)
2289 {
2290         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2291
2292         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2293             SLAB_CTOR_CONSTRUCTOR) {
2294                 inode_init_once(&p->vfs_inode);
2295 #ifdef CONFIG_TMPFS_POSIX_ACL
2296                 p->i_acl = NULL;
2297                 p->i_default_acl = NULL;
2298 #endif
2299         }
2300 }
2301
2302 static int init_inodecache(void)
2303 {
2304         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2305                                 sizeof(struct shmem_inode_info),
2306                                 0, 0, init_once, NULL);
2307         if (shmem_inode_cachep == NULL)
2308                 return -ENOMEM;
2309         return 0;
2310 }
2311
2312 static void destroy_inodecache(void)
2313 {
2314         kmem_cache_destroy(shmem_inode_cachep);
2315 }
2316
2317 static const struct address_space_operations shmem_aops = {
2318         .writepage      = shmem_writepage,
2319         .set_page_dirty = __set_page_dirty_no_writeback,
2320 #ifdef CONFIG_TMPFS
2321         .prepare_write  = shmem_prepare_write,
2322         .commit_write   = simple_commit_write,
2323 #endif
2324         .migratepage    = migrate_page,
2325 };
2326
2327 static const struct file_operations shmem_file_operations = {
2328         .mmap           = shmem_mmap,
2329 #ifdef CONFIG_TMPFS
2330         .llseek         = generic_file_llseek,
2331         .read           = shmem_file_read,
2332         .write          = shmem_file_write,
2333         .fsync          = simple_sync_file,
2334         .sendfile       = shmem_file_sendfile,
2335 #endif
2336 };
2337
2338 static const struct inode_operations shmem_inode_operations = {
2339         .truncate       = shmem_truncate,
2340         .setattr        = shmem_notify_change,
2341         .truncate_range = shmem_truncate_range,
2342 #ifdef CONFIG_TMPFS_POSIX_ACL
2343         .setxattr       = generic_setxattr,
2344         .getxattr       = generic_getxattr,
2345         .listxattr      = generic_listxattr,
2346         .removexattr    = generic_removexattr,
2347         .permission     = shmem_permission,
2348 #endif
2349
2350 };
2351
2352 static const struct inode_operations shmem_dir_inode_operations = {
2353 #ifdef CONFIG_TMPFS
2354         .create         = shmem_create,
2355         .lookup         = simple_lookup,
2356         .link           = shmem_link,
2357         .unlink         = shmem_unlink,
2358         .symlink        = shmem_symlink,
2359         .mkdir          = shmem_mkdir,
2360         .rmdir          = shmem_rmdir,
2361         .mknod          = shmem_mknod,
2362         .rename         = shmem_rename,
2363 #endif
2364 #ifdef CONFIG_TMPFS_POSIX_ACL
2365         .setattr        = shmem_notify_change,
2366         .setxattr       = generic_setxattr,
2367         .getxattr       = generic_getxattr,
2368         .listxattr      = generic_listxattr,
2369         .removexattr    = generic_removexattr,
2370         .permission     = shmem_permission,
2371 #endif
2372 };
2373
2374 static const struct inode_operations shmem_special_inode_operations = {
2375 #ifdef CONFIG_TMPFS_POSIX_ACL
2376         .setattr        = shmem_notify_change,
2377         .setxattr       = generic_setxattr,
2378         .getxattr       = generic_getxattr,
2379         .listxattr      = generic_listxattr,
2380         .removexattr    = generic_removexattr,
2381         .permission     = shmem_permission,
2382 #endif
2383 };
2384
2385 static struct super_operations shmem_ops = {
2386         .alloc_inode    = shmem_alloc_inode,
2387         .destroy_inode  = shmem_destroy_inode,
2388 #ifdef CONFIG_TMPFS
2389         .statfs         = shmem_statfs,
2390         .remount_fs     = shmem_remount_fs,
2391 #endif
2392         .delete_inode   = shmem_delete_inode,
2393         .drop_inode     = generic_delete_inode,
2394         .put_super      = shmem_put_super,
2395 };
2396
2397 static struct vm_operations_struct shmem_vm_ops = {
2398         .nopage         = shmem_nopage,
2399         .populate       = shmem_populate,
2400 #ifdef CONFIG_NUMA
2401         .set_policy     = shmem_set_policy,
