blob: 0cd7f66f1c668ceb7043f21f0715aff7881962b5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
Hugh Dickins0edd73b2005-06-21 17:15:04 -07009 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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 *
Matt Mackall853ac432009-01-06 14:40:20 -080017 * tiny-shmem:
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
19 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 * This file is released under the GPL.
21 */
22
Matt Mackall853ac432009-01-06 14:40:20 -080023#include <linux/fs.h>
24#include <linux/init.h>
25#include <linux/vfs.h>
26#include <linux/mount.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070027#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080028#include <linux/file.h>
29#include <linux/mm.h>
30#include <linux/module.h>
31#include <linux/swap.h>
32
33static struct vfsmount *shm_mnt;
34
35#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
37 * This virtual memory filesystem is heavily based on the ramfs. It
38 * extends ramfs by the ability to use swap and honor resource limits
39 * which makes it a completely usable filesystem.
40 */
41
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070042#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070043#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010044#include <linux/posix_acl.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070045#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/string.h>
48#include <linux/slab.h>
49#include <linux/backing-dev.h>
50#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/blkdev.h>
53#include <linux/security.h>
54#include <linux/swapops.h>
55#include <linux/mempolicy.h>
56#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000057#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070058#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070059#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080060#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040061#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070062
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#include <asm/uaccess.h>
64#include <asm/div64.h>
65#include <asm/pgtable.h>
66
Hugh Dickinscaefba12009-04-13 14:40:12 -070067/*
68 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
69 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
70 *
71 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
72 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
73 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
74 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
75 *
76 * We use / and * instead of shifts in the definitions below, so that the swap
77 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
78 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
Yuri Tikhonov61609d02009-04-13 14:40:11 -070080#define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
Hugh Dickinscaefba12009-04-13 14:40:12 -070081
82#define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
83#define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
84
85#define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
86#define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
90
91/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
92#define SHMEM_PAGEIN VM_READ
93#define SHMEM_TRUNCATE VM_WRITE
94
95/* Definition to limit shmem_truncate's steps between cond_rescheds */
96#define LATENCY_LIMIT 64
97
98/* Pretend that each entry is of this size in directory's i_size */
99#define BOGO_DIRENT_SIZE 20
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
102enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 SGP_READ, /* don't exceed i_size, don't allocate page */
104 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800105 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 SGP_WRITE, /* may exceed i_size, may allocate page */
107};
108
Andrew Mortonb76db732008-02-08 04:21:49 -0800109#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800110static unsigned long shmem_default_max_blocks(void)
111{
112 return totalram_pages / 2;
113}
114
115static unsigned long shmem_default_max_inodes(void)
116{
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118}
Andrew Mortonb76db732008-02-08 04:21:49 -0800119#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121static int shmem_getpage(struct inode *inode, unsigned long idx,
122 struct page **pagep, enum sgp_type sgp, int *type);
123
Al Viro6daa0e22005-10-21 03:18:50 -0400124static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
126 /*
127 * The above definition of ENTRIES_PER_PAGE, and the use of
128 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
129 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -0700130 *
Mel Gormane12ba742007-10-16 01:25:52 -0700131 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 */
Mel Gormane12ba742007-10-16 01:25:52 -0700133 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700134 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
137static inline void shmem_dir_free(struct page *page)
138{
139 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
140}
141
142static struct page **shmem_dir_map(struct page *page)
143{
144 return (struct page **)kmap_atomic(page, KM_USER0);
145}
146
147static inline void shmem_dir_unmap(struct page **dir)
148{
149 kunmap_atomic(dir, KM_USER0);
150}
151
152static swp_entry_t *shmem_swp_map(struct page *page)
153{
154 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
155}
156
157static inline void shmem_swp_balance_unmap(void)
158{
159 /*
160 * When passing a pointer to an i_direct entry, to code which
161 * also handles indirect entries and so will shmem_swp_unmap,
162 * we must arrange for the preempt count to remain in balance.
163 * What kmap_atomic of a lowmem page does depends on config
164 * and architecture, so pretend to kmap_atomic some lowmem page.
165 */
166 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
167}
168
169static inline void shmem_swp_unmap(swp_entry_t *entry)
170{
171 kunmap_atomic(entry, KM_USER1);
172}
173
174static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
175{
176 return sb->s_fs_info;
177}
178
179/*
180 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
181 * for shared memory and for shared anonymous (/dev/zero) mappings
182 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
183 * consistent with the pre-accounting of private mappings ...
184 */
185static inline int shmem_acct_size(unsigned long flags, loff_t size)
186{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000187 return (flags & VM_NORESERVE) ?
188 0 : security_vm_enough_memory_kern(VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190
191static inline void shmem_unacct_size(unsigned long flags, loff_t size)
192{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000193 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 vm_unacct_memory(VM_ACCT(size));
195}
196
197/*
198 * ... whereas tmpfs objects are accounted incrementally as
199 * pages are allocated, in order to allow huge sparse files.
200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
202 */
203static inline int shmem_acct_block(unsigned long flags)
204{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000205 return (flags & VM_NORESERVE) ?
206 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static inline void shmem_unacct_blocks(unsigned long flags, long pages)
210{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000211 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
213}
214
Hugh Dickins759b9772007-03-05 00:30:28 -0800215static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700216static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800217static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800218static const struct inode_operations shmem_inode_operations;
219static const struct inode_operations shmem_dir_inode_operations;
220static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400221static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700223static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700225 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 .unplug_io_fn = default_unplug_io_fn,
227};
228
229static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800230static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232static void shmem_free_blocks(struct inode *inode, long pages)
233{
234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700235 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 spin_lock(&sbinfo->stat_lock);
237 sbinfo->free_blocks += pages;
238 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
239 spin_unlock(&sbinfo->stat_lock);
240 }
241}
242
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800243static int shmem_reserve_inode(struct super_block *sb)
244{
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 if (!sbinfo->free_inodes) {
249 spin_unlock(&sbinfo->stat_lock);
250 return -ENOSPC;
251 }
252 sbinfo->free_inodes--;
253 spin_unlock(&sbinfo->stat_lock);
254 }
255 return 0;
256}
257
258static void shmem_free_inode(struct super_block *sb)
259{
260 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
261 if (sbinfo->max_inodes) {
262 spin_lock(&sbinfo->stat_lock);
263 sbinfo->free_inodes++;
264 spin_unlock(&sbinfo->stat_lock);
265 }
266}
267
Randy Dunlap46711812008-03-19 17:00:41 -0700268/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 * shmem_recalc_inode - recalculate the size of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * @inode: inode to recalc
271 *
272 * We have to calculate the free blocks since the mm can drop
273 * undirtied hole pages behind our back.
274 *
275 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
276 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
277 *
278 * It has to be called with the spinlock held.
279 */
280static void shmem_recalc_inode(struct inode *inode)
281{
282 struct shmem_inode_info *info = SHMEM_I(inode);
283 long freed;
284
285 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
286 if (freed > 0) {
287 info->alloced -= freed;
288 shmem_unacct_blocks(info->flags, freed);
289 shmem_free_blocks(inode, freed);
290 }
291}
292
Randy Dunlap46711812008-03-19 17:00:41 -0700293/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * shmem_swp_entry - find the swap vector position in the info structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * @info: info structure for the inode
296 * @index: index of the page to find
297 * @page: optional page to add to the structure. Has to be preset to
298 * all zeros
299 *
300 * If there is no space allocated yet it will return NULL when
301 * page is NULL, else it will use the page for the needed block,
302 * setting it to NULL on return to indicate that it has been used.
303 *
304 * The swap vector is organized the following way:
305 *
306 * There are SHMEM_NR_DIRECT entries directly stored in the
307 * shmem_inode_info structure. So small files do not need an addional
308 * allocation.
309 *
310 * For pages with index > SHMEM_NR_DIRECT there is the pointer
311 * i_indirect which points to a page which holds in the first half
312 * doubly indirect blocks, in the second half triple indirect blocks:
313 *
314 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
315 * following layout (for SHMEM_NR_DIRECT == 16):
316 *
317 * i_indirect -> dir --> 16-19
318 * | +-> 20-23
319 * |
320 * +-->dir2 --> 24-27
321 * | +-> 28-31
322 * | +-> 32-35
323 * | +-> 36-39
324 * |
325 * +-->dir3 --> 40-43
326 * +-> 44-47
327 * +-> 48-51
328 * +-> 52-55
329 */
330static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
331{
332 unsigned long offset;
333 struct page **dir;
334 struct page *subdir;
335
336 if (index < SHMEM_NR_DIRECT) {
337 shmem_swp_balance_unmap();
338 return info->i_direct+index;
339 }
340 if (!info->i_indirect) {
341 if (page) {
342 info->i_indirect = *page;
343 *page = NULL;
344 }
345 return NULL; /* need another page */
346 }
347
348 index -= SHMEM_NR_DIRECT;
349 offset = index % ENTRIES_PER_PAGE;
350 index /= ENTRIES_PER_PAGE;
351 dir = shmem_dir_map(info->i_indirect);
352
353 if (index >= ENTRIES_PER_PAGE/2) {
354 index -= ENTRIES_PER_PAGE/2;
355 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
356 index %= ENTRIES_PER_PAGE;
357 subdir = *dir;
358 if (!subdir) {
359 if (page) {
360 *dir = *page;
361 *page = NULL;
362 }
363 shmem_dir_unmap(dir);
364 return NULL; /* need another page */
365 }
366 shmem_dir_unmap(dir);
367 dir = shmem_dir_map(subdir);
368 }
369
370 dir += index;
371 subdir = *dir;
372 if (!subdir) {
373 if (!page || !(subdir = *page)) {
374 shmem_dir_unmap(dir);
375 return NULL; /* need a page */
376 }
377 *dir = subdir;
378 *page = NULL;
379 }
380 shmem_dir_unmap(dir);
381 return shmem_swp_map(subdir) + offset;
382}
383
384static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
385{
386 long incdec = value? 1: -1;
387
388 entry->val = value;
389 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700390 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
391 struct page *page = kmap_atomic_to_page(entry);
392 set_page_private(page, page_private(page) + incdec);
393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
Randy Dunlap46711812008-03-19 17:00:41 -0700396/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 * shmem_swp_alloc - get the position of the swap entry for the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * @info: info structure for the inode
399 * @index: index of the page to find
400 * @sgp: check and recheck i_size? skip allocation?
