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