blob: 91ce9a1024d7f87abab6c68b9d53e73e857f1fc0 [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>
Tim Chen7e496292010-08-09 17:19:05 -070031#include <linux/percpu_counter.h>
Matt Mackall853ac432009-01-06 14:40:20 -080032#include <linux/swap.h>
33
34static struct vfsmount *shm_mnt;
35
36#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
41 */
42
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070043#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070044#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010045#include <linux/posix_acl.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070046#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/string.h>
49#include <linux/slab.h>
50#include <linux/backing-dev.h>
51#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/blkdev.h>
54#include <linux/security.h>
55#include <linux/swapops.h>
56#include <linux/mempolicy.h>
57#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000058#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070059#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070060#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080061#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040062#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <asm/uaccess.h>
65#include <asm/div64.h>
66#include <asm/pgtable.h>
67
Hugh Dickinscaefba12009-04-13 14:40:12 -070068/*
69 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
70 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
71 *
72 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
73 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
74 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
75 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
76 *
77 * We use / and * instead of shifts in the definitions below, so that the swap
78 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
79 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
Yuri Tikhonov61609d02009-04-13 14:40:11 -070081#define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
Hugh Dickinscaefba12009-04-13 14:40:12 -070082
83#define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
84#define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
85
86#define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
87#define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
91
92/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
93#define SHMEM_PAGEIN VM_READ
94#define SHMEM_TRUNCATE VM_WRITE
95
96/* Definition to limit shmem_truncate's steps between cond_rescheds */
97#define LATENCY_LIMIT 64
98
99/* Pretend that each entry is of this size in directory's i_size */
100#define BOGO_DIRENT_SIZE 20
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
103enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 SGP_READ, /* don't exceed i_size, don't allocate page */
105 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800106 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 SGP_WRITE, /* may exceed i_size, may allocate page */
108};
109
Andrew Mortonb76db732008-02-08 04:21:49 -0800110#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800111static unsigned long shmem_default_max_blocks(void)
112{
113 return totalram_pages / 2;
114}
115
116static unsigned long shmem_default_max_inodes(void)
117{
118 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
119}
Andrew Mortonb76db732008-02-08 04:21:49 -0800120#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122static int shmem_getpage(struct inode *inode, unsigned long idx,
123 struct page **pagep, enum sgp_type sgp, int *type);
124
Al Viro6daa0e22005-10-21 03:18:50 -0400125static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
127 /*
128 * The above definition of ENTRIES_PER_PAGE, and the use of
129 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
130 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -0700131 *
Mel Gormane12ba742007-10-16 01:25:52 -0700132 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 */
Mel Gormane12ba742007-10-16 01:25:52 -0700134 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700135 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
138static inline void shmem_dir_free(struct page *page)
139{
140 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
141}
142
143static struct page **shmem_dir_map(struct page *page)
144{
145 return (struct page **)kmap_atomic(page, KM_USER0);
146}
147
148static inline void shmem_dir_unmap(struct page **dir)
149{
150 kunmap_atomic(dir, KM_USER0);
151}
152
153static swp_entry_t *shmem_swp_map(struct page *page)
154{
155 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
156}
157
158static inline void shmem_swp_balance_unmap(void)
159{
160 /*
161 * When passing a pointer to an i_direct entry, to code which
162 * also handles indirect entries and so will shmem_swp_unmap,
163 * we must arrange for the preempt count to remain in balance.
164 * What kmap_atomic of a lowmem page does depends on config
165 * and architecture, so pretend to kmap_atomic some lowmem page.
166 */
167 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
168}
169
170static inline void shmem_swp_unmap(swp_entry_t *entry)
171{
172 kunmap_atomic(entry, KM_USER1);
173}
174
175static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
176{
177 return sb->s_fs_info;
178}
179
180/*
181 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
182 * for shared memory and for shared anonymous (/dev/zero) mappings
183 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
184 * consistent with the pre-accounting of private mappings ...
185 */
186static inline int shmem_acct_size(unsigned long flags, loff_t size)
187{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000188 return (flags & VM_NORESERVE) ?
189 0 : security_vm_enough_memory_kern(VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
192static inline void shmem_unacct_size(unsigned long flags, loff_t size)
193{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000194 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 vm_unacct_memory(VM_ACCT(size));
196}
197
198/*
199 * ... whereas tmpfs objects are accounted incrementally as
200 * pages are allocated, in order to allow huge sparse files.
201 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
202 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
203 */
204static inline int shmem_acct_block(unsigned long flags)
205{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000206 return (flags & VM_NORESERVE) ?
207 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210static inline void shmem_unacct_blocks(unsigned long flags, long pages)
211{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000212 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
214}
215
Hugh Dickins759b9772007-03-05 00:30:28 -0800216static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700217static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800218static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800219static const struct inode_operations shmem_inode_operations;
220static const struct inode_operations shmem_dir_inode_operations;
221static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400222static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700224static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700226 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 .unplug_io_fn = default_unplug_io_fn,
228};
229
230static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800231static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233static void shmem_free_blocks(struct inode *inode, long pages)
234{
235 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700236 if (sbinfo->max_blocks) {
Tim Chen7e496292010-08-09 17:19:05 -0700237 percpu_counter_add(&sbinfo->used_blocks, -pages);
238 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
Tim Chen7e496292010-08-09 17:19:05 -0700240 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 }
242}
243
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800244static int shmem_reserve_inode(struct super_block *sb)
245{
246 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
247 if (sbinfo->max_inodes) {
248 spin_lock(&sbinfo->stat_lock);
249 if (!sbinfo->free_inodes) {
250 spin_unlock(&sbinfo->stat_lock);
251 return -ENOSPC;
252 }
253 sbinfo->free_inodes--;
254 spin_unlock(&sbinfo->stat_lock);
255 }
256 return 0;
257}
258
259static void shmem_free_inode(struct super_block *sb)
260{
261 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
262 if (sbinfo->max_inodes) {
263 spin_lock(&sbinfo->stat_lock);
264 sbinfo->free_inodes++;
265 spin_unlock(&sbinfo->stat_lock);
266 }
267}
268
Randy Dunlap46711812008-03-19 17:00:41 -0700269/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * shmem_recalc_inode - recalculate the size of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 * @inode: inode to recalc
272 *
273 * We have to calculate the free blocks since the mm can drop
274 * undirtied hole pages behind our back.
275 *
276 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
277 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
278 *
279 * It has to be called with the spinlock held.
280 */
281static void shmem_recalc_inode(struct inode *inode)
282{
283 struct shmem_inode_info *info = SHMEM_I(inode);
284 long freed;
285
286 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
287 if (freed > 0) {
288 info->alloced -= freed;
289 shmem_unacct_blocks(info->flags, freed);
290 shmem_free_blocks(inode, freed);
291 }
292}
293
Randy Dunlap46711812008-03-19 17:00:41 -0700294/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * shmem_swp_entry - find the swap vector position in the info structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 * @info: info structure for the inode
297 * @index: index of the page to find
298 * @page: optional page to add to the structure. Has to be preset to
299 * all zeros
300 *
301 * If there is no space allocated yet it will return NULL when
302 * page is NULL, else it will use the page for the needed block,
303 * setting it to NULL on return to indicate that it has been used.
304 *
305 * The swap vector is organized the following way:
306 *
307 * There are SHMEM_NR_DIRECT entries directly stored in the
308 * shmem_inode_info structure. So small files do not need an addional
309 * allocation.
310 *
311 * For pages with index > SHMEM_NR_DIRECT there is the pointer
312 * i_indirect which points to a page which holds in the first half
313 * doubly indirect blocks, in the second half triple indirect blocks:
314 *
315 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
316 * following layout (for SHMEM_NR_DIRECT == 16):
317 *
318 * i_indirect -> dir --> 16-19
319 * | +-> 20-23
320 * |
321 * +-->dir2 --> 24-27
322 * | +-> 28-31
323 * | +-> 32-35
324 * | +-> 36-39
325 * |
326 * +-->dir3 --> 40-43
327 * +-> 44-47
328 * +-> 48-51
329 * +-> 52-55
330 */
331static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
332{
333 unsigned long offset;
334 struct page **dir;
335 struct page *subdir;
336
337 if (index < SHMEM_NR_DIRECT) {
338 shmem_swp_balance_unmap();
339 return info->i_direct+index;
340 }
341 if (!info->i_indirect) {
342 if (page) {
343 info->i_indirect = *page;
344 *page = NULL;
345 }
346 return NULL; /* need another page */
347 }
348
349 index -= SHMEM_NR_DIRECT;
350 offset = index % ENTRIES_PER_PAGE;
351 index /= ENTRIES_PER_PAGE;
352 dir = shmem_dir_map(info->i_indirect);
353
354 if (index >= ENTRIES_PER_PAGE/2) {
355 index -= ENTRIES_PER_PAGE/2;
356 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
357 index %= ENTRIES_PER_PAGE;
358 subdir = *dir;
359 if (!subdir) {
360 if (page) {
361 *dir = *page;
362 *page = NULL;
363 }
364 shmem_dir_unmap(dir);
365 return NULL; /* need another page */
366 }
367 shmem_dir_unmap(dir);
368 dir = shmem_dir_map(subdir);
369 }
370
371 dir += index;
372 subdir = *dir;
373 if (!subdir) {
374 if (!page || !(subdir = *page)) {
375 shmem_dir_unmap(dir);
376 return NULL; /* need a page */
377 }
378 *dir = subdir;
379 *page = NULL;
380 }
381 shmem_dir_unmap(dir);
382 return shmem_swp_map(subdir) + offset;
383}
384
385static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
386{
387 long incdec = value? 1: -1;
388
389 entry->val = value;
390 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700391 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
392 struct page *page = kmap_atomic_to_page(entry);
393 set_page_private(page, page_private(page) + incdec);
394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Randy Dunlap46711812008-03-19 17:00:41 -0700397/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * shmem_swp_alloc - get the position of the swap entry for the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 * @info: info structure for the inode
400 * @index: index of the page to find
401 * @sgp: check and recheck i_size? skip allocation?