Randy Dunlap46711812008-03-19 17:00:41 -0700401 *
402 * If the entry does not exist, allocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
404static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
405{
406 struct inode *inode = &info->vfs_inode;
407 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
408 struct page *page = NULL;
409 swp_entry_t *entry;
410
411 if (sgp != SGP_WRITE &&
412 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
413 return ERR_PTR(-EINVAL);
414
415 while (!(entry = shmem_swp_entry(info, index, &page))) {
416 if (sgp == SGP_READ)
417 return shmem_swp_map(ZERO_PAGE(0));
418 /*
419 * Test free_blocks against 1 not 0, since we have 1 data
420 * page (and perhaps indirect index pages) yet to allocate:
421 * a waste to allocate index if we cannot allocate data.
422 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700423 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 spin_lock(&sbinfo->stat_lock);
425 if (sbinfo->free_blocks <= 1) {
426 spin_unlock(&sbinfo->stat_lock);
427 return ERR_PTR(-ENOSPC);
428 }
429 sbinfo->free_blocks--;
430 inode->i_blocks += BLOCKS_PER_PAGE;
431 spin_unlock(&sbinfo->stat_lock);
432 }
433
434 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700435 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700436 if (page)
437 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 spin_lock(&info->lock);
439
440 if (!page) {
441 shmem_free_blocks(inode, 1);
442 return ERR_PTR(-ENOMEM);
443 }
444 if (sgp != SGP_WRITE &&
445 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
446 entry = ERR_PTR(-EINVAL);
447 break;
448 }
449 if (info->next_index <= index)
450 info->next_index = index + 1;
451 }
452 if (page) {
453 /* another task gave its page, or truncated the file */
454 shmem_free_blocks(inode, 1);
455 shmem_dir_free(page);
456 }
457 if (info->next_index <= index && !IS_ERR(entry))
458 info->next_index = index + 1;
459 return entry;
460}
461
Randy Dunlap46711812008-03-19 17:00:41 -0700462/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 * shmem_free_swp - free some swap entries in a directory
Hugh Dickins1ae70002007-03-29 01:20:36 -0700464 * @dir: pointer to the directory
465 * @edir: pointer after last entry of the directory
466 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700468static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
469 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700471 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 swp_entry_t *ptr;
473 int freed = 0;
474
475 for (ptr = dir; ptr < edir; ptr++) {
476 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700477 if (unlikely(punch_lock)) {
478 punch_unlock = punch_lock;
479 punch_lock = NULL;
480 spin_lock(punch_unlock);
481 if (!ptr->val)
482 continue;
483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 free_swap_and_cache(*ptr);
485 *ptr = (swp_entry_t){0};
486 freed++;
487 }
488 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700489 if (punch_unlock)
490 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 return freed;
492}
493
Hugh Dickins1ae70002007-03-29 01:20:36 -0700494static int shmem_map_and_free_swp(struct page *subdir, int offset,
495 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
497 swp_entry_t *ptr;
498 int freed = 0;
499
500 ptr = shmem_swp_map(subdir);
501 for (; offset < limit; offset += LATENCY_LIMIT) {
502 int size = limit - offset;
503 if (size > LATENCY_LIMIT)
504 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700505 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
506 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (need_resched()) {
508 shmem_swp_unmap(ptr);
509 if (*dir) {
510 shmem_dir_unmap(*dir);
511 *dir = NULL;
512 }
513 cond_resched();
514 ptr = shmem_swp_map(subdir);
515 }
516 }
517 shmem_swp_unmap(ptr);
518 return freed;
519}
520
521static void shmem_free_pages(struct list_head *next)
522{
523 struct page *page;
524 int freed = 0;
525
526 do {
527 page = container_of(next, struct page, lru);
528 next = next->next;
529 shmem_dir_free(page);
530 freed++;
531 if (freed >= LATENCY_LIMIT) {
532 cond_resched();
533 freed = 0;
534 }
535 } while (next);
536}
537
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800538static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct shmem_inode_info *info = SHMEM_I(inode);
541 unsigned long idx;
542 unsigned long size;
543 unsigned long limit;
544 unsigned long stage;
545 unsigned long diroff;
546 struct page **dir;
547 struct page *topdir;
548 struct page *middir;
549 struct page *subdir;
550 swp_entry_t *ptr;
551 LIST_HEAD(pages_to_free);
552 long nr_pages_to_free = 0;
553 long nr_swaps_freed = 0;
554 int offset;
555 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700556 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700557 spinlock_t *needs_lock;
558 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700559 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800562 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 if (idx >= info->next_index)
564 return;
565
566 spin_lock(&info->lock);
567 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800568 if (likely(end == (loff_t) -1)) {
569 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700570 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800571 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700572 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700573 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800574 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700575 if (end + 1 >= inode->i_size) { /* we may free a little more */
576 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
577 PAGE_CACHE_SHIFT;
578 upper_limit = SHMEM_MAX_INDEX;
579 } else {
580 limit = (end + 1) >> PAGE_CACHE_SHIFT;
581 upper_limit = limit;
582 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700583 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800584 punch_hole = 1;
585 }
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800588 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 info->i_indirect = NULL;
590 nr_pages_to_free++;
591 list_add(&topdir->lru, &pages_to_free);
592 }
593 spin_unlock(&info->lock);
594
595 if (info->swapped && idx < SHMEM_NR_DIRECT) {
596 ptr = info->i_direct;
597 size = limit;
598 if (size > SHMEM_NR_DIRECT)
599 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700600 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800602
603 /*
604 * If there are no indirect blocks or we are punching a hole
605 * below indirect blocks, nothing to be done.
606 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700607 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 goto done2;
609
Hugh Dickins1ae70002007-03-29 01:20:36 -0700610 /*
611 * The truncation case has already dropped info->lock, and we're safe
612 * because i_size and next_index have already been lowered, preventing
613 * access beyond. But in the punch_hole case, we still need to take
614 * the lock when updating the swap directory, because there might be
615 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
616 * shmem_writepage. However, whenever we find we can remove a whole
617 * directory page (not at the misaligned start or end of the range),
618 * we first NULLify its pointer in the level above, and then have no
619 * need to take the lock when updating its contents: needs_lock and
620 * punch_lock (either pointing to info->lock or NULL) manage this.
621 */
622
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700623 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 limit -= SHMEM_NR_DIRECT;
625 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
626 offset = idx % ENTRIES_PER_PAGE;
627 idx -= offset;
628
629 dir = shmem_dir_map(topdir);
630 stage = ENTRIES_PER_PAGEPAGE/2;
631 if (idx < ENTRIES_PER_PAGEPAGE/2) {
632 middir = topdir;
633 diroff = idx/ENTRIES_PER_PAGE;
634 } else {
635 dir += ENTRIES_PER_PAGE/2;
636 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
637 while (stage <= idx)
638 stage += ENTRIES_PER_PAGEPAGE;
639 middir = *dir;
640 if (*dir) {
641 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
642 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700643 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700644 if (needs_lock) {
645 spin_lock(needs_lock);
646 *dir = NULL;
647 spin_unlock(needs_lock);
648 needs_lock = NULL;
649 } else
650 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 nr_pages_to_free++;
652 list_add(&middir->lru, &pages_to_free);
653 }
654 shmem_dir_unmap(dir);
655 dir = shmem_dir_map(middir);
656 } else {
657 diroff = 0;
658 offset = 0;
659 idx = stage;
660 }
661 }
662
663 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
664 if (unlikely(idx == stage)) {
665 shmem_dir_unmap(dir);
666 dir = shmem_dir_map(topdir) +
667 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
668 while (!*dir) {
669 dir++;
670 idx += ENTRIES_PER_PAGEPAGE;
671 if (idx >= limit)
672 goto done1;
673 }
674 stage = idx + ENTRIES_PER_PAGEPAGE;
675 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700676 if (punch_hole)
677 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700678 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700679 if (needs_lock) {
680 spin_lock(needs_lock);
681 *dir = NULL;
682 spin_unlock(needs_lock);
683 needs_lock = NULL;
684 } else
685 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700686 nr_pages_to_free++;
687 list_add(&middir->lru, &pages_to_free);
688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 shmem_dir_unmap(dir);
690 cond_resched();
691 dir = shmem_dir_map(middir);
692 diroff = 0;
693 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700694 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700696 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
697 if (needs_lock) {
698 spin_lock(needs_lock);
699 dir[diroff] = NULL;
700 spin_unlock(needs_lock);
701 punch_lock = NULL;
702 } else
703 dir[diroff] = NULL;
704 nr_pages_to_free++;
705 list_add(&subdir->lru, &pages_to_free);
706 }
707 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 size = limit - idx;
709 if (size > ENTRIES_PER_PAGE)
710 size = ENTRIES_PER_PAGE;
711 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700712 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (!dir)
714 dir = shmem_dir_map(middir);
715 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700716 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700718 set_page_private(subdir,
719 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700721 } else
722 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700724 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
726done1:
727 shmem_dir_unmap(dir);
728done2:
729 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
730 /*
731 * Call truncate_inode_pages again: racing shmem_unuse_inode
732 * may have swizzled a page in from swap since vmtruncate or
733 * generic_delete_inode did it, before we lowered next_index.
