blob: e281f0658cb86d3da24fa2e97e01f8a705b1e0d9 [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 Dickins6922c0c2011-08-03 16:21:25 -07009 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -070011 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 *
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 *
Matt Mackall853ac432009-01-06 14:40:20 -080018 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * This file is released under the GPL.
22 */
23
Matt Mackall853ac432009-01-06 14:40:20 -080024#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070028#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080029#include <linux/file.h>
30#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040031#include <linux/export.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>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070054#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070055#include <linux/percpu_counter.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070056#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/security.h>
58#include <linux/swapops.h>
59#include <linux/mempolicy.h>
60#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000061#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070062#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070063#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080064#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040065#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070066
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <asm/pgtable.h>
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073/* Pretend that each entry is of this size in directory's i_size */
74#define BOGO_DIRENT_SIZE 20
75
Hugh Dickins69f07ec2011-08-03 16:21:26 -070076/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
77#define SHORT_SYMLINK_LEN 128
78
Eric Parisb09e0fa2011-05-24 17:12:39 -070079struct shmem_xattr {
80 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
81 char *name; /* xattr name */
82 size_t size;
83 char value[0];
84};
85
Hugh Dickins285b2c42011-08-03 16:21:20 -070086/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -070087enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 SGP_READ, /* don't exceed i_size, don't allocate page */
89 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080090 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 SGP_WRITE, /* may exceed i_size, may allocate page */
92};
93
Andrew Mortonb76db732008-02-08 04:21:49 -080094#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080095static unsigned long shmem_default_max_blocks(void)
96{
97 return totalram_pages / 2;
98}
99
100static unsigned long shmem_default_max_inodes(void)
101{
102 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
103}
Andrew Mortonb76db732008-02-08 04:21:49 -0800104#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800105
Hugh Dickins68da9f02011-07-25 17:12:34 -0700106static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
107 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
108
109static inline int shmem_getpage(struct inode *inode, pgoff_t index,
110 struct page **pagep, enum sgp_type sgp, int *fault_type)
111{
112 return shmem_getpage_gfp(inode, index, pagep, sgp,
113 mapping_gfp_mask(inode->i_mapping), fault_type);
114}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
117{
118 return sb->s_fs_info;
119}
120
121/*
122 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
123 * for shared memory and for shared anonymous (/dev/zero) mappings
124 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
125 * consistent with the pre-accounting of private mappings ...
126 */
127static inline int shmem_acct_size(unsigned long flags, loff_t size)
128{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000129 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000130 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
133static inline void shmem_unacct_size(unsigned long flags, loff_t size)
134{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000135 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 vm_unacct_memory(VM_ACCT(size));
137}
138
139/*
140 * ... whereas tmpfs objects are accounted incrementally as
141 * pages are allocated, in order to allow huge sparse files.
142 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
143 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
144 */
145static inline int shmem_acct_block(unsigned long flags)
146{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000147 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000148 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151static inline void shmem_unacct_blocks(unsigned long flags, long pages)
152{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000153 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
155}
156
Hugh Dickins759b9772007-03-05 00:30:28 -0800157static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700158static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800159static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800160static const struct inode_operations shmem_inode_operations;
161static const struct inode_operations shmem_dir_inode_operations;
162static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400163static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700165static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700167 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168};
169
170static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800171static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800173static int shmem_reserve_inode(struct super_block *sb)
174{
175 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
176 if (sbinfo->max_inodes) {
177 spin_lock(&sbinfo->stat_lock);
178 if (!sbinfo->free_inodes) {
179 spin_unlock(&sbinfo->stat_lock);
180 return -ENOSPC;
181 }
182 sbinfo->free_inodes--;
183 spin_unlock(&sbinfo->stat_lock);
184 }
185 return 0;
186}
187
188static void shmem_free_inode(struct super_block *sb)
189{
190 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
191 if (sbinfo->max_inodes) {
192 spin_lock(&sbinfo->stat_lock);
193 sbinfo->free_inodes++;
194 spin_unlock(&sbinfo->stat_lock);
195 }
196}
197
Randy Dunlap46711812008-03-19 17:00:41 -0700198/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700199 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 * @inode: inode to recalc
201 *
202 * We have to calculate the free blocks since the mm can drop
203 * undirtied hole pages behind our back.
204 *
205 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
206 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
207 *
208 * It has to be called with the spinlock held.
209 */
210static void shmem_recalc_inode(struct inode *inode)
211{
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 long freed;
214
215 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
216 if (freed > 0) {
Hugh Dickins54af6042011-08-03 16:21:24 -0700217 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
218 if (sbinfo->max_blocks)
219 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 info->alloced -= freed;
Hugh Dickins54af6042011-08-03 16:21:24 -0700221 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 }
224}
225
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700226/*
227 * Replace item expected in radix tree by a new item, while holding tree lock.