Randy Dunlap46711812008-03-19 17:00:41 -0700402 *
403 * If the entry does not exist, allocate it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
405static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
406{
407 struct inode *inode = &info->vfs_inode;
408 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
409 struct page *page = NULL;
410 swp_entry_t *entry;
411
412 if (sgp != SGP_WRITE &&
413 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
414 return ERR_PTR(-EINVAL);
415
416 while (!(entry = shmem_swp_entry(info, index, &page))) {
417 if (sgp == SGP_READ)
418 return shmem_swp_map(ZERO_PAGE(0));
419 /*
Tim Chen7e496292010-08-09 17:19:05 -0700420 * Test used_blocks against 1 less max_blocks, since we have 1 data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 * page (and perhaps indirect index pages) yet to allocate:
422 * a waste to allocate index if we cannot allocate data.
423 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700424 if (sbinfo->max_blocks) {
Tim Chen7e496292010-08-09 17:19:05 -0700425 if (percpu_counter_compare(&sbinfo->used_blocks, (sbinfo->max_blocks - 1)) > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 return ERR_PTR(-ENOSPC);
Tim Chen7e496292010-08-09 17:19:05 -0700427 percpu_counter_inc(&sbinfo->used_blocks);
428 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 inode->i_blocks += BLOCKS_PER_PAGE;
Tim Chen7e496292010-08-09 17:19:05 -0700430 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 }
432
433 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700434 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 spin_lock(&info->lock);
436
437 if (!page) {
438 shmem_free_blocks(inode, 1);
439 return ERR_PTR(-ENOMEM);
440 }
441 if (sgp != SGP_WRITE &&
442 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
443 entry = ERR_PTR(-EINVAL);
444 break;
445 }
446 if (info->next_index <= index)
447 info->next_index = index + 1;
448 }
449 if (page) {
450 /* another task gave its page, or truncated the file */
451 shmem_free_blocks(inode, 1);
452 shmem_dir_free(page);
453 }
454 if (info->next_index <= index && !IS_ERR(entry))
455 info->next_index = index + 1;
456 return entry;
457}
458
Randy Dunlap46711812008-03-19 17:00:41 -0700459/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 * shmem_free_swp - free some swap entries in a directory
Hugh Dickins1ae70002007-03-29 01:20:36 -0700461 * @dir: pointer to the directory
462 * @edir: pointer after last entry of the directory
463 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700465static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
466 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700468 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 swp_entry_t *ptr;
470 int freed = 0;
471
472 for (ptr = dir; ptr < edir; ptr++) {
473 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700474 if (unlikely(punch_lock)) {
475 punch_unlock = punch_lock;
476 punch_lock = NULL;
477 spin_lock(punch_unlock);
478 if (!ptr->val)
479 continue;
480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 free_swap_and_cache(*ptr);
482 *ptr = (swp_entry_t){0};
483 freed++;
484 }
485 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700486 if (punch_unlock)
487 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 return freed;
489}
490
Hugh Dickins1ae70002007-03-29 01:20:36 -0700491static int shmem_map_and_free_swp(struct page *subdir, int offset,
492 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 swp_entry_t *ptr;
495 int freed = 0;
496
497 ptr = shmem_swp_map(subdir);
498 for (; offset < limit; offset += LATENCY_LIMIT) {
499 int size = limit - offset;
500 if (size > LATENCY_LIMIT)
501 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700502 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
503 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 if (need_resched()) {
505 shmem_swp_unmap(ptr);
506 if (*dir) {
507 shmem_dir_unmap(*dir);
508 *dir = NULL;
509 }
510 cond_resched();
511 ptr = shmem_swp_map(subdir);
512 }
513 }
514 shmem_swp_unmap(ptr);
515 return freed;
516}
517
518static void shmem_free_pages(struct list_head *next)
519{
520 struct page *page;
521 int freed = 0;
522
523 do {
524 page = container_of(next, struct page, lru);
525 next = next->next;
526 shmem_dir_free(page);
527 freed++;
528 if (freed >= LATENCY_LIMIT) {
529 cond_resched();
530 freed = 0;
531 }
532 } while (next);
533}
534
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800535static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
537 struct shmem_inode_info *info = SHMEM_I(inode);
538 unsigned long idx;
539 unsigned long size;
540 unsigned long limit;
541 unsigned long stage;
542 unsigned long diroff;
543 struct page **dir;
544 struct page *topdir;
545 struct page *middir;
546 struct page *subdir;
547 swp_entry_t *ptr;
548 LIST_HEAD(pages_to_free);
549 long nr_pages_to_free = 0;
550 long nr_swaps_freed = 0;
551 int offset;
552 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700553 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700554 spinlock_t *needs_lock;
555 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700556 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800559 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 if (idx >= info->next_index)
561 return;
562
563 spin_lock(&info->lock);
564 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800565 if (likely(end == (loff_t) -1)) {
566 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700567 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800568 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700569 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700570 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800571 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700572 if (end + 1 >= inode->i_size) { /* we may free a little more */
573 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
574 PAGE_CACHE_SHIFT;
575 upper_limit = SHMEM_MAX_INDEX;
576 } else {
577 limit = (end + 1) >> PAGE_CACHE_SHIFT;
578 upper_limit = limit;
579 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700580 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800581 punch_hole = 1;
582 }
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800585 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 info->i_indirect = NULL;
587 nr_pages_to_free++;
588 list_add(&topdir->lru, &pages_to_free);
589 }
590 spin_unlock(&info->lock);
591
592 if (info->swapped && idx < SHMEM_NR_DIRECT) {
593 ptr = info->i_direct;
594 size = limit;
595 if (size > SHMEM_NR_DIRECT)
596 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700597 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800599
600 /*
601 * If there are no indirect blocks or we are punching a hole
602 * below indirect blocks, nothing to be done.
603 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700604 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 goto done2;
606
Hugh Dickins1ae70002007-03-29 01:20:36 -0700607 /*
608 * The truncation case has already dropped info->lock, and we're safe
609 * because i_size and next_index have already been lowered, preventing
610 * access beyond. But in the punch_hole case, we still need to take
611 * the lock when updating the swap directory, because there might be
612 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
613 * shmem_writepage. However, whenever we find we can remove a whole
614 * directory page (not at the misaligned start or end of the range),
615 * we first NULLify its pointer in the level above, and then have no
616 * need to take the lock when updating its contents: needs_lock and
617 * punch_lock (either pointing to info->lock or NULL) manage this.
618 */
619
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700620 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 limit -= SHMEM_NR_DIRECT;
622 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
623 offset = idx % ENTRIES_PER_PAGE;
624 idx -= offset;
625
626 dir = shmem_dir_map(topdir);
627 stage = ENTRIES_PER_PAGEPAGE/2;
628 if (idx < ENTRIES_PER_PAGEPAGE/2) {
629 middir = topdir;
630 diroff = idx/ENTRIES_PER_PAGE;
631 } else {
632 dir += ENTRIES_PER_PAGE/2;
633 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
634 while (stage <= idx)
635 stage += ENTRIES_PER_PAGEPAGE;
636 middir = *dir;
637 if (*dir) {
638 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
639 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700640 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700641 if (needs_lock) {
642 spin_lock(needs_lock);
643 *dir = NULL;
644 spin_unlock(needs_lock);
645 needs_lock = NULL;
646 } else
647 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 nr_pages_to_free++;
649 list_add(&middir->lru, &pages_to_free);
650 }
651 shmem_dir_unmap(dir);
652 dir = shmem_dir_map(middir);
653 } else {
654 diroff = 0;
655 offset = 0;
656 idx = stage;
657 }
658 }
659
660 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
661 if (unlikely(idx == stage)) {
662 shmem_dir_unmap(dir);
663 dir = shmem_dir_map(topdir) +
664 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
665 while (!*dir) {
666 dir++;
667 idx += ENTRIES_PER_PAGEPAGE;
668 if (idx >= limit)
669 goto done1;
670 }
671 stage = idx + ENTRIES_PER_PAGEPAGE;
672 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700673 if (punch_hole)
674 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700675 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700676 if (needs_lock) {
677 spin_lock(needs_lock);
678 *dir = NULL;
679 spin_unlock(needs_lock);
680 needs_lock = NULL;
681 } else
682 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700683 nr_pages_to_free++;
684 list_add(&middir->lru, &pages_to_free);
685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 shmem_dir_unmap(dir);
687 cond_resched();
688 dir = shmem_dir_map(middir);
689 diroff = 0;
690 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700691 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700693 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
694 if (needs_lock) {
695 spin_lock(needs_lock);
696 dir[diroff] = NULL;
697 spin_unlock(needs_lock);
698 punch_lock = NULL;
699 } else
700 dir[diroff] = NULL;
701 nr_pages_to_free++;
702 list_add(&subdir->lru, &pages_to_free);
703 }
704 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 size = limit - idx;
706 if (size > ENTRIES_PER_PAGE)
707 size = ENTRIES_PER_PAGE;
708 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700709 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 if (!dir)
711 dir = shmem_dir_map(middir);
712 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700713 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700715 set_page_private(subdir,
716 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700718 } else
719 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700721 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723done1:
724 shmem_dir_unmap(dir);
725done2:
726 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
727 /*
728 * Call truncate_inode_pages again: racing shmem_unuse_inode
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000729 * may have swizzled a page in from swap since
730 * truncate_pagecache or generic_delete_inode did it, before we
731 * lowered next_index. Also, though shmem_getpage checks
732 * i_size before adding to cache, no recheck after: so fix the
733 * narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700734 *
735 * Recalling truncate_inode_pages_range and unmap_mapping_range
736 * every time for punch_hole (which never got a chance to clear
737 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
738 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800740 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700741 if (punch_hole)
742 unmap_mapping_range(inode->i_mapping, start,
743 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
745
746 spin_lock(&info->lock);
747 info->flags &= ~SHMEM_TRUNCATE;
748 info->swapped -= nr_swaps_freed;
749 if (nr_pages_to_free)
750 shmem_free_blocks(inode, nr_pages_to_free);
751 shmem_recalc_inode(inode);
752 spin_unlock(&info->lock);
753
754 /*
755 * Empty swap vector directory pages to be freed?