734 * Also, though shmem_getpage checks i_size before adding to
735 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700736 *
737 * Recalling truncate_inode_pages_range and unmap_mapping_range
738 * every time for punch_hole (which never got a chance to clear
739 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
740 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800742 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700743 if (punch_hole)
744 unmap_mapping_range(inode->i_mapping, start,
745 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
747
748 spin_lock(&info->lock);
749 info->flags &= ~SHMEM_TRUNCATE;
750 info->swapped -= nr_swaps_freed;
751 if (nr_pages_to_free)
752 shmem_free_blocks(inode, nr_pages_to_free);
753 shmem_recalc_inode(inode);
754 spin_unlock(&info->lock);
755
756 /*
757 * Empty swap vector directory pages to be freed?
758 */
759 if (!list_empty(&pages_to_free)) {
760 pages_to_free.prev->next = NULL;
761 shmem_free_pages(pages_to_free.next);
762 }
763}
764
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800765static void shmem_truncate(struct inode *inode)
766{
767 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
768}
769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
771{
772 struct inode *inode = dentry->d_inode;
773 struct page *page = NULL;
774 int error;
775
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700776 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 if (attr->ia_size < inode->i_size) {
778 /*
779 * If truncating down to a partial page, then
780 * if that page is already allocated, hold it
781 * in memory until the truncation is over, so
782 * truncate_partial_page cannnot miss it were
783 * it assigned to swap.
784 */
785 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
786 (void) shmem_getpage(inode,
787 attr->ia_size>>PAGE_CACHE_SHIFT,
788 &page, SGP_READ, NULL);
Hugh Dickinsd3602442008-02-04 22:28:44 -0800789 if (page)
790 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792 /*
793 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
794 * detect if any pages might have been added to cache
795 * after truncate_inode_pages. But we needn't bother
796 * if it's being fully truncated to zero-length: the
797 * nrpages check is efficient enough in that case.
798 */
799 if (attr->ia_size) {
800 struct shmem_inode_info *info = SHMEM_I(inode);
801 spin_lock(&info->lock);
802 info->flags &= ~SHMEM_PAGEIN;
803 spin_unlock(&info->lock);
804 }
805 }
806 }
807
808 error = inode_change_ok(inode, attr);
809 if (!error)
810 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700811#ifdef CONFIG_TMPFS_POSIX_ACL
812 if (!error && (attr->ia_valid & ATTR_MODE))
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100813 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700814#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (page)
816 page_cache_release(page);
817 return error;
818}
819
820static void shmem_delete_inode(struct inode *inode)
821{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 struct shmem_inode_info *info = SHMEM_I(inode);
823
824 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700825 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 shmem_unacct_size(info->flags, inode->i_size);
827 inode->i_size = 0;
828 shmem_truncate(inode);
829 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800830 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800832 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 }
834 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700835 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800836 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 clear_inode(inode);
838}
839
840static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
841{
842 swp_entry_t *ptr;
843
844 for (ptr = dir; ptr < edir; ptr++) {
845 if (ptr->val == entry.val)
846 return ptr - dir;
847 }
848 return -1;
849}
850
851static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
852{
853 struct inode *inode;
854 unsigned long idx;
855 unsigned long size;
856 unsigned long limit;
857 unsigned long stage;
858 struct page **dir;
859 struct page *subdir;
860 swp_entry_t *ptr;
861 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800862 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 idx = 0;
865 ptr = info->i_direct;
866 spin_lock(&info->lock);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800867 if (!info->swapped) {
868 list_del_init(&info->swaplist);
869 goto lost2;
870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 limit = info->next_index;
872 size = limit;
873 if (size > SHMEM_NR_DIRECT)
874 size = SHMEM_NR_DIRECT;
875 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800876 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (!info->i_indirect)
879 goto lost2;
880
881 dir = shmem_dir_map(info->i_indirect);
882 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
883
884 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
885 if (unlikely(idx == stage)) {
886 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800887 if (cond_resched_lock(&info->lock)) {
888 /* check it has not been truncated */
889 if (limit > info->next_index) {
890 limit = info->next_index;
891 if (idx >= limit)
892 goto lost2;
893 }
894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 dir = shmem_dir_map(info->i_indirect) +
896 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
897 while (!*dir) {
898 dir++;
899 idx += ENTRIES_PER_PAGEPAGE;
900 if (idx >= limit)
901 goto lost1;
902 }
903 stage = idx + ENTRIES_PER_PAGEPAGE;
904 subdir = *dir;
905 shmem_dir_unmap(dir);
906 dir = shmem_dir_map(subdir);
907 }
908 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700909 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 ptr = shmem_swp_map(subdir);
911 size = limit - idx;
912 if (size > ENTRIES_PER_PAGE)
913 size = ENTRIES_PER_PAGE;
914 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800915 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (offset >= 0) {
917 shmem_dir_unmap(dir);
918 goto found;
919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
921 }
922lost1:
923 shmem_dir_unmap(dir-1);
924lost2:
925 spin_unlock(&info->lock);
926 return 0;
927found:
928 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800929 inode = igrab(&info->vfs_inode);
930 spin_unlock(&info->lock);
931
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800932 /*
933 * Move _head_ to start search for next from here.
934 * But be careful: shmem_delete_inode checks list_empty without taking
935 * mutex, and there's an instant in list_move_tail when info->swaplist
936 * would appear empty, if it were the only one on shmem_swaplist. We
937 * could avoid doing it if inode NULL; or use this minor optimization.
938 */
939 if (shmem_swaplist.next != &info->swaplist)
940 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800941 mutex_unlock(&shmem_swaplist_mutex);
942
943 error = 1;
944 if (!inode)
945 goto out;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800946 /*
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -0800947 * Charge page using GFP_KERNEL while we can wait.
948 * Charged back to the user(not to caller) when swap account is used.
949 * add_to_page_cache() will be called with GFP_NOWAIT.
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800950 */
Hugh Dickins82369552008-02-07 00:14:22 -0800951 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800952 if (error)
953 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800954 error = radix_tree_preload(GFP_KERNEL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700955 if (error) {
956 mem_cgroup_uncharge_cache_page(page);
957 goto out;
958 }
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800959 error = 1;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800960
961 spin_lock(&info->lock);
962 ptr = shmem_swp_entry(info, idx, NULL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700963 if (ptr && ptr->val == entry.val) {
Nick Piggine2867812008-07-25 19:45:30 -0700964 error = add_to_page_cache_locked(page, inode->i_mapping,
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800965 idx, GFP_NOWAIT);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700966 /* does mem_cgroup_uncharge_cache_page on error */
967 } else /* we must compensate for our precharge above */
968 mem_cgroup_uncharge_cache_page(page);
969
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800970 if (error == -EEXIST) {
971 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800972 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800973 if (filepage) {
974 /*
975 * There might be a more uptodate page coming down
976 * from a stacked writepage: forget our swappage if so.
977 */
978 if (PageUptodate(filepage))
979 error = 0;
980 page_cache_release(filepage);
981 }
982 }
983 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800984 delete_from_swap_cache(page);
985 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800987 shmem_swp_set(info, ptr, 0);
988 swap_free(entry);
989 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800991 if (ptr)
992 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 spin_unlock(&info->lock);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800994 radix_tree_preload_end();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800995out:
996 unlock_page(page);
997 page_cache_release(page);
998 iput(inode); /* allows for NULL */
999 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
1002/*
1003 * shmem_unuse() search for an eventually swapped out shmem page.