228 */
229static int shmem_radix_tree_replace(struct address_space *mapping,
230 pgoff_t index, void *expected, void *replacement)
231{
232 void **pslot;
233 void *item = NULL;
234
235 VM_BUG_ON(!expected);
236 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
237 if (pslot)
238 item = radix_tree_deref_slot_protected(pslot,
239 &mapping->tree_lock);
240 if (item != expected)
241 return -ENOENT;
242 if (replacement)
243 radix_tree_replace_slot(pslot, replacement);
244 else
245 radix_tree_delete(&mapping->page_tree, index);
246 return 0;
247}
248
249/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700250 * Like add_to_page_cache_locked, but error if expected item has gone.
251 */
252static int shmem_add_to_page_cache(struct page *page,
253 struct address_space *mapping,
254 pgoff_t index, gfp_t gfp, void *expected)
255{
Hugh Dickinsaa3b1892011-08-03 16:21:24 -0700256 int error = 0;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700257
258 VM_BUG_ON(!PageLocked(page));
259 VM_BUG_ON(!PageSwapBacked(page));
260
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700261 if (!expected)
262 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
263 if (!error) {
264 page_cache_get(page);
265 page->mapping = mapping;
266 page->index = index;
267
268 spin_lock_irq(&mapping->tree_lock);
269 if (!expected)
270 error = radix_tree_insert(&mapping->page_tree,
271 index, page);
272 else
273 error = shmem_radix_tree_replace(mapping, index,
274 expected, page);
275 if (!error) {
276 mapping->nrpages++;
277 __inc_zone_page_state(page, NR_FILE_PAGES);
278 __inc_zone_page_state(page, NR_SHMEM);
279 spin_unlock_irq(&mapping->tree_lock);
280 } else {
281 page->mapping = NULL;
282 spin_unlock_irq(&mapping->tree_lock);
283 page_cache_release(page);
284 }
285 if (!expected)
286 radix_tree_preload_end();
287 }
288 if (error)
289 mem_cgroup_uncharge_cache_page(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700290 return error;
291}
292
293/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700294 * Like delete_from_page_cache, but substitutes swap for page.
295 */
296static void shmem_delete_from_page_cache(struct page *page, void *radswap)
297{
298 struct address_space *mapping = page->mapping;
299 int error;
300
301 spin_lock_irq(&mapping->tree_lock);
302 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
303 page->mapping = NULL;
304 mapping->nrpages--;
305 __dec_zone_page_state(page, NR_FILE_PAGES);
306 __dec_zone_page_state(page, NR_SHMEM);
307 spin_unlock_irq(&mapping->tree_lock);
308 page_cache_release(page);
309 BUG_ON(error);
310}
311
312/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700313 * Like find_get_pages, but collecting swap entries as well as pages.
314 */
315static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
316 pgoff_t start, unsigned int nr_pages,
317 struct page **pages, pgoff_t *indices)
318{
319 unsigned int i;
320 unsigned int ret;
321 unsigned int nr_found;
322
323 rcu_read_lock();
324restart:
325 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
326 (void ***)pages, indices, start, nr_pages);
327 ret = 0;
328 for (i = 0; i < nr_found; i++) {
329 struct page *page;
330repeat:
331 page = radix_tree_deref_slot((void **)pages[i]);
332 if (unlikely(!page))
333 continue;
334 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700335 if (radix_tree_deref_retry(page))
336 goto restart;
337 /*
338 * Otherwise, we must be storing a swap entry
339 * here as an exceptional entry: so return it
340 * without attempting to raise page count.
341 */
342 goto export;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700343 }
344 if (!page_cache_get_speculative(page))
345 goto repeat;
346
347 /* Has the page moved? */
348 if (unlikely(page != *((void **)pages[i]))) {
349 page_cache_release(page);
350 goto repeat;
351 }
352export:
353 indices[ret] = indices[i];
354 pages[ret] = page;
355 ret++;
356 }
357 if (unlikely(!ret && nr_found))
358 goto restart;
359 rcu_read_unlock();
360 return ret;
361}
362
363/*
364 * Remove swap entry from radix tree, free the swap and its page cache.
365 */
366static int shmem_free_swap(struct address_space *mapping,
367 pgoff_t index, void *radswap)
368{
369 int error;
370
371 spin_lock_irq(&mapping->tree_lock);
372 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
373 spin_unlock_irq(&mapping->tree_lock);
374 if (!error)
375 free_swap_and_cache(radix_to_swp_entry(radswap));
376 return error;
377}
378
379/*
380 * Pagevec may contain swap entries, so shuffle up pages before releasing.
381 */
Hugh Dickins24513262012-01-20 14:34:21 -0800382static void shmem_deswap_pagevec(struct pagevec *pvec)
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700383{
384 int i, j;
385
386 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
387 struct page *page = pvec->pages[i];
388 if (!radix_tree_exceptional_entry(page))
389 pvec->pages[j++] = page;
390 }
391 pvec->nr = j;
Hugh Dickins24513262012-01-20 14:34:21 -0800392}
393
394/*
395 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
396 */
397void shmem_unlock_mapping(struct address_space *mapping)
398{
399 struct pagevec pvec;
400 pgoff_t indices[PAGEVEC_SIZE];
401 pgoff_t index = 0;
402
403 pagevec_init(&pvec, 0);
404 /*
405 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
406 */
407 while (!mapping_unevictable(mapping)) {
408 /*
409 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
410 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
411 */
412 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
413 PAGEVEC_SIZE, pvec.pages, indices);
414 if (!pvec.nr)
415 break;
416 index = indices[pvec.nr - 1] + 1;
417 shmem_deswap_pagevec(&pvec);
418 check_move_unevictable_pages(pvec.pages, pvec.nr);
419 pagevec_release(&pvec);
420 cond_resched();
421 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700422}
423
424/*
425 * Remove range of pages and swap entries from radix tree, and free them.