756 */
757 if (!list_empty(&pages_to_free)) {
758 pages_to_free.prev->next = NULL;
759 shmem_free_pages(pages_to_free.next);
760 }
761}
762
763static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
764{
765 struct inode *inode = dentry->d_inode;
Nick Pigginaf5a30d2010-06-03 22:01:46 +1000766 loff_t newsize = attr->ia_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 int error;
768
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200769 error = inode_change_ok(inode, attr);
770 if (error)
771 return error;
772
Nick Pigginaf5a30d2010-06-03 22:01:46 +1000773 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)
774 && newsize != inode->i_size) {
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000775 struct page *page = NULL;
776
777 if (newsize < inode->i_size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 /*
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
Justin P. Mattockae0e47f2011-03-01 15:06:02 +0100782 * truncate_partial_page cannot miss it were
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 * it assigned to swap.
784 */
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000785 if (newsize & (PAGE_CACHE_SIZE-1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 (void) shmem_getpage(inode,
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000787 newsize >> PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 &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 */
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000799 if (newsize) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 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 }
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000806
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200807 /* XXX(truncate): truncate_setsize should be called last */
808 truncate_setsize(inode, newsize);
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000809 if (page)
810 page_cache_release(page);
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000811 shmem_truncate_range(inode, newsize, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
813
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200814 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700815#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200816 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100817 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700818#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return error;
820}
821
Al Viro1f895f72010-06-05 19:10:41 -0400822static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 struct shmem_inode_info *info = SHMEM_I(inode);
825
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000826 if (inode->i_mapping->a_ops == &shmem_aops) {
Mark Fashehfef26652005-09-09 13:01:31 -0700827 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 shmem_unacct_size(info->flags, inode->i_size);
829 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000830 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800832 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800834 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 }
836 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700837 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800838 shmem_free_inode(inode->i_sb);
Al Viro1f895f72010-06-05 19:10:41 -0400839 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840}
841
842static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
843{
844 swp_entry_t *ptr;
845
846 for (ptr = dir; ptr < edir; ptr++) {
847 if (ptr->val == entry.val)
848 return ptr - dir;
849 }
850 return -1;
851}
852
853static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
854{
855 struct inode *inode;
856 unsigned long idx;
857 unsigned long size;
858 unsigned long limit;
859 unsigned long stage;
860 struct page **dir;
861 struct page *subdir;
862 swp_entry_t *ptr;
863 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800864 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
866 idx = 0;
867 ptr = info->i_direct;
868 spin_lock(&info->lock);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800869 if (!info->swapped) {
870 list_del_init(&info->swaplist);
871 goto lost2;
872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 limit = info->next_index;
874 size = limit;
875 if (size > SHMEM_NR_DIRECT)
876 size = SHMEM_NR_DIRECT;
877 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800878 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 if (!info->i_indirect)
881 goto lost2;
882
883 dir = shmem_dir_map(info->i_indirect);
884 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
885
886 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
887 if (unlikely(idx == stage)) {
888 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800889 if (cond_resched_lock(&info->lock)) {
890 /* check it has not been truncated */
891 if (limit > info->next_index) {
892 limit = info->next_index;
893 if (idx >= limit)
894 goto lost2;
895 }
896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 dir = shmem_dir_map(info->i_indirect) +
898 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
899 while (!*dir) {
900 dir++;
901 idx += ENTRIES_PER_PAGEPAGE;
902 if (idx >= limit)
903 goto lost1;
904 }
905 stage = idx + ENTRIES_PER_PAGEPAGE;
906 subdir = *dir;
907 shmem_dir_unmap(dir);
908 dir = shmem_dir_map(subdir);
909 }
910 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700911 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 ptr = shmem_swp_map(subdir);
913 size = limit - idx;
914 if (size > ENTRIES_PER_PAGE)
915 size = ENTRIES_PER_PAGE;
916 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800917 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 if (offset >= 0) {
919 shmem_dir_unmap(dir);
920 goto found;
921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
923 }
924lost1:
925 shmem_dir_unmap(dir-1);
926lost2:
927 spin_unlock(&info->lock);
928 return 0;
929found:
930 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800931 inode = igrab(&info->vfs_inode);
932 spin_unlock(&info->lock);
933
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800934 /*
935 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400936 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800937 * mutex, and there's an instant in list_move_tail when info->swaplist
938 * would appear empty, if it were the only one on shmem_swaplist. We
939 * could avoid doing it if inode NULL; or use this minor optimization.
940 */
941 if (shmem_swaplist.next != &info->swaplist)
942 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800943 mutex_unlock(&shmem_swaplist_mutex);
944
945 error = 1;
946 if (!inode)
947 goto out;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800948 /*
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -0800949 * Charge page using GFP_KERNEL while we can wait.
950 * Charged back to the user(not to caller) when swap account is used.
951 * add_to_page_cache() will be called with GFP_NOWAIT.
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800952 */
Hugh Dickins82369552008-02-07 00:14:22 -0800953 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800954 if (error)
955 goto out;
Hugh Dickins82369552008-02-07 00:14:22 -0800956 error = radix_tree_preload(GFP_KERNEL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700957 if (error) {
958 mem_cgroup_uncharge_cache_page(page);
959 goto out;
960 }
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800961 error = 1;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800962
963 spin_lock(&info->lock);
964 ptr = shmem_swp_entry(info, idx, NULL);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700965 if (ptr && ptr->val == entry.val) {
Nick Piggine2867812008-07-25 19:45:30 -0700966 error = add_to_page_cache_locked(page, inode->i_mapping,
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800967 idx, GFP_NOWAIT);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700968 /* does mem_cgroup_uncharge_cache_page on error */
969 } else /* we must compensate for our precharge above */
970 mem_cgroup_uncharge_cache_page(page);
971
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800972 if (error == -EEXIST) {
973 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800974 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800975 if (filepage) {
976 /*
977 * There might be a more uptodate page coming down
978 * from a stacked writepage: forget our swappage if so.
979 */
980 if (PageUptodate(filepage))
981 error = 0;
982 page_cache_release(filepage);
983 }
984 }
985 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800986 delete_from_swap_cache(page);
987 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800989 shmem_swp_set(info, ptr, 0);
990 swap_free(entry);
991 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800993 if (ptr)
994 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 spin_unlock(&info->lock);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800996 radix_tree_preload_end();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800997out:
998 unlock_page(page);
999 page_cache_release(page);
1000 iput(inode); /* allows for NULL */
1001 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002}
1003
1004/*
1005 * shmem_unuse() search for an eventually swapped out shmem page.
1006 */
1007int shmem_unuse(swp_entry_t entry, struct page *page)
1008{
1009 struct list_head *p, *next;
1010 struct shmem_inode_info *info;
1011 int found = 0;
1012
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001013 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 list_for_each_safe(p, next, &shmem_swaplist) {
1015 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001016 found = shmem_unuse_inode(info, entry, page);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001017 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001018 if (found)
1019 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001021 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001022 /*
1023 * Can some race bring us here? We've been holding page lock,
1024 * so I think not; but would rather try again later than BUG()
1025 */
1026 unlock_page(page);
1027 page_cache_release(page);
1028out:
1029 return (found < 0) ? found : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
1032/*
1033 * Move the page from the page cache to the swap cache.
1034 */
1035static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1036{
1037 struct shmem_inode_info *info;
1038 swp_entry_t *entry, swap;
1039 struct address_space *mapping;
1040 unsigned long index;
1041 struct inode *inode;
1042
1043 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 mapping = page->mapping;
1045 index = page->index;
1046 inode = mapping->host;
1047 info = SHMEM_I(inode);
1048 if (info->flags & VM_LOCKED)
1049 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001050 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 goto redirty;
1052
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001053 /*
1054 * shmem_backing_dev_info's capabilities prevent regular writeback or
1055 * sync from ever calling shmem_writepage; but a stacking filesystem
1056 * may use the ->writepage of its underlying filesystem, in which case
1057 * tmpfs should write out to swap only in response to memory pressure,
Jens Axboe5b0830c2009-09-23 19:37:09 +02001058 * and not for the writeback threads or sync. However, in those cases,
1059 * we do still want to check if there's a redundant swappage to be
1060 * discarded.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001061 */
1062 if (wbc->for_reclaim)
1063 swap = get_swap_page();
1064 else
1065 swap.val = 0;
1066
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 if (index >= info->next_index) {
1069 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1070 goto unlock;
1071 }
1072 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001073 if (entry->val) {
1074 /*
1075 * The more uptodate page coming down from a stacked
1076 * writepage should replace our old swappage.