1004 */
1005int shmem_unuse(swp_entry_t entry, struct page *page)
1006{
1007 struct list_head *p, *next;
1008 struct shmem_inode_info *info;
1009 int found = 0;
1010
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001011 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 list_for_each_safe(p, next, &shmem_swaplist) {
1013 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001014 found = shmem_unuse_inode(info, entry, page);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001015 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001016 if (found)
1017 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001019 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001020 /*
1021 * Can some race bring us here? We've been holding page lock,
1022 * so I think not; but would rather try again later than BUG()
1023 */
1024 unlock_page(page);
1025 page_cache_release(page);
1026out:
1027 return (found < 0) ? found : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028}
1029
1030/*
1031 * Move the page from the page cache to the swap cache.
1032 */
1033static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1034{
1035 struct shmem_inode_info *info;
1036 swp_entry_t *entry, swap;
1037 struct address_space *mapping;
1038 unsigned long index;
1039 struct inode *inode;
1040
1041 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 mapping = page->mapping;
1043 index = page->index;
1044 inode = mapping->host;
1045 info = SHMEM_I(inode);
1046 if (info->flags & VM_LOCKED)
1047 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001048 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 goto redirty;
1050
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001051 /*
1052 * shmem_backing_dev_info's capabilities prevent regular writeback or
1053 * sync from ever calling shmem_writepage; but a stacking filesystem
1054 * may use the ->writepage of its underlying filesystem, in which case
1055 * tmpfs should write out to swap only in response to memory pressure,
Jens Axboe5b0830c2009-09-23 19:37:09 +02001056 * and not for the writeback threads or sync. However, in those cases,
1057 * we do still want to check if there's a redundant swappage to be
1058 * discarded.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001059 */
1060 if (wbc->for_reclaim)
1061 swap = get_swap_page();
1062 else
1063 swap.val = 0;
1064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (index >= info->next_index) {
1067 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1068 goto unlock;
1069 }
1070 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001071 if (entry->val) {
1072 /*
1073 * The more uptodate page coming down from a stacked
1074 * writepage should replace our old swappage.
1075 */
1076 free_swap_and_cache(*entry);
1077 shmem_swp_set(info, entry, 0);
1078 }
1079 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001081 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins73b12622008-02-04 22:28:50 -08001082 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 shmem_swp_set(info, entry, swap.val);
1084 shmem_swp_unmap(entry);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001085 if (list_empty(&info->swaplist))
1086 inode = igrab(inode);
1087 else
1088 inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 spin_unlock(&info->lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001090 swap_shmem_alloc(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001091 BUG_ON(page_mapped(page));
Hugh Dickins73b12622008-02-04 22:28:50 -08001092 page_cache_release(page); /* pagecache ref */
Hugh Dickins9fab5612009-03-31 15:23:33 -07001093 swap_writepage(page, wbc);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001094 if (inode) {
1095 mutex_lock(&shmem_swaplist_mutex);
1096 /* move instead of add in case we're racing */
1097 list_move_tail(&info->swaplist, &shmem_swaplist);
1098 mutex_unlock(&shmem_swaplist_mutex);
1099 iput(inode);
1100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 return 0;
1102 }
1103
1104 shmem_swp_unmap(entry);
1105unlock:
1106 spin_unlock(&info->lock);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -07001107 /*
1108 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1109 * clear SWAP_HAS_CACHE flag.
1110 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07001111 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112redirty:
1113 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001114 if (wbc->for_reclaim)
1115 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1116 unlock_page(page);
1117 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
1120#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001121#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001122static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001123{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001124 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001125
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001126 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001127 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001128
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001129 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001130
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001131 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001132}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001133
1134static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1135{
1136 struct mempolicy *mpol = NULL;
1137 if (sbinfo->mpol) {
1138 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1139 mpol = sbinfo->mpol;
1140 mpol_get(mpol);
1141 spin_unlock(&sbinfo->stat_lock);
1142 }
1143 return mpol;
1144}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001145#endif /* CONFIG_TMPFS */
1146
Hugh Dickins02098fe2008-02-04 22:28:42 -08001147static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1148 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001150 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001152 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001154 spol = mpol_cond_copy(&mpol,
1155 mpol_shared_policy_lookup(&info->policy, idx));
1156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001158 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001160 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001161 pvma.vm_policy = spol;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001162 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 return page;
1164}
1165
Hugh Dickins02098fe2008-02-04 22:28:42 -08001166static struct page *shmem_alloc_page(gfp_t gfp,
1167 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
1169 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001171 /* Create a pseudo vma that just contains the policy */
1172 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001174 pvma.vm_ops = NULL;
1175 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001176
1177 /*
1178 * alloc_page_vma() will drop the shared policy reference
1179 */
1180 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001182#else /* !CONFIG_NUMA */
1183#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001184static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001185{
1186}
1187#endif /* CONFIG_TMPFS */
1188
Hugh Dickins02098fe2008-02-04 22:28:42 -08001189static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1190 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001192 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193}
1194
Hugh Dickins02098fe2008-02-04 22:28:42 -08001195static inline struct page *shmem_alloc_page(gfp_t gfp,
1196 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001198 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001200#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001202#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1203static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1204{
1205 return NULL;
1206}
1207#endif
1208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209/*
1210 * shmem_getpage - either get the page from swap or allocate a new one
1211 *
1212 * If we allocate a new one we do not mark it dirty. That's up to the
1213 * vm. If we swap it in we mark it dirty since we also free the swap
1214 * entry since a page cannot live in both the swap and page cache
1215 */
1216static int shmem_getpage(struct inode *inode, unsigned long idx,
1217 struct page **pagep, enum sgp_type sgp, int *type)
1218{
1219 struct address_space *mapping = inode->i_mapping;
1220 struct shmem_inode_info *info = SHMEM_I(inode);
1221 struct shmem_sb_info *sbinfo;
1222 struct page *filepage = *pagep;
1223 struct page *swappage;
1224 swp_entry_t *entry;
1225 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001226 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 int error;
1228
1229 if (idx >= SHMEM_MAX_INDEX)
1230 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001231
1232 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001233 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 /*
1236 * Normally, filepage is NULL on entry, and either found
1237 * uptodate immediately, or allocated and zeroed, or read
1238 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001239 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001240 * filepage, which may be found not uptodate by other callers
1241 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
1243repeat:
1244 if (!filepage)
1245 filepage = find_lock_page(mapping, idx);
1246 if (filepage && PageUptodate(filepage))
1247 goto done;
1248 error = 0;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001249 gfp = mapping_gfp_mask(mapping);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001250 if (!filepage) {
1251 /*
1252 * Try to preload while we can wait, to not make a habit of
1253 * draining atomic reserves; but don't latch on to this cpu.
1254 */
1255 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1256 if (error)
1257 goto failed;
1258 radix_tree_preload_end();
1259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 spin_lock(&info->lock);
1262 shmem_recalc_inode(inode);
1263 entry = shmem_swp_alloc(info, idx, sgp);
1264 if (IS_ERR(entry)) {
1265 spin_unlock(&info->lock);
1266 error = PTR_ERR(entry);
1267 goto failed;
1268 }
1269 swap = *entry;
1270
1271 if (swap.val) {
1272 /* Look it up and read it in.. */
1273 swappage = lookup_swap_cache(swap);
1274 if (!swappage) {
1275 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001277 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001278 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001279 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001281 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001282 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (!swappage) {
1284 spin_lock(&info->lock);
1285 entry = shmem_swp_alloc(info, idx, sgp);
1286 if (IS_ERR(entry))
1287 error = PTR_ERR(entry);
1288 else {
1289 if (entry->val == swap.val)
1290 error = -ENOMEM;
1291 shmem_swp_unmap(entry);
1292 }
1293 spin_unlock(&info->lock);
1294 if (error)
1295 goto failed;
1296 goto repeat;
1297 }
1298 wait_on_page_locked(swappage);
1299 page_cache_release(swappage);
1300 goto repeat;
1301 }
1302
1303 /* We have to do this with page locked to prevent races */
Nick Piggin529ae9a2008-08-02 12:01:03 +02001304 if (!trylock_page(swappage)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 shmem_swp_unmap(entry);
1306 spin_unlock(&info->lock);
1307 wait_on_page_locked(swappage);
1308 page_cache_release(swappage);
1309 goto repeat;
1310 }
1311 if (PageWriteback(swappage)) {
1312 shmem_swp_unmap(entry);
1313 spin_unlock(&info->lock);
1314 wait_on_page_writeback(swappage);
1315 unlock_page(swappage);
1316 page_cache_release(swappage);
1317 goto repeat;
1318 }
1319 if (!PageUptodate(swappage)) {
1320 shmem_swp_unmap(entry);
1321 spin_unlock(&info->lock);
1322 unlock_page(swappage);
1323 page_cache_release(swappage);
1324 error = -EIO;
1325 goto failed;
1326 }
1327
1328 if (filepage) {
1329 shmem_swp_set(info, entry, 0);
1330 shmem_swp_unmap(entry);
1331 delete_from_swap_cache(swappage);
1332 spin_unlock(&info->lock);
1333 copy_highpage(filepage, swappage);
1334 unlock_page(swappage);
1335 page_cache_release(swappage);
1336 flush_dcache_page(filepage);
1337 SetPageUptodate(filepage);
1338 set_page_dirty(filepage);
1339 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -07001340 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1341 idx, GFP_NOWAIT))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 info->flags |= SHMEM_PAGEIN;
1343 shmem_swp_set(info, entry, 0);
1344 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001345 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 spin_unlock(&info->lock);
1347 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001348 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 swap_free(swap);
1350 } else {
1351 shmem_swp_unmap(entry);
1352 spin_unlock(&info->lock);
Hugh Dickins82369552008-02-07 00:14:22 -08001353 if (error == -ENOMEM) {
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001354 /*
1355 * reclaim from proper memory cgroup and
1356 * call memcg's OOM if needed.