2402         .get_policy     = shmem_get_policy,
2403 #endif
2404 };
2405
2406
2407 static int shmem_get_sb(struct file_system_type *fs_type,
2408         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2409 {
2410         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2411 }
2412
2413 static struct file_system_type tmpfs_fs_type = {
2414         .owner          = THIS_MODULE,
2415         .name           = "tmpfs",
2416         .get_sb         = shmem_get_sb,
2417         .kill_sb        = kill_litter_super,
2418 };
2419 static struct vfsmount *shm_mnt;
2420
2421 static int __init init_tmpfs(void)
2422 {
2423         int error;
2424
2425         error = init_inodecache();
2426         if (error)
2427                 goto out3;
2428
2429         error = register_filesystem(&tmpfs_fs_type);
2430         if (error) {
2431                 printk(KERN_ERR "Could not register tmpfs\n");
2432                 goto out2;
2433         }
2434
2435         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2436                                 tmpfs_fs_type.name, NULL);
2437         if (IS_ERR(shm_mnt)) {
2438                 error = PTR_ERR(shm_mnt);
2439                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2440                 goto out1;
2441         }
2442         return 0;
2443
2444 out1:
2445         unregister_filesystem(&tmpfs_fs_type);
2446 out2:
2447         destroy_inodecache();
2448 out3:
2449         shm_mnt = ERR_PTR(error);
2450         return error;
2451 }
2452 module_init(init_tmpfs)
2453
2454 /*
2455  * shmem_file_setup - get an unlinked file living in tmpfs
2456  *
2457  * @name: name for dentry (to be seen in /proc/<pid>/maps
2458  * @size: size to be set for the file
2459  *
2460  */
2461 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2462 {
2463         int error;
2464         struct file *file;
2465         struct inode *inode;
2466         struct dentry *dentry, *root;
2467         struct qstr this;
2468
2469         if (IS_ERR(shm_mnt))
2470                 return (void *)shm_mnt;
2471
2472         if (size < 0 || size > SHMEM_MAX_BYTES)
2473                 return ERR_PTR(-EINVAL);
2474
2475         if (shmem_acct_size(flags, size))
2476                 return ERR_PTR(-ENOMEM);
2477
2478         error = -ENOMEM;
2479         this.name = name;
2480         this.len = strlen(name);
2481         this.hash = 0; /* will go */
2482         root = shm_mnt->mnt_root;
2483         dentry = d_alloc(root, &this);
2484         if (!dentry)
2485                 goto put_memory;
2486
2487         error = -ENFILE;
2488         file = get_empty_filp();
2489         if (!file)
2490                 goto put_dentry;
2491
2492         error = -ENOSPC;
2493         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2494         if (!inode)
2495                 goto close_file;
2496
2497         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2498         d_instantiate(dentry, inode);
2499         inode->i_size = size;
2500         inode->i_nlink = 0;     /* It is unlinked */
2501         file->f_path.mnt = mntget(shm_mnt);
2502         file->f_path.dentry = dentry;
2503         file->f_mapping = inode->i_mapping;
2504         file->f_op = &shmem_file_operations;
2505         file->f_mode = FMODE_WRITE | FMODE_READ;
2506         return file;
2507
2508 close_file:
2509         put_filp(file);
2510 put_dentry:
2511         dput(dentry);
2512 put_memory:
2513         shmem_unacct_size(flags, size);
2514         return ERR_PTR(error);
2515 }
2516
2517 /*
2518  * shmem_zero_setup - setup a shared anonymous mapping
2519  *
2520  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2521  */
2522 int shmem_zero_setup(struct vm_area_struct *vma)
2523 {
2524         struct file *file;
2525         loff_t size = vma->vm_end - vma->vm_start;
2526
2527         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2528         if (IS_ERR(file))
2529                 return PTR_ERR(file);
2530
2531         if (vma->vm_file)
2532                 fput(vma->vm_file);
2533         vma->vm_file = file;
2534         vma->vm_ops = &shmem_vm_ops;
2535         return 0;
2536 }