1077 */
1078 free_swap_and_cache(*entry);
1079 shmem_swp_set(info, entry, 0);
1080 }
1081 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001083 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Minchan Kim4c73b1b2011-03-22 16:32:40 -07001084 delete_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 shmem_swp_set(info, entry, swap.val);
1086 shmem_swp_unmap(entry);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001087 if (list_empty(&info->swaplist))
1088 inode = igrab(inode);
1089 else
1090 inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 spin_unlock(&info->lock);
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001092 swap_shmem_alloc(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001093 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001094 swap_writepage(page, wbc);
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001095 if (inode) {
1096 mutex_lock(&shmem_swaplist_mutex);
1097 /* move instead of add in case we're racing */
1098 list_move_tail(&info->swaplist, &shmem_swaplist);
1099 mutex_unlock(&shmem_swaplist_mutex);
1100 iput(inode);
1101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return 0;
1103 }
1104
1105 shmem_swp_unmap(entry);
1106unlock:
1107 spin_unlock(&info->lock);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -07001108 /*
1109 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1110 * clear SWAP_HAS_CACHE flag.
1111 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -07001112 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113redirty:
1114 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001115 if (wbc->for_reclaim)
1116 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1117 unlock_page(page);
1118 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
1121#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001122#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001123static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001124{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001125 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001126
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001127 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001128 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001129
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001130 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001131
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001132 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001133}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001134
1135static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1136{
1137 struct mempolicy *mpol = NULL;
1138 if (sbinfo->mpol) {
1139 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1140 mpol = sbinfo->mpol;
1141 mpol_get(mpol);
1142 spin_unlock(&sbinfo->stat_lock);
1143 }
1144 return mpol;
1145}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001146#endif /* CONFIG_TMPFS */
1147
Hugh Dickins02098fe2008-02-04 22:28:42 -08001148static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1149 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001151 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001153 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001155 spol = mpol_cond_copy(&mpol,
1156 mpol_shared_policy_lookup(&info->policy, idx));
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001159 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001161 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001162 pvma.vm_policy = spol;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001163 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 return page;
1165}
1166
Hugh Dickins02098fe2008-02-04 22:28:42 -08001167static struct page *shmem_alloc_page(gfp_t gfp,
1168 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169{
1170 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001172 /* Create a pseudo vma that just contains the policy */
1173 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001175 pvma.vm_ops = NULL;
1176 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001177
1178 /*
1179 * alloc_page_vma() will drop the shared policy reference
1180 */
1181 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001183#else /* !CONFIG_NUMA */
1184#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001185static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001186{
1187}
1188#endif /* CONFIG_TMPFS */
1189
Hugh Dickins02098fe2008-02-04 22:28:42 -08001190static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1191 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001193 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194}
1195
Hugh Dickins02098fe2008-02-04 22:28:42 -08001196static inline struct page *shmem_alloc_page(gfp_t gfp,
1197 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001199 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001201#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001203#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1204static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1205{
1206 return NULL;
1207}
1208#endif
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210/*
1211 * shmem_getpage - either get the page from swap or allocate a new one
1212 *
1213 * If we allocate a new one we do not mark it dirty. That's up to the
1214 * vm. If we swap it in we mark it dirty since we also free the swap
1215 * entry since a page cannot live in both the swap and page cache
1216 */
1217static int shmem_getpage(struct inode *inode, unsigned long idx,
1218 struct page **pagep, enum sgp_type sgp, int *type)
1219{
1220 struct address_space *mapping = inode->i_mapping;
1221 struct shmem_inode_info *info = SHMEM_I(inode);
1222 struct shmem_sb_info *sbinfo;
1223 struct page *filepage = *pagep;
1224 struct page *swappage;
Shaohua Liff36b802010-08-09 17:19:06 -07001225 struct page *prealloc_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 swp_entry_t *entry;
1227 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001228 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 int error;
1230
1231 if (idx >= SHMEM_MAX_INDEX)
1232 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001233
1234 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001235 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 /*
1238 * Normally, filepage is NULL on entry, and either found
1239 * uptodate immediately, or allocated and zeroed, or read
1240 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001241 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001242 * filepage, which may be found not uptodate by other callers
1243 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 */
1245repeat:
1246 if (!filepage)
1247 filepage = find_lock_page(mapping, idx);
1248 if (filepage && PageUptodate(filepage))
1249 goto done;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001250 gfp = mapping_gfp_mask(mapping);
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001251 if (!filepage) {
1252 /*
1253 * Try to preload while we can wait, to not make a habit of
1254 * draining atomic reserves; but don't latch on to this cpu.
1255 */
1256 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1257 if (error)
1258 goto failed;
1259 radix_tree_preload_end();
Shaohua Liff36b802010-08-09 17:19:06 -07001260 if (sgp != SGP_READ && !prealloc_page) {
1261 /* We don't care if this fails */
1262 prealloc_page = shmem_alloc_page(gfp, info, idx);
1263 if (prealloc_page) {
1264 if (mem_cgroup_cache_charge(prealloc_page,
1265 current->mm, GFP_KERNEL)) {
1266 page_cache_release(prealloc_page);
1267 prealloc_page = NULL;
1268 }
1269 }
1270 }
Hugh Dickinsb409f9f2008-02-04 22:28:54 -08001271 }
Shaohua Liff36b802010-08-09 17:19:06 -07001272 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
1274 spin_lock(&info->lock);
1275 shmem_recalc_inode(inode);
1276 entry = shmem_swp_alloc(info, idx, sgp);
1277 if (IS_ERR(entry)) {
1278 spin_unlock(&info->lock);
1279 error = PTR_ERR(entry);
1280 goto failed;
1281 }
1282 swap = *entry;
1283
1284 if (swap.val) {
1285 /* Look it up and read it in.. */
1286 swappage = lookup_swap_cache(swap);
1287 if (!swappage) {
1288 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001290 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001291 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001292 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001294 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001295 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 if (!swappage) {
1297 spin_lock(&info->lock);
1298 entry = shmem_swp_alloc(info, idx, sgp);
1299 if (IS_ERR(entry))
1300 error = PTR_ERR(entry);
1301 else {
1302 if (entry->val == swap.val)
1303 error = -ENOMEM;
1304 shmem_swp_unmap(entry);
1305 }
1306 spin_unlock(&info->lock);
1307 if (error)
1308 goto failed;
1309 goto repeat;
1310 }
1311 wait_on_page_locked(swappage);
1312 page_cache_release(swappage);
1313 goto repeat;
1314 }
1315
1316 /* We have to do this with page locked to prevent races */
Nick Piggin529ae9a2008-08-02 12:01:03 +02001317 if (!trylock_page(swappage)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 shmem_swp_unmap(entry);
1319 spin_unlock(&info->lock);
1320 wait_on_page_locked(swappage);
1321 page_cache_release(swappage);
1322 goto repeat;
1323 }
1324 if (PageWriteback(swappage)) {
1325 shmem_swp_unmap(entry);
1326 spin_unlock(&info->lock);
1327 wait_on_page_writeback(swappage);
1328 unlock_page(swappage);
1329 page_cache_release(swappage);
1330 goto repeat;
1331 }
1332 if (!PageUptodate(swappage)) {
1333 shmem_swp_unmap(entry);
1334 spin_unlock(&info->lock);
1335 unlock_page(swappage);
1336 page_cache_release(swappage);
1337 error = -EIO;
1338 goto failed;
1339 }
1340
1341 if (filepage) {
1342 shmem_swp_set(info, entry, 0);
1343 shmem_swp_unmap(entry);
1344 delete_from_swap_cache(swappage);
1345 spin_unlock(&info->lock);
1346 copy_highpage(filepage, swappage);
1347 unlock_page(swappage);
1348 page_cache_release(swappage);
1349 flush_dcache_page(filepage);
1350 SetPageUptodate(filepage);
1351 set_page_dirty(filepage);
1352 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -07001353 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1354 idx, GFP_NOWAIT))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 info->flags |= SHMEM_PAGEIN;
1356 shmem_swp_set(info, entry, 0);
1357 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001358 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 spin_unlock(&info->lock);
1360 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001361 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 swap_free(swap);
1363 } else {
1364 shmem_swp_unmap(entry);
1365 spin_unlock(&info->lock);
Hugh Dickins82369552008-02-07 00:14:22 -08001366 if (error == -ENOMEM) {
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001367 /*
1368 * reclaim from proper memory cgroup and
1369 * call memcg's OOM if needed.