1357 */
1358 error = mem_cgroup_shmem_charge_fallback(
1359 swappage,
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001360 current->mm,
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001361 gfp);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001362 if (error) {
1363 unlock_page(swappage);
1364 page_cache_release(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001365 goto failed;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001366 }
Hugh Dickins82369552008-02-07 00:14:22 -08001367 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001368 unlock_page(swappage);
1369 page_cache_release(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 goto repeat;
1371 }
1372 } else if (sgp == SGP_READ && !filepage) {
1373 shmem_swp_unmap(entry);
1374 filepage = find_get_page(mapping, idx);
1375 if (filepage &&
Nick Piggin529ae9a2008-08-02 12:01:03 +02001376 (!PageUptodate(filepage) || !trylock_page(filepage))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 spin_unlock(&info->lock);
1378 wait_on_page_locked(filepage);
1379 page_cache_release(filepage);
1380 filepage = NULL;
1381 goto repeat;
1382 }
1383 spin_unlock(&info->lock);
1384 } else {
1385 shmem_swp_unmap(entry);
1386 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001387 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 spin_lock(&sbinfo->stat_lock);
1389 if (sbinfo->free_blocks == 0 ||
1390 shmem_acct_block(info->flags)) {
1391 spin_unlock(&sbinfo->stat_lock);
1392 spin_unlock(&info->lock);
1393 error = -ENOSPC;
1394 goto failed;
1395 }
1396 sbinfo->free_blocks--;
1397 inode->i_blocks += BLOCKS_PER_PAGE;
1398 spin_unlock(&sbinfo->stat_lock);
1399 } else if (shmem_acct_block(info->flags)) {
1400 spin_unlock(&info->lock);
1401 error = -ENOSPC;
1402 goto failed;
1403 }
1404
1405 if (!filepage) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001406 int ret;
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001409 filepage = shmem_alloc_page(gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 if (!filepage) {
1411 shmem_unacct_blocks(info->flags, 1);
1412 shmem_free_blocks(inode, 1);
1413 error = -ENOMEM;
1414 goto failed;
1415 }
Rik van Rielb2e18532008-10-18 20:26:30 -07001416 SetPageSwapBacked(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Hugh Dickins82369552008-02-07 00:14:22 -08001418 /* Precharge page while we can wait, compensate after */
1419 error = mem_cgroup_cache_charge(filepage, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001420 GFP_KERNEL);
Hugh Dickins82369552008-02-07 00:14:22 -08001421 if (error) {
1422 page_cache_release(filepage);
1423 shmem_unacct_blocks(info->flags, 1);
1424 shmem_free_blocks(inode, 1);
1425 filepage = NULL;
1426 goto failed;
1427 }
1428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 spin_lock(&info->lock);
1430 entry = shmem_swp_alloc(info, idx, sgp);
1431 if (IS_ERR(entry))
1432 error = PTR_ERR(entry);
1433 else {
1434 swap = *entry;
1435 shmem_swp_unmap(entry);
1436 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001437 ret = error || swap.val;
1438 if (ret)
1439 mem_cgroup_uncharge_cache_page(filepage);
1440 else
1441 ret = add_to_page_cache_lru(filepage, mapping,
1442 idx, GFP_NOWAIT);
1443 /*
1444 * At add_to_page_cache_lru() failure, uncharge will
1445 * be done automatically.
1446 */
1447 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 spin_unlock(&info->lock);
1449 page_cache_release(filepage);
1450 shmem_unacct_blocks(info->flags, 1);
1451 shmem_free_blocks(inode, 1);
1452 filepage = NULL;
1453 if (error)
1454 goto failed;
1455 goto repeat;
1456 }
1457 info->flags |= SHMEM_PAGEIN;
1458 }
1459
1460 info->alloced++;
1461 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001462 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 flush_dcache_page(filepage);
1464 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001465 if (sgp == SGP_DIRTY)
1466 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
1468done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001469 *pagep = filepage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 return 0;
1471
1472failed:
1473 if (*pagep != filepage) {
1474 unlock_page(filepage);
1475 page_cache_release(filepage);
1476 }
1477 return error;
1478}
1479
Nick Piggind0217ac2007-07-19 01:47:03 -07001480static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001482 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001484 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Nick Piggind0217ac2007-07-19 01:47:03 -07001486 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1487 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001488
Hugh Dickins27d54b32008-02-04 22:28:43 -08001489 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001490 if (error)
1491 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Nick Piggin83c54072007-07-19 01:47:05 -07001493 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494}
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001497static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001499 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1501}
1502
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001503static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1504 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001506 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 unsigned long idx;
1508
1509 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1510 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1511}
1512#endif
1513
1514int shmem_lock(struct file *file, int lock, struct user_struct *user)
1515{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001516 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 struct shmem_inode_info *info = SHMEM_I(inode);
1518 int retval = -ENOMEM;
1519
1520 spin_lock(&info->lock);
1521 if (lock && !(info->flags & VM_LOCKED)) {
1522 if (!user_shm_lock(inode->i_size, user))
1523 goto out_nomem;
1524 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001525 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
1527 if (!lock && (info->flags & VM_LOCKED) && user) {
1528 user_shm_unlock(inode->i_size, user);
1529 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001530 mapping_clear_unevictable(file->f_mapping);
1531 scan_mapping_unevictable_pages(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 }
1533 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535out_nomem:
1536 spin_unlock(&info->lock);
1537 return retval;
1538}
1539
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001540static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
1542 file_accessed(file);
1543 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001544 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return 0;
1546}
1547
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001548static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1549 int mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 struct inode *inode;
1552 struct shmem_inode_info *info;
1553 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1554
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001555 if (shmem_reserve_inode(sb))
1556 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
1558 inode = new_inode(sb);
1559 if (inode) {
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001560 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1563 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001564 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 info = SHMEM_I(inode);
1566 memset(info, 0, (char *)inode - (char *)info);
1567 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001568 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 INIT_LIST_HEAD(&info->swaplist);
Al Viro72c04902009-06-24 16:58:48 -04001570 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572 switch (mode & S_IFMT) {
1573 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001574 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 init_special_inode(inode, mode, dev);
1576 break;
1577 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001578 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 inode->i_op = &shmem_inode_operations;
1580 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001581 mpol_shared_policy_init(&info->policy,
1582 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 break;
1584 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001585 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 /* Some things misbehave if size == 0 on a directory */
1587 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1588 inode->i_op = &shmem_dir_inode_operations;
1589 inode->i_fop = &simple_dir_operations;
1590 break;
1591 case S_IFLNK:
1592 /*
1593 * Must not load anything in the rbtree,
1594 * mpol_free_shared_policy will not be called.
1595 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001596 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 break;
1598 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001599 } else
1600 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 return inode;
1602}
1603
1604#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001605static const struct inode_operations shmem_symlink_inode_operations;
1606static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001609 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001610 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1611 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001613static int shmem_readpage(struct file *file, struct page *page)
1614{
1615 struct inode *inode = page->mapping->host;
1616 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1617 unlock_page(page);
1618 return error;
1619}
1620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001622shmem_write_begin(struct file *file, struct address_space *mapping,
1623 loff_t pos, unsigned len, unsigned flags,
1624 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625{
Nick Piggin800d15a2007-10-16 01:25:03 -07001626 struct inode *inode = mapping->host;
1627 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1628 *pagep = NULL;
1629 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1630}
1631
1632static int
1633shmem_write_end(struct file *file, struct address_space *mapping,
1634 loff_t pos, unsigned len, unsigned copied,
1635 struct page *page, void *fsdata)
1636{
1637 struct inode *inode = mapping->host;
1638
Hugh Dickinsd3602442008-02-04 22:28:44 -08001639 if (pos + copied > inode->i_size)
1640 i_size_write(inode, pos + copied);
1641
Nick Piggin800d15a2007-10-16 01:25:03 -07001642 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001643 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001644 page_cache_release(page);
1645
Nick Piggin800d15a2007-10-16 01:25:03 -07001646 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647}
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1650{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001651 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 struct address_space *mapping = inode->i_mapping;
1653 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001654 enum sgp_type sgp = SGP_READ;
1655
1656 /*
1657 * Might this read be for a stacking filesystem? Then when reading
1658 * holes of a sparse file, we actually need to allocate those pages,
1659 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1660 */
1661 if (segment_eq(get_fs(), KERNEL_DS))
1662 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
1664 index = *ppos >> PAGE_CACHE_SHIFT;
1665 offset = *ppos & ~PAGE_CACHE_MASK;
1666
1667 for (;;) {
1668 struct page *page = NULL;
1669 unsigned long end_index, nr, ret;
1670 loff_t i_size = i_size_read(inode);
1671
1672 end_index = i_size >> PAGE_CACHE_SHIFT;
1673 if (index > end_index)
1674 break;
1675 if (index == end_index) {
1676 nr = i_size & ~PAGE_CACHE_MASK;
1677 if (nr <= offset)
1678 break;
1679 }
1680
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001681 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 if (desc->error) {
1683 if (desc->error == -EINVAL)
1684 desc->error = 0;
1685 break;
1686 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001687 if (page)
1688 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
1690 /*
1691 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001692 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 */
1694 nr = PAGE_CACHE_SIZE;
1695 i_size = i_size_read(inode);
1696 end_index = i_size >> PAGE_CACHE_SHIFT;
1697 if (index == end_index) {
1698 nr = i_size & ~PAGE_CACHE_MASK;
1699 if (nr <= offset) {
1700 if (page)
1701 page_cache_release(page);
1702 break;
1703 }
1704 }
1705 nr -= offset;
1706
1707 if (page) {
1708 /*
1709 * If users can be writing to this page using arbitrary
1710 * virtual addresses, take care about potential aliasing
1711 * before reading the page on the kernel side.
1712 */
1713 if (mapping_writably_mapped(mapping))
1714 flush_dcache_page(page);
1715 /*
1716 * Mark the page accessed if we read the beginning.
1717 */
1718 if (!offset)
1719 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001720 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001722 page_cache_get(page);
1723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 /*
1726 * Ok, we have the page, and it's up-to-date, so
1727 * now we can copy it to user space...
1728 *
1729 * The actor routine returns how many bytes were actually used..
1730 * NOTE! This may not be the same as how much of a user buffer
1731 * we filled up (we may be padding etc), so we can only update
1732 * "pos" here (the actor routine has to update the user buffer
1733 * pointers and the remaining count).
1734 */
1735 ret = actor(desc, page, offset, nr);
1736 offset += ret;
1737 index += offset >> PAGE_CACHE_SHIFT;
1738 offset &= ~PAGE_CACHE_MASK;
1739
1740 page_cache_release(page);
1741 if (ret != nr || !desc->count)
1742 break;
1743
1744 cond_resched();
1745 }
1746
1747 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1748 file_accessed(filp);
1749}
1750
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001751static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1752 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001754 struct file *filp = iocb->ki_filp;
1755 ssize_t retval;
1756 unsigned long seg;
1757 size_t count;
1758 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001760 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1761 if (retval)
1762 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001764 for (seg = 0; seg < nr_segs; seg++) {
1765 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001767 desc.written = 0;
1768 desc.arg.buf = iov[seg].iov_base;
1769 desc.count = iov[seg].iov_len;
1770 if (desc.count == 0)
1771 continue;
1772 desc.error = 0;
1773 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1774 retval += desc.written;
1775 if (desc.error) {
1776 retval = retval ?: desc.error;
1777 break;
1778 }
1779 if (desc.count > 0)
1780 break;
1781 }
1782 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783}
1784
David Howells726c3342006-06-23 02:02:58 -07001785static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786{
David Howells726c3342006-06-23 02:02:58 -07001787 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
1789 buf->f_type = TMPFS_MAGIC;
1790 buf->f_bsize = PAGE_CACHE_SIZE;
1791 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001792 spin_lock(&sbinfo->stat_lock);
1793 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 buf->f_blocks = sbinfo->max_blocks;
1795 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001796 }
1797 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 buf->f_files = sbinfo->max_inodes;
1799 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001802 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return 0;
1804}
1805
1806/*
1807 * File creation. Allocate an inode, and we're done..
1808 */
1809static int
1810shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1811{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001812 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 int error = -ENOSPC;
1814
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001815 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001817 error = security_inode_init_security(inode, dir, NULL, NULL,
1818 NULL);
1819 if (error) {
1820 if (error != -EOPNOTSUPP) {
1821 iput(inode);
1822 return error;
1823 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001824 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001825#ifdef CONFIG_TMPFS_POSIX_ACL
1826 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001827 if (error) {
1828 iput(inode);
1829 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001830 }
Al Viro718deb62009-12-16 19:35:36 -05001831#else
1832 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001833#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 dir->i_size += BOGO_DIRENT_SIZE;
1835 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1836 d_instantiate(dentry, inode);
1837 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 }
1839 return error;
1840}
1841
1842static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1843{
1844 int error;
1845
1846 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1847 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001848 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 return 0;
1850}
1851
1852static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1853 struct nameidata *nd)
1854{
1855 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1856}
1857
1858/*
1859 * Link a file..
1860 */
1861static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1862{
1863 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001864 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
1866 /*
1867 * No ordinary (disk based) filesystem counts links as inodes;
1868 * but each new link needs a new dentry, pinning lowmem, and
1869 * tmpfs dentries cannot be pruned until they are unlinked.
1870 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001871 ret = shmem_reserve_inode(inode->i_sb);
1872 if (ret)
1873 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
1875 dir->i_size += BOGO_DIRENT_SIZE;
1876 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001877 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 atomic_inc(&inode->i_count); /* New dentry reference */
1879 dget(dentry); /* Extra pinning count for the created dentry */
1880 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001881out:
1882 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883}
1884
1885static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1886{
1887 struct inode *inode = dentry->d_inode;
1888
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001889 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1890 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
1892 dir->i_size -= BOGO_DIRENT_SIZE;
1893 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001894 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 dput(dentry); /* Undo the count from "create" - this does all the work */
1896 return 0;
1897}
1898
1899static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1900{
1901 if (!simple_empty(dentry))
1902 return -ENOTEMPTY;
1903
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001904 drop_nlink(dentry->d_inode);
1905 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 return shmem_unlink(dir, dentry);
1907}
1908
1909/*
1910 * The VFS layer already does all the dentry stuff for rename,
1911 * we just have to decrement the usage count for the target if
1912 * it exists so that the VFS layer correctly free's it when it
1913 * gets overwritten.
1914 */
1915static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1916{
1917 struct inode *inode = old_dentry->d_inode;
1918 int they_are_dirs = S_ISDIR(inode->i_mode);
1919
1920 if (!simple_empty(new_dentry))
1921 return -ENOTEMPTY;
1922
1923 if (new_dentry->d_inode) {
1924 (void) shmem_unlink(new_dir, new_dentry);
1925 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001926 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001928 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001929 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 }
1931
1932 old_dir->i_size -= BOGO_DIRENT_SIZE;
1933 new_dir->i_size += BOGO_DIRENT_SIZE;
1934 old_dir->i_ctime = old_dir->i_mtime =
1935 new_dir->i_ctime = new_dir->i_mtime =
1936 inode->i_ctime = CURRENT_TIME;
1937 return 0;
1938}
1939
1940static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1941{
1942 int error;
1943 int len;
1944 struct inode *inode;
1945 struct page *page = NULL;
1946 char *kaddr;
1947 struct shmem_inode_info *info;
1948
1949 len = strlen(symname) + 1;
1950 if (len > PAGE_CACHE_SIZE)
1951 return -ENAMETOOLONG;
1952
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001953 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 if (!inode)
1955 return -ENOSPC;
1956
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001957 error = security_inode_init_security(inode, dir, NULL, NULL,
1958 NULL);
1959 if (error) {
1960 if (error != -EOPNOTSUPP) {
1961 iput(inode);
1962 return error;
1963 }
1964 error = 0;
1965 }
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 info = SHMEM_I(inode);
1968 inode->i_size = len-1;
1969 if (len <= (char *)inode - (char *)info) {
1970 /* do it inline */
1971 memcpy(info, symname, len);
1972 inode->i_op = &shmem_symlink_inline_operations;
1973 } else {
1974 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1975 if (error) {
1976 iput(inode);
1977 return error;
1978 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07001979 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 inode->i_op = &shmem_symlink_inode_operations;
1981 kaddr = kmap_atomic(page, KM_USER0);
1982 memcpy(kaddr, symname, len);
1983 kunmap_atomic(kaddr, KM_USER0);
1984 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001985 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 page_cache_release(page);
1987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 dir->i_size += BOGO_DIRENT_SIZE;
1989 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1990 d_instantiate(dentry, inode);
1991 dget(dentry);
1992 return 0;
1993}
1994
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001995static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996{
1997 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001998 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999}
2000
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002001static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002{
2003 struct page *page = NULL;
2004 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2005 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002006 if (page)
2007 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002008 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009}
2010
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002011static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012{
2013 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002014 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 kunmap(page);
2016 mark_page_accessed(page);
2017 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 }
2019}
2020
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002021static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 .readlink = generic_readlink,
2023 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024};
2025
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002026static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 .truncate = shmem_truncate,
2028 .readlink = generic_readlink,
2029 .follow_link = shmem_follow_link,
2030 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031};
2032
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002033#ifdef CONFIG_TMPFS_POSIX_ACL
Randy Dunlap46711812008-03-19 17:00:41 -07002034/*
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002035 * Superblocks without xattr inode operations will get security.* xattr
2036 * support from the VFS "for free". As soon as we have any other xattrs
2037 * like ACLs, we also need to implement the security.* handlers at
2038 * filesystem level, though.