1370 */
1371 error = mem_cgroup_shmem_charge_fallback(
1372 swappage,
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001373 current->mm,
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001374 gfp);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001375 if (error) {
1376 unlock_page(swappage);
1377 page_cache_release(swappage);
Hugh Dickins82369552008-02-07 00:14:22 -08001378 goto failed;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001379 }
Hugh Dickins82369552008-02-07 00:14:22 -08001380 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001381 unlock_page(swappage);
1382 page_cache_release(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 goto repeat;
1384 }
1385 } else if (sgp == SGP_READ && !filepage) {
1386 shmem_swp_unmap(entry);
1387 filepage = find_get_page(mapping, idx);
1388 if (filepage &&
Nick Piggin529ae9a2008-08-02 12:01:03 +02001389 (!PageUptodate(filepage) || !trylock_page(filepage))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 spin_unlock(&info->lock);
1391 wait_on_page_locked(filepage);
1392 page_cache_release(filepage);
1393 filepage = NULL;
1394 goto repeat;
1395 }
1396 spin_unlock(&info->lock);
1397 } else {
1398 shmem_swp_unmap(entry);
1399 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001400 if (sbinfo->max_blocks) {
Tim Chen7e496292010-08-09 17:19:05 -07001401 if ((percpu_counter_compare(&sbinfo->used_blocks, sbinfo->max_blocks) > 0) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 shmem_acct_block(info->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 spin_unlock(&info->lock);
1404 error = -ENOSPC;
1405 goto failed;
1406 }
Tim Chen7e496292010-08-09 17:19:05 -07001407 percpu_counter_inc(&sbinfo->used_blocks);
1408 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 inode->i_blocks += BLOCKS_PER_PAGE;
Tim Chen7e496292010-08-09 17:19:05 -07001410 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 } else if (shmem_acct_block(info->flags)) {
1412 spin_unlock(&info->lock);
1413 error = -ENOSPC;
1414 goto failed;
1415 }
1416
1417 if (!filepage) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001418 int ret;
1419
Shaohua Liff36b802010-08-09 17:19:06 -07001420 if (!prealloc_page) {
1421 spin_unlock(&info->lock);
1422 filepage = shmem_alloc_page(gfp, info, idx);
1423 if (!filepage) {
1424 shmem_unacct_blocks(info->flags, 1);
1425 shmem_free_blocks(inode, 1);
1426 error = -ENOMEM;
1427 goto failed;
1428 }
1429 SetPageSwapBacked(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
Shaohua Liff36b802010-08-09 17:19:06 -07001431 /*
1432 * Precharge page while we can wait, compensate
1433 * after
1434 */
1435 error = mem_cgroup_cache_charge(filepage,
1436 current->mm, GFP_KERNEL);
1437 if (error) {
1438 page_cache_release(filepage);
1439 shmem_unacct_blocks(info->flags, 1);
1440 shmem_free_blocks(inode, 1);
1441 filepage = NULL;
1442 goto failed;
1443 }
1444
1445 spin_lock(&info->lock);
1446 } else {
1447 filepage = prealloc_page;
1448 prealloc_page = NULL;
1449 SetPageSwapBacked(filepage);
Hugh Dickins82369552008-02-07 00:14:22 -08001450 }
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 entry = shmem_swp_alloc(info, idx, sgp);
1453 if (IS_ERR(entry))
1454 error = PTR_ERR(entry);
1455 else {
1456 swap = *entry;
1457 shmem_swp_unmap(entry);
1458 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001459 ret = error || swap.val;
1460 if (ret)
1461 mem_cgroup_uncharge_cache_page(filepage);
1462 else
1463 ret = add_to_page_cache_lru(filepage, mapping,
1464 idx, GFP_NOWAIT);
1465 /*
1466 * At add_to_page_cache_lru() failure, uncharge will
1467 * be done automatically.
1468 */
1469 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 spin_unlock(&info->lock);
1471 page_cache_release(filepage);
1472 shmem_unacct_blocks(info->flags, 1);
1473 shmem_free_blocks(inode, 1);
1474 filepage = NULL;
1475 if (error)
1476 goto failed;
1477 goto repeat;
1478 }
1479 info->flags |= SHMEM_PAGEIN;
1480 }
1481
1482 info->alloced++;
1483 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001484 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 flush_dcache_page(filepage);
1486 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001487 if (sgp == SGP_DIRTY)
1488 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 }
1490done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001491 *pagep = filepage;
Shaohua Liff36b802010-08-09 17:19:06 -07001492 error = 0;
1493 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495failed:
1496 if (*pagep != filepage) {
1497 unlock_page(filepage);
1498 page_cache_release(filepage);
1499 }
Shaohua Liff36b802010-08-09 17:19:06 -07001500out:
1501 if (prealloc_page) {
1502 mem_cgroup_uncharge_cache_page(prealloc_page);
1503 page_cache_release(prealloc_page);
1504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 return error;
1506}
1507
Nick Piggind0217ac2007-07-19 01:47:03 -07001508static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001510 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001512 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Nick Piggind0217ac2007-07-19 01:47:03 -07001514 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1515 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001516
Hugh Dickins27d54b32008-02-04 22:28:43 -08001517 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001518 if (error)
1519 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
Nick Piggin83c54072007-07-19 01:47:05 -07001521 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522}
1523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001525static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001527 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1529}
1530
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001531static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1532 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001534 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 unsigned long idx;
1536
1537 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1538 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1539}
1540#endif
1541
1542int shmem_lock(struct file *file, int lock, struct user_struct *user)
1543{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001544 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 struct shmem_inode_info *info = SHMEM_I(inode);
1546 int retval = -ENOMEM;
1547
1548 spin_lock(&info->lock);
1549 if (lock && !(info->flags & VM_LOCKED)) {
1550 if (!user_shm_lock(inode->i_size, user))
1551 goto out_nomem;
1552 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001553 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 }
1555 if (!lock && (info->flags & VM_LOCKED) && user) {
1556 user_shm_unlock(inode->i_size, user);
1557 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001558 mapping_clear_unevictable(file->f_mapping);
1559 scan_mapping_unevictable_pages(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 }
1561 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563out_nomem:
1564 spin_unlock(&info->lock);
1565 return retval;
1566}
1567
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001568static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
1570 file_accessed(file);
1571 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001572 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 return 0;
1574}
1575
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001576static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1577 int mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
1579 struct inode *inode;
1580 struct shmem_inode_info *info;
1581 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1582
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001583 if (shmem_reserve_inode(sb))
1584 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 inode = new_inode(sb);
1587 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001588 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001589 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1592 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001593 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 info = SHMEM_I(inode);
1595 memset(info, 0, (char *)inode - (char *)info);
1596 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001597 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 INIT_LIST_HEAD(&info->swaplist);
Al Viro72c04902009-06-24 16:58:48 -04001599 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 switch (mode & S_IFMT) {
1602 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001603 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 init_special_inode(inode, mode, dev);
1605 break;
1606 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001607 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 inode->i_op = &shmem_inode_operations;
1609 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001610 mpol_shared_policy_init(&info->policy,
1611 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 break;
1613 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001614 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 /* Some things misbehave if size == 0 on a directory */
1616 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1617 inode->i_op = &shmem_dir_inode_operations;
1618 inode->i_fop = &simple_dir_operations;
1619 break;
1620 case S_IFLNK:
1621 /*
1622 * Must not load anything in the rbtree,
1623 * mpol_free_shared_policy will not be called.
1624 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001625 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 break;
1627 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001628 } else
1629 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 return inode;
1631}
1632
1633#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001634static const struct inode_operations shmem_symlink_inode_operations;
1635static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001638 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001639 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1640 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001642static int shmem_readpage(struct file *file, struct page *page)
1643{
1644 struct inode *inode = page->mapping->host;
1645 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1646 unlock_page(page);
1647 return error;
1648}
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001651shmem_write_begin(struct file *file, struct address_space *mapping,
1652 loff_t pos, unsigned len, unsigned flags,
1653 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
Nick Piggin800d15a2007-10-16 01:25:03 -07001655 struct inode *inode = mapping->host;
1656 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1657 *pagep = NULL;
1658 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1659}
1660
1661static int
1662shmem_write_end(struct file *file, struct address_space *mapping,
1663 loff_t pos, unsigned len, unsigned copied,
1664 struct page *page, void *fsdata)
1665{
1666 struct inode *inode = mapping->host;
1667
Hugh Dickinsd3602442008-02-04 22:28:44 -08001668 if (pos + copied > inode->i_size)
1669 i_size_write(inode, pos + copied);
1670
Nick Piggin800d15a2007-10-16 01:25:03 -07001671 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001672 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001673 page_cache_release(page);
1674
Nick Piggin800d15a2007-10-16 01:25:03 -07001675 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676}
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1679{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001680 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 struct address_space *mapping = inode->i_mapping;
1682 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001683 enum sgp_type sgp = SGP_READ;
1684
1685 /*
1686 * Might this read be for a stacking filesystem? Then when reading
1687 * holes of a sparse file, we actually need to allocate those pages,
1688 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1689 */
1690 if (segment_eq(get_fs(), KERNEL_DS))
1691 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693 index = *ppos >> PAGE_CACHE_SHIFT;
1694 offset = *ppos & ~PAGE_CACHE_MASK;
1695
1696 for (;;) {
1697 struct page *page = NULL;
1698 unsigned long end_index, nr, ret;
1699 loff_t i_size = i_size_read(inode);
1700
1701 end_index = i_size >> PAGE_CACHE_SHIFT;
1702 if (index > end_index)
1703 break;
1704 if (index == end_index) {
1705 nr = i_size & ~PAGE_CACHE_MASK;
1706 if (nr <= offset)
1707 break;
1708 }
1709
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001710 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 if (desc->error) {
1712 if (desc->error == -EINVAL)
1713 desc->error = 0;
1714 break;
1715 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001716 if (page)
1717 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 /*
1720 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001721 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 */
1723 nr = PAGE_CACHE_SIZE;
1724 i_size = i_size_read(inode);
1725 end_index = i_size >> PAGE_CACHE_SHIFT;
1726 if (index == end_index) {
1727 nr = i_size & ~PAGE_CACHE_MASK;
1728 if (nr <= offset) {
1729 if (page)
1730 page_cache_release(page);
1731 break;
1732 }
1733 }
1734 nr -= offset;
1735
1736 if (page) {
1737 /*
1738 * If users can be writing to this page using arbitrary
1739 * virtual addresses, take care about potential aliasing
1740 * before reading the page on the kernel side.