2039 */
2040
Christoph Hellwig431547b2009-11-13 09:52:56 +00002041static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002042 size_t list_len, const char *name,
Christoph Hellwig431547b2009-11-13 09:52:56 +00002043 size_t name_len, int handler_flags)
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002044{
Christoph Hellwig431547b2009-11-13 09:52:56 +00002045 return security_inode_listsecurity(dentry->d_inode, list, list_len);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002046}
2047
Christoph Hellwig431547b2009-11-13 09:52:56 +00002048static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2049 void *buffer, size_t size, int handler_flags)
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002050{
2051 if (strcmp(name, "") == 0)
2052 return -EINVAL;
Christoph Hellwig431547b2009-11-13 09:52:56 +00002053 return xattr_getsecurity(dentry->d_inode, name, buffer, size);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002054}
2055
Christoph Hellwig431547b2009-11-13 09:52:56 +00002056static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2057 const void *value, size_t size, int flags, int handler_flags)
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002058{
2059 if (strcmp(name, "") == 0)
2060 return -EINVAL;
Christoph Hellwig431547b2009-11-13 09:52:56 +00002061 return security_inode_setsecurity(dentry->d_inode, name, value,
2062 size, flags);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002063}
2064
Stephen Hemmingerbb435452010-05-13 17:53:14 -07002065static const struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002066 .prefix = XATTR_SECURITY_PREFIX,
2067 .list = shmem_xattr_security_list,
2068 .get = shmem_xattr_security_get,
2069 .set = shmem_xattr_security_set,
2070};
2071
Stephen Hemmingerbb435452010-05-13 17:53:14 -07002072static const struct xattr_handler *shmem_xattr_handlers[] = {
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002073 &generic_acl_access_handler,
2074 &generic_acl_default_handler,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002075 &shmem_xattr_security_handler,
2076 NULL
2077};
2078#endif
2079
David M. Grimes91828a42006-10-17 00:09:45 -07002080static struct dentry *shmem_get_parent(struct dentry *child)
2081{
2082 return ERR_PTR(-ESTALE);
2083}
2084
2085static int shmem_match(struct inode *ino, void *vfh)
2086{
2087 __u32 *fh = vfh;
2088 __u64 inum = fh[2];
2089 inum = (inum << 32) | fh[1];
2090 return ino->i_ino == inum && fh[0] == ino->i_generation;
2091}
2092
Christoph Hellwig480b1162007-10-21 16:42:13 -07002093static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2094 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002095{
David M. Grimes91828a42006-10-17 00:09:45 -07002096 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002097 struct dentry *dentry = NULL;
2098 u64 inum = fid->raw[2];
2099 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002100
Christoph Hellwig480b1162007-10-21 16:42:13 -07002101 if (fh_len < 3)
2102 return NULL;
2103
2104 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2105 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002106 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002107 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002108 iput(inode);
2109 }
2110
Christoph Hellwig480b1162007-10-21 16:42:13 -07002111 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002112}
2113
2114static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2115 int connectable)
2116{
2117 struct inode *inode = dentry->d_inode;
2118
2119 if (*len < 3)
2120 return 255;
2121
2122 if (hlist_unhashed(&inode->i_hash)) {
2123 /* Unfortunately insert_inode_hash is not idempotent,
2124 * so as we hash inodes here rather than at creation
2125 * time, we need a lock to ensure we only try
2126 * to do it once
2127 */
2128 static DEFINE_SPINLOCK(lock);
2129 spin_lock(&lock);
2130 if (hlist_unhashed(&inode->i_hash))
2131 __insert_inode_hash(inode,
2132 inode->i_ino + inode->i_generation);
2133 spin_unlock(&lock);
2134 }
2135
2136 fh[0] = inode->i_generation;
2137 fh[1] = inode->i_ino;
2138 fh[2] = ((__u64)inode->i_ino) >> 32;
2139
2140 *len = 3;
2141 return 1;
2142}
2143
Christoph Hellwig39655162007-10-21 16:42:17 -07002144static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002145 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002146 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002147 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002148};
2149
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002150static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2151 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
2153 char *this_char, *value, *rest;
2154
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002155 while (options != NULL) {
2156 this_char = options;
2157 for (;;) {
2158 /*
2159 * NUL-terminate this option: unfortunately,
2160 * mount options form a comma-separated list,
2161 * but mpol's nodelist may also contain commas.
2162 */
2163 options = strchr(options, ',');
2164 if (options == NULL)
2165 break;
2166 options++;
2167 if (!isdigit(*options)) {
2168 options[-1] = '\0';
2169 break;
2170 }
2171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 if (!*this_char)
2173 continue;
2174 if ((value = strchr(this_char,'=')) != NULL) {
2175 *value++ = 0;
2176 } else {
2177 printk(KERN_ERR
2178 "tmpfs: No value for mount option '%s'\n",
2179 this_char);
2180 return 1;
2181 }
2182
2183 if (!strcmp(this_char,"size")) {
2184 unsigned long long size;
2185 size = memparse(value,&rest);
2186 if (*rest == '%') {
2187 size <<= PAGE_SHIFT;
2188 size *= totalram_pages;
2189 do_div(size, 100);
2190 rest++;
2191 }
2192 if (*rest)
2193 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002194 sbinfo->max_blocks =
2195 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002197 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 if (*rest)
2199 goto bad_val;
2200 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002201 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 if (*rest)
2203 goto bad_val;
2204 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002205 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002207 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 if (*rest)
2209 goto bad_val;
2210 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002211 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002213 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 if (*rest)
2215 goto bad_val;
2216 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002217 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002219 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 if (*rest)
2221 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002222 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002223 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002224 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 } else {
2226 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2227 this_char);
2228 return 1;
2229 }
2230 }
2231 return 0;
2232
2233bad_val:
2234 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2235 value, this_char);
2236 return 1;
2237
2238}
2239
2240static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2241{
2242 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002243 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002244 unsigned long blocks;
2245 unsigned long inodes;
2246 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002248 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002249 return error;
2250
2251 spin_lock(&sbinfo->stat_lock);
2252 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2253 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002254 if (config.max_blocks < blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002255 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002256 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002257 goto out;
2258 /*
2259 * Those tests also disallow limited->unlimited while any are in
2260 * use, so i_blocks will always be zero when max_blocks is zero;
2261 * but we must separately disallow unlimited->limited, because
2262 * in that case we have no record of how much is already in use.
2263 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002264 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002265 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002266 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002267 goto out;
2268
2269 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002270 sbinfo->max_blocks = config.max_blocks;
2271 sbinfo->free_blocks = config.max_blocks - blocks;
2272 sbinfo->max_inodes = config.max_inodes;
2273 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002274
2275 mpol_put(sbinfo->mpol);
2276 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002277out:
2278 spin_unlock(&sbinfo->stat_lock);
2279 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002281
2282static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2283{
2284 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2285
2286 if (sbinfo->max_blocks != shmem_default_max_blocks())
2287 seq_printf(seq, ",size=%luk",
2288 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2289 if (sbinfo->max_inodes != shmem_default_max_inodes())
2290 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2291 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2292 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2293 if (sbinfo->uid != 0)
2294 seq_printf(seq, ",uid=%u", sbinfo->uid);
2295 if (sbinfo->gid != 0)
2296 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002297 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002298 return 0;
2299}
2300#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
2302static void shmem_put_super(struct super_block *sb)
2303{
2304 kfree(sb->s_fs_info);
2305 sb->s_fs_info = NULL;
2306}
2307
Kay Sievers2b2af542009-04-30 15:23:42 +02002308int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
2310 struct inode *inode;
2311 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002312 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002313 int err = -ENOMEM;
2314
2315 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002316 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002317 L1_CACHE_BYTES), GFP_KERNEL);
2318 if (!sbinfo)
2319 return -ENOMEM;
2320
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002321 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002322 sbinfo->uid = current_fsuid();
2323 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002324 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002326#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 /*
2328 * Per default we only allow half of the physical ram per
2329 * tmpfs instance, limiting inodes to one per page of lowmem;
2330 * but the internal instance is left unlimited.