1741 */
1742 if (mapping_writably_mapped(mapping))
1743 flush_dcache_page(page);
1744 /*
1745 * Mark the page accessed if we read the beginning.
1746 */
1747 if (!offset)
1748 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001749 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001751 page_cache_get(page);
1752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
1754 /*
1755 * Ok, we have the page, and it's up-to-date, so
1756 * now we can copy it to user space...
1757 *
1758 * The actor routine returns how many bytes were actually used..
1759 * NOTE! This may not be the same as how much of a user buffer
1760 * we filled up (we may be padding etc), so we can only update
1761 * "pos" here (the actor routine has to update the user buffer
1762 * pointers and the remaining count).
1763 */
1764 ret = actor(desc, page, offset, nr);
1765 offset += ret;
1766 index += offset >> PAGE_CACHE_SHIFT;
1767 offset &= ~PAGE_CACHE_MASK;
1768
1769 page_cache_release(page);
1770 if (ret != nr || !desc->count)
1771 break;
1772
1773 cond_resched();
1774 }
1775
1776 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1777 file_accessed(filp);
1778}
1779
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001780static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1781 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001783 struct file *filp = iocb->ki_filp;
1784 ssize_t retval;
1785 unsigned long seg;
1786 size_t count;
1787 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001789 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1790 if (retval)
1791 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001793 for (seg = 0; seg < nr_segs; seg++) {
1794 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001796 desc.written = 0;
1797 desc.arg.buf = iov[seg].iov_base;
1798 desc.count = iov[seg].iov_len;
1799 if (desc.count == 0)
1800 continue;
1801 desc.error = 0;
1802 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1803 retval += desc.written;
1804 if (desc.error) {
1805 retval = retval ?: desc.error;
1806 break;
1807 }
1808 if (desc.count > 0)
1809 break;
1810 }
1811 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812}
1813
David Howells726c3342006-06-23 02:02:58 -07001814static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815{
David Howells726c3342006-06-23 02:02:58 -07001816 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
1818 buf->f_type = TMPFS_MAGIC;
1819 buf->f_bsize = PAGE_CACHE_SIZE;
1820 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001821 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 buf->f_blocks = sbinfo->max_blocks;
Tim Chen7e496292010-08-09 17:19:05 -07001823 buf->f_bavail = buf->f_bfree =
1824 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001825 }
1826 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 buf->f_files = sbinfo->max_inodes;
1828 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 }
1830 /* else leave those fields 0 like simple_statfs */
1831 return 0;
1832}
1833
1834/*
1835 * File creation. Allocate an inode, and we're done..
1836 */
1837static int
1838shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1839{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001840 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 int error = -ENOSPC;
1842
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001843 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001845 error = security_inode_init_security(inode, dir,
1846 &dentry->d_name, NULL,
1847 NULL, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001848 if (error) {
1849 if (error != -EOPNOTSUPP) {
1850 iput(inode);
1851 return error;
1852 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001853 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001854#ifdef CONFIG_TMPFS_POSIX_ACL
1855 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001856 if (error) {
1857 iput(inode);
1858 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001859 }
Al Viro718deb62009-12-16 19:35:36 -05001860#else
1861 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001862#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 dir->i_size += BOGO_DIRENT_SIZE;
1864 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1865 d_instantiate(dentry, inode);
1866 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
1868 return error;
1869}
1870
1871static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1872{
1873 int error;
1874
1875 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1876 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001877 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 return 0;
1879}
1880
1881static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1882 struct nameidata *nd)
1883{
1884 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1885}
1886
1887/*
1888 * Link a file..
1889 */
1890static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1891{
1892 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001893 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
1895 /*
1896 * No ordinary (disk based) filesystem counts links as inodes;
1897 * but each new link needs a new dentry, pinning lowmem, and
1898 * tmpfs dentries cannot be pruned until they are unlinked.
1899 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001900 ret = shmem_reserve_inode(inode->i_sb);
1901 if (ret)
1902 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 dir->i_size += BOGO_DIRENT_SIZE;
1905 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001906 inc_nlink(inode);
Al Viro7de9c6e2010-10-23 11:11:40 -04001907 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 dget(dentry); /* Extra pinning count for the created dentry */
1909 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001910out:
1911 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912}
1913
1914static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1915{
1916 struct inode *inode = dentry->d_inode;
1917
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001918 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1919 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921 dir->i_size -= BOGO_DIRENT_SIZE;
1922 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001923 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 dput(dentry); /* Undo the count from "create" - this does all the work */
1925 return 0;
1926}
1927
1928static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1929{
1930 if (!simple_empty(dentry))
1931 return -ENOTEMPTY;
1932
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001933 drop_nlink(dentry->d_inode);
1934 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return shmem_unlink(dir, dentry);
1936}
1937
1938/*
1939 * The VFS layer already does all the dentry stuff for rename,
1940 * we just have to decrement the usage count for the target if
1941 * it exists so that the VFS layer correctly free's it when it
1942 * gets overwritten.
1943 */
1944static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1945{
1946 struct inode *inode = old_dentry->d_inode;
1947 int they_are_dirs = S_ISDIR(inode->i_mode);
1948
1949 if (!simple_empty(new_dentry))
1950 return -ENOTEMPTY;
1951
1952 if (new_dentry->d_inode) {
1953 (void) shmem_unlink(new_dir, new_dentry);
1954 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001955 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001957 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001958 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 }
1960
1961 old_dir->i_size -= BOGO_DIRENT_SIZE;
1962 new_dir->i_size += BOGO_DIRENT_SIZE;
1963 old_dir->i_ctime = old_dir->i_mtime =
1964 new_dir->i_ctime = new_dir->i_mtime =
1965 inode->i_ctime = CURRENT_TIME;
1966 return 0;
1967}
1968
1969static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1970{
1971 int error;
1972 int len;
1973 struct inode *inode;
1974 struct page *page = NULL;
1975 char *kaddr;
1976 struct shmem_inode_info *info;
1977
1978 len = strlen(symname) + 1;
1979 if (len > PAGE_CACHE_SIZE)
1980 return -ENAMETOOLONG;
1981
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001982 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (!inode)
1984 return -ENOSPC;
1985
Eric Paris2a7dba32011-02-01 11:05:39 -05001986 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
1987 NULL, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001988 if (error) {
1989 if (error != -EOPNOTSUPP) {
1990 iput(inode);
1991 return error;
1992 }
1993 error = 0;
1994 }
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 info = SHMEM_I(inode);
1997 inode->i_size = len-1;
1998 if (len <= (char *)inode - (char *)info) {
1999 /* do it inline */
2000 memcpy(info, symname, len);
2001 inode->i_op = &shmem_symlink_inline_operations;
2002 } else {
2003 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2004 if (error) {
2005 iput(inode);
2006 return error;
2007 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002008 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 inode->i_op = &shmem_symlink_inode_operations;
2010 kaddr = kmap_atomic(page, KM_USER0);
2011 memcpy(kaddr, symname, len);
2012 kunmap_atomic(kaddr, KM_USER0);
2013 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002014 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 page_cache_release(page);
2016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 dir->i_size += BOGO_DIRENT_SIZE;
2018 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2019 d_instantiate(dentry, inode);
2020 dget(dentry);
2021 return 0;
2022}
2023
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002024static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
2026 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002027 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028}
2029
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002030static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031{
2032 struct page *page = NULL;
2033 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2034 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002035 if (page)
2036 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002037 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038}
2039
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002040static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
2042 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002043 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 kunmap(page);
2045 mark_page_accessed(page);
2046 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 }
2048}
2049
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002050static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 .readlink = generic_readlink,
2052 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053};
2054
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002055static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 .readlink = generic_readlink,
2057 .follow_link = shmem_follow_link,
2058 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059};
2060
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002061#ifdef CONFIG_TMPFS_POSIX_ACL
Randy Dunlap46711812008-03-19 17:00:41 -07002062/*
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002063 * Superblocks without xattr inode operations will get security.* xattr
2064 * support from the VFS "for free". As soon as we have any other xattrs
2065 * like ACLs, we also need to implement the security.* handlers at
2066 * filesystem level, though.
2067 */
2068
Christoph Hellwig431547b2009-11-13 09:52:56 +00002069static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002070 size_t list_len, const char *name,
Christoph Hellwig431547b2009-11-13 09:52:56 +00002071 size_t name_len, int handler_flags)
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002072{
Christoph Hellwig431547b2009-11-13 09:52:56 +00002073 return security_inode_listsecurity(dentry->d_inode, list, list_len);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002074}
2075
Christoph Hellwig431547b2009-11-13 09:52:56 +00002076static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2077 void *buffer, size_t size, int handler_flags)
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002078{
2079 if (strcmp(name, "") == 0)
2080 return -EINVAL;
Christoph Hellwig431547b2009-11-13 09:52:56 +00002081 return xattr_getsecurity(dentry->d_inode, name, buffer, size);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002082}
2083
Christoph Hellwig431547b2009-11-13 09:52:56 +00002084static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2085 const void *value, size_t size, int flags, int handler_flags)
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002086{
2087 if (strcmp(name, "") == 0)
2088 return -EINVAL;
Christoph Hellwig431547b2009-11-13 09:52:56 +00002089 return security_inode_setsecurity(dentry->d_inode, name, value,
2090 size, flags);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002091}
2092
Stephen Hemmingerbb435452010-05-13 17:53:14 -07002093static const struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002094 .prefix = XATTR_SECURITY_PREFIX,
2095 .list = shmem_xattr_security_list,
2096 .get = shmem_xattr_security_get,
2097 .set = shmem_xattr_security_set,
2098};
2099
Stephen Hemmingerbb435452010-05-13 17:53:14 -07002100static const struct xattr_handler *shmem_xattr_handlers[] = {
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002101 &generic_acl_access_handler,
2102 &generic_acl_default_handler,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002103 &shmem_xattr_security_handler,
2104 NULL
2105};
2106#endif
2107
David M. Grimes91828a42006-10-17 00:09:45 -07002108static struct dentry *shmem_get_parent(struct dentry *child)
2109{
2110 return ERR_PTR(-ESTALE);
2111}
2112
2113static int shmem_match(struct inode *ino, void *vfh)
2114{
2115 __u32 *fh = vfh;
2116 __u64 inum = fh[2];
2117 inum = (inum << 32) | fh[1];
2118 return ino->i_ino == inum && fh[0] == ino->i_generation;
2119}
2120
Christoph Hellwig480b1162007-10-21 16:42:13 -07002121static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2122 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002123{
David M. Grimes91828a42006-10-17 00:09:45 -07002124 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002125 struct dentry *dentry = NULL;
2126 u64 inum = fid->raw[2];
2127 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002128
Christoph Hellwig480b1162007-10-21 16:42:13 -07002129 if (fh_len < 3)
2130 return NULL;
2131
2132 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2133 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002134 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002135 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002136 iput(inode);
2137 }
2138
Christoph Hellwig480b1162007-10-21 16:42:13 -07002139 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002140}
2141
2142static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2143 int connectable)
2144{
2145 struct inode *inode = dentry->d_inode;
2146
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302147 if (*len < 3) {
2148 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07002149 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302150 }
David M. Grimes91828a42006-10-17 00:09:45 -07002151
Al Viro1d3382c2010-10-23 15:19:20 -04002152 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002153 /* Unfortunately insert_inode_hash is not idempotent,
2154 * so as we hash inodes here rather than at creation
2155 * time, we need a lock to ensure we only try
2156 * to do it once
2157 */
2158 static DEFINE_SPINLOCK(lock);
2159 spin_lock(&lock);
Al Viro1d3382c2010-10-23 15:19:20 -04002160 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002161 __insert_inode_hash(inode,
2162 inode->i_ino + inode->i_generation);
2163 spin_unlock(&lock);
2164 }
2165
2166 fh[0] = inode->i_generation;
2167 fh[1] = inode->i_ino;
2168 fh[2] = ((__u64)inode->i_ino) >> 32;
2169
2170 *len = 3;
2171 return 1;
2172}
2173
Christoph Hellwig39655162007-10-21 16:42:17 -07002174static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002175 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002176 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002177 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002178};
2179
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002180static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2181 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182{
2183 char *this_char, *value, *rest;
2184
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002185 while (options != NULL) {
2186 this_char = options;
2187 for (;;) {
2188 /*
2189 * NUL-terminate this option: unfortunately,
2190 * mount options form a comma-separated list,
2191 * but mpol's nodelist may also contain commas.
2192 */
2193 options = strchr(options, ',');
2194 if (options == NULL)
2195 break;
2196 options++;
2197 if (!isdigit(*options)) {
2198 options[-1] = '\0';
2199 break;
2200 }
2201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 if (!*this_char)
2203 continue;
2204 if ((value = strchr(this_char,'=')) != NULL) {
2205 *value++ = 0;
2206 } else {
2207 printk(KERN_ERR
2208 "tmpfs: No value for mount option '%s'\n",
2209 this_char);
2210 return 1;
2211 }
2212
2213 if (!strcmp(this_char,"size")) {
2214 unsigned long long size;
2215 size = memparse(value,&rest);
2216 if (*rest == '%') {
2217 size <<= PAGE_SHIFT;
2218 size *= totalram_pages;
2219 do_div(size, 100);
2220 rest++;
2221 }
2222 if (*rest)
2223 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002224 sbinfo->max_blocks =
2225 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002227 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 if (*rest)
2229 goto bad_val;
2230 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002231 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 if (*rest)
2233 goto bad_val;
2234 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002235 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002237 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 if (*rest)
2239 goto bad_val;
2240 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002241 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002243 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 if (*rest)
2245 goto bad_val;
2246 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002247 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002249 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 if (*rest)
2251 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002252 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002253 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002254 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 } else {
2256 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2257 this_char);
2258 return 1;
2259 }
2260 }
2261 return 0;
2262
2263bad_val:
2264 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2265 value, this_char);
2266 return 1;
2267
2268}
2269
2270static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2271{
2272 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002273 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002274 unsigned long inodes;
2275 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002277 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002278 return error;
2279
2280 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002281 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002282 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002283 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002284 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002285 goto out;
2286 /*
2287 * Those tests also disallow limited->unlimited while any are in
2288 * use, so i_blocks will always be zero when max_blocks is zero;
2289 * but we must separately disallow unlimited->limited, because
2290 * in that case we have no record of how much is already in use.
2291 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002292 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002293 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002294 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002295 goto out;
2296
2297 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002298 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002299 sbinfo->max_inodes = config.max_inodes;
2300 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002301
2302 mpol_put(sbinfo->mpol);
2303 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002304out:
2305 spin_unlock(&sbinfo->stat_lock);
2306 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002308
2309static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2310{
2311 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2312
2313 if (sbinfo->max_blocks != shmem_default_max_blocks())
2314 seq_printf(seq, ",size=%luk",
2315 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2316 if (sbinfo->max_inodes != shmem_default_max_inodes())
2317 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2318 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2319 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2320 if (sbinfo->uid != 0)
2321 seq_printf(seq, ",uid=%u", sbinfo->uid);
2322 if (sbinfo->gid != 0)
2323 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002324 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002325 return 0;
2326}
2327#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
2329static void shmem_put_super(struct super_block *sb)
2330{
Hugh Dickins602586a2010-08-17 15:23:56 -07002331 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2332
2333 percpu_counter_destroy(&sbinfo->used_blocks);
2334 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 sb->s_fs_info = NULL;
2336}
2337
Kay Sievers2b2af542009-04-30 15:23:42 +02002338int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339{
2340 struct inode *inode;
2341 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002342 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002343 int err = -ENOMEM;
2344
2345 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002346 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002347 L1_CACHE_BYTES), GFP_KERNEL);
2348 if (!sbinfo)
2349 return -ENOMEM;
2350
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002351 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002352 sbinfo->uid = current_fsuid();
2353 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002354 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002356#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 /*
2358 * Per default we only allow half of the physical ram per
2359 * tmpfs instance, limiting inodes to one per page of lowmem;
2360 * but the internal instance is left unlimited.
2361 */
2362 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002363 sbinfo->max_blocks = shmem_default_max_blocks();
2364 sbinfo->max_inodes = shmem_default_max_inodes();
2365 if (shmem_parse_options(data, sbinfo, false)) {
2366 err = -EINVAL;
2367 goto failed;
2368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 }
David M. Grimes91828a42006-10-17 00:09:45 -07002370 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371#else
2372 sb->s_flags |= MS_NOUSER;
2373#endif
2374
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002375 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002376 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2377 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002378 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 sb->s_maxbytes = SHMEM_MAX_BYTES;
2381 sb->s_blocksize = PAGE_CACHE_SIZE;
2382 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2383 sb->s_magic = TMPFS_MAGIC;
2384 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002385 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002386#ifdef CONFIG_TMPFS_POSIX_ACL
2387 sb->s_xattr = shmem_xattr_handlers;
2388 sb->s_flags |= MS_POSIXACL;
2389#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002390
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002391 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 if (!inode)
2393 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002394 inode->i_uid = sbinfo->uid;
2395 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 root = d_alloc_root(inode);
2397 if (!root)
2398 goto failed_iput;
2399 sb->s_root = root;
2400 return 0;
2401
2402failed_iput:
2403 iput(inode);
2404failed:
2405 shmem_put_super(sb);
2406 return err;
2407}
2408
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002409static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
2411static struct inode *shmem_alloc_inode(struct super_block *sb)
2412{
2413 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002414 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (!p)
2416 return NULL;
2417 return &p->vfs_inode;
2418}
2419
Nick Pigginfa0d7e32011-01-07 17:49:49 +11002420static void shmem_i_callback(struct rcu_head *head)
2421{
2422 struct inode *inode = container_of(head, struct inode, i_rcu);
2423 INIT_LIST_HEAD(&inode->i_dentry);