2331 */
2332 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002333 sbinfo->max_blocks = shmem_default_max_blocks();
2334 sbinfo->max_inodes = shmem_default_max_inodes();
2335 if (shmem_parse_options(data, sbinfo, false)) {
2336 err = -EINVAL;
2337 goto failed;
2338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 }
David M. Grimes91828a42006-10-17 00:09:45 -07002340 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341#else
2342 sb->s_flags |= MS_NOUSER;
2343#endif
2344
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002345 spin_lock_init(&sbinfo->stat_lock);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002346 sbinfo->free_blocks = sbinfo->max_blocks;
2347 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 sb->s_maxbytes = SHMEM_MAX_BYTES;
2350 sb->s_blocksize = PAGE_CACHE_SIZE;
2351 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2352 sb->s_magic = TMPFS_MAGIC;
2353 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002354 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002355#ifdef CONFIG_TMPFS_POSIX_ACL
2356 sb->s_xattr = shmem_xattr_handlers;
2357 sb->s_flags |= MS_POSIXACL;
2358#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002359
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002360 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 if (!inode)
2362 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002363 inode->i_uid = sbinfo->uid;
2364 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 root = d_alloc_root(inode);
2366 if (!root)
2367 goto failed_iput;
2368 sb->s_root = root;
2369 return 0;
2370
2371failed_iput:
2372 iput(inode);
2373failed:
2374 shmem_put_super(sb);
2375 return err;
2376}
2377
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002378static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
2380static struct inode *shmem_alloc_inode(struct super_block *sb)
2381{
2382 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002383 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 if (!p)
2385 return NULL;
2386 return &p->vfs_inode;
2387}
2388
2389static void shmem_destroy_inode(struct inode *inode)
2390{
2391 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2392 /* only struct inode is valid if it's an inline symlink */
2393 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2394 }
2395 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2396}
2397
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002398static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399{
2400 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2401
Christoph Lametera35afb82007-05-16 22:10:57 -07002402 inode_init_once(&p->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403}
2404
2405static int init_inodecache(void)
2406{
2407 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2408 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002409 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 return 0;
2411}
2412
2413static void destroy_inodecache(void)
2414{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002415 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416}
2417
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002418static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002420 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002422 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002423 .write_begin = shmem_write_begin,
2424 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002426 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002427 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428};
2429
Helge Deller15ad7cd2006-12-06 20:40:36 -08002430static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 .mmap = shmem_mmap,
2432#ifdef CONFIG_TMPFS
2433 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002434 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002435 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002436 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002437 .aio_write = generic_file_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 .fsync = simple_sync_file,
Hugh Dickinsae976412007-06-04 10:00:39 +02002439 .splice_read = generic_file_splice_read,
2440 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441#endif
2442};
2443
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002444static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 .truncate = shmem_truncate,
2446 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002447 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002448#ifdef CONFIG_TMPFS_POSIX_ACL
2449 .setxattr = generic_setxattr,
2450 .getxattr = generic_getxattr,
2451 .listxattr = generic_listxattr,
2452 .removexattr = generic_removexattr,
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002453 .check_acl = generic_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002454#endif
2455
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456};
2457
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002458static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459#ifdef CONFIG_TMPFS
2460 .create = shmem_create,
2461 .lookup = simple_lookup,
2462 .link = shmem_link,
2463 .unlink = shmem_unlink,
2464 .symlink = shmem_symlink,
2465 .mkdir = shmem_mkdir,
2466 .rmdir = shmem_rmdir,
2467 .mknod = shmem_mknod,
2468 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002470#ifdef CONFIG_TMPFS_POSIX_ACL
2471 .setattr = shmem_notify_change,
2472 .setxattr = generic_setxattr,
2473 .getxattr = generic_getxattr,
2474 .listxattr = generic_listxattr,
2475 .removexattr = generic_removexattr,
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002476 .check_acl = generic_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002477#endif
2478};
2479
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002480static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002481#ifdef CONFIG_TMPFS_POSIX_ACL
2482 .setattr = shmem_notify_change,
2483 .setxattr = generic_setxattr,
2484 .getxattr = generic_getxattr,
2485 .listxattr = generic_listxattr,
2486 .removexattr = generic_removexattr,
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002487 .check_acl = generic_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002488#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489};
2490
Hugh Dickins759b9772007-03-05 00:30:28 -08002491static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 .alloc_inode = shmem_alloc_inode,
2493 .destroy_inode = shmem_destroy_inode,
2494#ifdef CONFIG_TMPFS
2495 .statfs = shmem_statfs,
2496 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002497 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498#endif
2499 .delete_inode = shmem_delete_inode,
2500 .drop_inode = generic_delete_inode,
2501 .put_super = shmem_put_super,
2502};
2503
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002504static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002505 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506#ifdef CONFIG_NUMA
2507 .set_policy = shmem_set_policy,
2508 .get_policy = shmem_get_policy,
2509#endif
2510};
2511
2512
David Howells454e2392006-06-23 02:02:57 -07002513static int shmem_get_sb(struct file_system_type *fs_type,
2514 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515{
David Howells454e2392006-06-23 02:02:57 -07002516 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517}
2518
2519static struct file_system_type tmpfs_fs_type = {
2520 .owner = THIS_MODULE,
2521 .name = "tmpfs",
2522 .get_sb = shmem_get_sb,
2523 .kill_sb = kill_litter_super,
2524};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
Kay Sievers2b2af542009-04-30 15:23:42 +02002526int __init init_tmpfs(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527{
2528 int error;
2529
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002530 error = bdi_init(&shmem_backing_dev_info);
2531 if (error)
2532 goto out4;
2533
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 error = init_inodecache();
2535 if (error)
2536 goto out3;
2537
2538 error = register_filesystem(&tmpfs_fs_type);
2539 if (error) {
2540 printk(KERN_ERR "Could not register tmpfs\n");
2541 goto out2;
2542 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002543
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002544 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 tmpfs_fs_type.name, NULL);
2546 if (IS_ERR(shm_mnt)) {
2547 error = PTR_ERR(shm_mnt);
2548 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2549 goto out1;
2550 }
2551 return 0;
2552
2553out1:
2554 unregister_filesystem(&tmpfs_fs_type);
2555out2:
2556 destroy_inodecache();
2557out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002558 bdi_destroy(&shmem_backing_dev_info);
2559out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 shm_mnt = ERR_PTR(error);
2561 return error;
2562}
Matt Mackall853ac432009-01-06 14:40:20 -08002563
2564#else /* !CONFIG_SHMEM */
2565
2566/*
2567 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2568 *
2569 * This is intended for small system where the benefits of the full
2570 * shmem code (swap-backed and resource-limited) are outweighed by
2571 * their complexity. On systems without swap this code should be
2572 * effectively equivalent, but much lighter weight.
2573 */
2574
2575#include <linux/ramfs.h>
2576
2577static struct file_system_type tmpfs_fs_type = {
2578 .name = "tmpfs",
2579 .get_sb = ramfs_get_sb,
2580 .kill_sb = kill_litter_super,
2581};
2582
Kay Sievers2b2af542009-04-30 15:23:42 +02002583int __init init_tmpfs(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002584{
2585 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2586
2587 shm_mnt = kern_mount(&tmpfs_fs_type);
2588 BUG_ON(IS_ERR(shm_mnt));
2589
2590 return 0;
2591}
2592
2593int shmem_unuse(swp_entry_t entry, struct page *page)
2594{
2595 return 0;
2596}
2597
Hugh Dickins3f96b792009-09-21 17:03:37 -07002598int shmem_lock(struct file *file, int lock, struct user_struct *user)
2599{
2600 return 0;
2601}
2602
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002603#define shmem_vm_ops generic_file_vm_ops
2604#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002605#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002606#define shmem_acct_size(flags, size) 0
2607#define shmem_unacct_size(flags, size) do {} while (0)
Hugh Dickinscaefba12009-04-13 14:40:12 -07002608#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
Matt Mackall853ac432009-01-06 14:40:20 -08002609
2610#endif /* CONFIG_SHMEM */
2611
2612/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Randy Dunlap46711812008-03-19 17:00:41 -07002614/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 * @name: name for dentry (to be seen in /proc/<pid>/maps
2617 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002618 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002620struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621{
2622 int error;
2623 struct file *file;
2624 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002625 struct path path;
2626 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 struct qstr this;
2628
2629 if (IS_ERR(shm_mnt))
2630 return (void *)shm_mnt;
2631
2632 if (size < 0 || size > SHMEM_MAX_BYTES)
2633 return ERR_PTR(-EINVAL);
2634
2635 if (shmem_acct_size(flags, size))
2636 return ERR_PTR(-ENOMEM);
2637
2638 error = -ENOMEM;
2639 this.name = name;
2640 this.len = strlen(name);
2641 this.hash = 0; /* will go */
2642 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002643 path.dentry = d_alloc(root, &this);
2644 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002646 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002649 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002651 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Al Viro2c48b9c2009-08-09 00:52:35 +04002653 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 inode->i_size = size;
2655 inode->i_nlink = 0; /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002656#ifndef CONFIG_MMU
2657 error = ramfs_nommu_expand_for_mapping(inode, size);
2658 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002659 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002660#endif
Al Viro4b42af82009-08-05 18:25:56 +04002661
2662 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002663 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002664 &shmem_file_operations);
2665 if (!file)
2666 goto put_dentry;
2667
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 return file;
2669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002671 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672put_memory:
2673 shmem_unacct_size(flags, size);
2674 return ERR_PTR(error);
2675}
Keith Packard395e0dd2008-06-20 00:08:06 -07002676EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677
Randy Dunlap46711812008-03-19 17:00:41 -07002678/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2681 */
2682int shmem_zero_setup(struct vm_area_struct *vma)
2683{
2684 struct file *file;
2685 loff_t size = vma->vm_end - vma->vm_start;
2686
2687 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2688 if (IS_ERR(file))
2689 return PTR_ERR(file);
2690
2691 if (vma->vm_file)
2692 fput(vma->vm_file);
2693 vma->vm_file = file;
2694 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 return 0;
2696}