2424 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2425}
2426
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427static void shmem_destroy_inode(struct inode *inode)
2428{
2429 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2430 /* only struct inode is valid if it's an inline symlink */
2431 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2432 }
Nick Pigginfa0d7e32011-01-07 17:49:49 +11002433 call_rcu(&inode->i_rcu, shmem_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434}
2435
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002436static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437{
2438 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2439
Christoph Lametera35afb82007-05-16 22:10:57 -07002440 inode_init_once(&p->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
2442
2443static int init_inodecache(void)
2444{
2445 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2446 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002447 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 return 0;
2449}
2450
2451static void destroy_inodecache(void)
2452{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002453 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454}
2455
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002456static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002458 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002460 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002461 .write_begin = shmem_write_begin,
2462 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002464 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002465 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466};
2467
Helge Deller15ad7cd2006-12-06 20:40:36 -08002468static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 .mmap = shmem_mmap,
2470#ifdef CONFIG_TMPFS
2471 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002472 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002473 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002474 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002475 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002476 .fsync = noop_fsync,
Hugh Dickinsae976412007-06-04 10:00:39 +02002477 .splice_read = generic_file_splice_read,
2478 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479#endif
2480};
2481
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002482static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002484 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002485#ifdef CONFIG_TMPFS_POSIX_ACL
2486 .setxattr = generic_setxattr,
2487 .getxattr = generic_getxattr,
2488 .listxattr = generic_listxattr,
2489 .removexattr = generic_removexattr,
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002490 .check_acl = generic_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002491#endif
2492
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493};
2494
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002495static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496#ifdef CONFIG_TMPFS
2497 .create = shmem_create,
2498 .lookup = simple_lookup,
2499 .link = shmem_link,
2500 .unlink = shmem_unlink,
2501 .symlink = shmem_symlink,
2502 .mkdir = shmem_mkdir,
2503 .rmdir = shmem_rmdir,
2504 .mknod = shmem_mknod,
2505 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002507#ifdef CONFIG_TMPFS_POSIX_ACL
2508 .setattr = shmem_notify_change,
2509 .setxattr = generic_setxattr,
2510 .getxattr = generic_getxattr,
2511 .listxattr = generic_listxattr,
2512 .removexattr = generic_removexattr,
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002513 .check_acl = generic_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002514#endif
2515};
2516
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002517static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002518#ifdef CONFIG_TMPFS_POSIX_ACL
2519 .setattr = shmem_notify_change,
2520 .setxattr = generic_setxattr,
2521 .getxattr = generic_getxattr,
2522 .listxattr = generic_listxattr,
2523 .removexattr = generic_removexattr,
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01002524 .check_acl = generic_check_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002525#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526};
2527
Hugh Dickins759b9772007-03-05 00:30:28 -08002528static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 .alloc_inode = shmem_alloc_inode,
2530 .destroy_inode = shmem_destroy_inode,
2531#ifdef CONFIG_TMPFS
2532 .statfs = shmem_statfs,
2533 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002534 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535#endif
Al Viro1f895f72010-06-05 19:10:41 -04002536 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 .drop_inode = generic_delete_inode,
2538 .put_super = shmem_put_super,
2539};
2540
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002541static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002542 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543#ifdef CONFIG_NUMA
2544 .set_policy = shmem_set_policy,
2545 .get_policy = shmem_get_policy,
2546#endif
2547};
2548
2549
Al Viro3c26ff62010-07-25 11:46:36 +04002550static struct dentry *shmem_mount(struct file_system_type *fs_type,
2551 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552{
Al Viro3c26ff62010-07-25 11:46:36 +04002553 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554}
2555
2556static struct file_system_type tmpfs_fs_type = {
2557 .owner = THIS_MODULE,
2558 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002559 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 .kill_sb = kill_litter_super,
2561};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Kay Sievers2b2af542009-04-30 15:23:42 +02002563int __init init_tmpfs(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564{
2565 int error;
2566
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002567 error = bdi_init(&shmem_backing_dev_info);
2568 if (error)
2569 goto out4;
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 error = init_inodecache();
2572 if (error)
2573 goto out3;
2574
2575 error = register_filesystem(&tmpfs_fs_type);
2576 if (error) {
2577 printk(KERN_ERR "Could not register tmpfs\n");
2578 goto out2;
2579 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002580
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002581 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 tmpfs_fs_type.name, NULL);
2583 if (IS_ERR(shm_mnt)) {
2584 error = PTR_ERR(shm_mnt);
2585 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2586 goto out1;
2587 }
2588 return 0;
2589
2590out1:
2591 unregister_filesystem(&tmpfs_fs_type);
2592out2:
2593 destroy_inodecache();
2594out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002595 bdi_destroy(&shmem_backing_dev_info);
2596out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 shm_mnt = ERR_PTR(error);
2598 return error;
2599}
Matt Mackall853ac432009-01-06 14:40:20 -08002600
Daisuke Nishimura87946a72010-05-26 14:42:39 -07002601#ifdef CONFIG_CGROUP_MEM_RES_CTLR
2602/**
2603 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2604 * @inode: the inode to be searched
2605 * @pgoff: the offset to be searched
2606 * @pagep: the pointer for the found page to be stored
2607 * @ent: the pointer for the found swap entry to be stored
2608 *
2609 * If a page is found, refcount of it is incremented. Callers should handle
2610 * these refcount.
2611 */
2612void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2613 struct page **pagep, swp_entry_t *ent)
2614{
2615 swp_entry_t entry = { .val = 0 }, *ptr;
2616 struct page *page = NULL;
2617 struct shmem_inode_info *info = SHMEM_I(inode);
2618
2619 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2620 goto out;
2621
2622 spin_lock(&info->lock);
2623 ptr = shmem_swp_entry(info, pgoff, NULL);
2624#ifdef CONFIG_SWAP
2625 if (ptr && ptr->val) {
2626 entry.val = ptr->val;
2627 page = find_get_page(&swapper_space, entry.val);
2628 } else
2629#endif
2630 page = find_get_page(inode->i_mapping, pgoff);
2631 if (ptr)
2632 shmem_swp_unmap(ptr);
2633 spin_unlock(&info->lock);
2634out:
2635 *pagep = page;
2636 *ent = entry;
2637}
2638#endif
2639
Matt Mackall853ac432009-01-06 14:40:20 -08002640#else /* !CONFIG_SHMEM */
2641
2642/*
2643 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2644 *
2645 * This is intended for small system where the benefits of the full
2646 * shmem code (swap-backed and resource-limited) are outweighed by
2647 * their complexity. On systems without swap this code should be
2648 * effectively equivalent, but much lighter weight.
2649 */
2650
2651#include <linux/ramfs.h>
2652
2653static struct file_system_type tmpfs_fs_type = {
2654 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002655 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002656 .kill_sb = kill_litter_super,
2657};
2658
Kay Sievers2b2af542009-04-30 15:23:42 +02002659int __init init_tmpfs(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002660{
2661 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2662
2663 shm_mnt = kern_mount(&tmpfs_fs_type);
2664 BUG_ON(IS_ERR(shm_mnt));
2665
2666 return 0;
2667}
2668
2669int shmem_unuse(swp_entry_t entry, struct page *page)
2670{
2671 return 0;
2672}
2673
Hugh Dickins3f96b792009-09-21 17:03:37 -07002674int shmem_lock(struct file *file, int lock, struct user_struct *user)
2675{
2676 return 0;
2677}
2678
Daisuke Nishimura87946a72010-05-26 14:42:39 -07002679#ifdef CONFIG_CGROUP_MEM_RES_CTLR
2680/**
2681 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2682 * @inode: the inode to be searched
2683 * @pgoff: the offset to be searched
2684 * @pagep: the pointer for the found page to be stored
2685 * @ent: the pointer for the found swap entry to be stored
2686 *
2687 * If a page is found, refcount of it is incremented. Callers should handle
2688 * these refcount.
2689 */
2690void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2691 struct page **pagep, swp_entry_t *ent)
2692{
2693 struct page *page = NULL;
2694
2695 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2696 goto out;
2697 page = find_get_page(inode->i_mapping, pgoff);
2698out:
2699 *pagep = page;
2700 *ent = (swp_entry_t){ .val = 0 };
2701}
2702#endif
2703
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002704#define shmem_vm_ops generic_file_vm_ops
2705#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002706#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002707#define shmem_acct_size(flags, size) 0
2708#define shmem_unacct_size(flags, size) do {} while (0)
Hugh Dickinscaefba12009-04-13 14:40:12 -07002709#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
Matt Mackall853ac432009-01-06 14:40:20 -08002710
2711#endif /* CONFIG_SHMEM */
2712
2713/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Randy Dunlap46711812008-03-19 17:00:41 -07002715/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 * @name: name for dentry (to be seen in /proc/<pid>/maps
2718 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002719 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002721struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
2723 int error;
2724 struct file *file;
2725 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002726 struct path path;
2727 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 struct qstr this;
2729
2730 if (IS_ERR(shm_mnt))
2731 return (void *)shm_mnt;
2732
2733 if (size < 0 || size > SHMEM_MAX_BYTES)
2734 return ERR_PTR(-EINVAL);
2735
2736 if (shmem_acct_size(flags, size))
2737 return ERR_PTR(-ENOMEM);
2738
2739 error = -ENOMEM;
2740 this.name = name;
2741 this.len = strlen(name);
2742 this.hash = 0; /* will go */
2743 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002744 path.dentry = d_alloc(root, &this);
2745 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002747 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002750 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002752 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
Al Viro2c48b9c2009-08-09 00:52:35 +04002754 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 inode->i_size = size;
2756 inode->i_nlink = 0; /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002757#ifndef CONFIG_MMU
2758 error = ramfs_nommu_expand_for_mapping(inode, size);
2759 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002760 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002761#endif
Al Viro4b42af82009-08-05 18:25:56 +04002762
2763 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002764 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002765 &shmem_file_operations);
2766 if (!file)
2767 goto put_dentry;
2768
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 return file;
2770
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002772 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773put_memory:
2774 shmem_unacct_size(flags, size);
2775 return ERR_PTR(error);
2776}
Keith Packard395e0dd2008-06-20 00:08:06 -07002777EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778
Randy Dunlap46711812008-03-19 17:00:41 -07002779/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2782 */
2783int shmem_zero_setup(struct vm_area_struct *vma)
2784{
2785 struct file *file;
2786 loff_t size = vma->vm_end - vma->vm_start;
2787
2788 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2789 if (IS_ERR(file))
2790 return PTR_ERR(file);
2791
2792 if (vma->vm_file)
2793 fput(vma->vm_file);
2794 vma->vm_file = file;
2795 vma->vm_ops = &shmem_vm_ops;
Hugh Dickinsbee4c362011-03-22 16:33:43 -07002796 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 return 0;
2798}