426 */
Hugh Dickins285b2c42011-08-03 16:21:20 -0700427void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700429 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700431 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700432 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700433 pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700434 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700435 pgoff_t indices[PAGEVEC_SIZE];
436 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700437 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700438 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700440 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
441
442 pagevec_init(&pvec, 0);
443 index = start;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700444 while (index <= end) {
445 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
446 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
447 pvec.pages, indices);
448 if (!pvec.nr)
449 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700450 mem_cgroup_uncharge_start();
451 for (i = 0; i < pagevec_count(&pvec); i++) {
452 struct page *page = pvec.pages[i];
453
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700454 index = indices[i];
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700455 if (index > end)
456 break;
457
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700458 if (radix_tree_exceptional_entry(page)) {
459 nr_swaps_freed += !shmem_free_swap(mapping,
460 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700461 continue;
462 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700463
464 if (!trylock_page(page))
465 continue;
466 if (page->mapping == mapping) {
467 VM_BUG_ON(PageWriteback(page));
468 truncate_inode_page(mapping, page);
469 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700470 unlock_page(page);
471 }
Hugh Dickins24513262012-01-20 14:34:21 -0800472 shmem_deswap_pagevec(&pvec);
473 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700474 mem_cgroup_uncharge_end();
475 cond_resched();
476 index++;
477 }
478
479 if (partial) {
480 struct page *page = NULL;
481 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
482 if (page) {
483 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
484 set_page_dirty(page);
485 unlock_page(page);
486 page_cache_release(page);
487 }
488 }
489
490 index = start;
491 for ( ; ; ) {
492 cond_resched();
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700493 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
494 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
495 pvec.pages, indices);
496 if (!pvec.nr) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700497 if (index == start)
498 break;
499 index = start;
500 continue;
501 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700502 if (index == start && indices[0] > end) {
Hugh Dickins24513262012-01-20 14:34:21 -0800503 shmem_deswap_pagevec(&pvec);
504 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700505 break;
506 }
507 mem_cgroup_uncharge_start();
508 for (i = 0; i < pagevec_count(&pvec); i++) {
509 struct page *page = pvec.pages[i];
510
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700511 index = indices[i];
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700512 if (index > end)
513 break;
514
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700515 if (radix_tree_exceptional_entry(page)) {
516 nr_swaps_freed += !shmem_free_swap(mapping,
517 index, page);
518 continue;
519 }
520
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700521 lock_page(page);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700522 if (page->mapping == mapping) {
523 VM_BUG_ON(PageWriteback(page));
524 truncate_inode_page(mapping, page);
525 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700526 unlock_page(page);
527 }
Hugh Dickins24513262012-01-20 14:34:21 -0800528 shmem_deswap_pagevec(&pvec);
529 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700530 mem_cgroup_uncharge_end();
531 index++;
532 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700535 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 shmem_recalc_inode(inode);
537 spin_unlock(&info->lock);
538
Hugh Dickins285b2c42011-08-03 16:21:20 -0700539 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700541EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Hugh Dickins94c1e622011-06-27 16:18:03 -0700543static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
545 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 int error;
547
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200548 error = inode_change_ok(inode, attr);
549 if (error)
550 return error;
551
Hugh Dickins94c1e622011-06-27 16:18:03 -0700552 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
553 loff_t oldsize = inode->i_size;
554 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000555
Hugh Dickins94c1e622011-06-27 16:18:03 -0700556 if (newsize != oldsize) {
557 i_size_write(inode, newsize);
558 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
559 }
560 if (newsize < oldsize) {
561 loff_t holebegin = round_up(newsize, PAGE_SIZE);
562 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
563 shmem_truncate_range(inode, newsize, (loff_t)-1);
564 /* unmap again to remove racily COWed private pages */
565 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200569 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700570#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200571 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100572 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700573#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 return error;
575}
576
Al Viro1f895f72010-06-05 19:10:41 -0400577static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 struct shmem_inode_info *info = SHMEM_I(inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700580 struct shmem_xattr *xattr, *nxattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000582 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 shmem_unacct_size(info->flags, inode->i_size);
584 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000585 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800587 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800589 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700591 } else
592 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700593
594 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
595 kfree(xattr->name);
596 kfree(xattr);
597 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700598 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800599 shmem_free_inode(inode->i_sb);
Al Viro1f895f72010-06-05 19:10:41 -0400600 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700603/*
604 * If swap found in inode, free it and move page from swapcache to filecache.
605 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700606static int shmem_unuse_inode(struct shmem_inode_info *info,
607 swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700609 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700610 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700611 pgoff_t index;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800612 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700614 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700615 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700616 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700617 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800618
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800619 /*
620 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400621 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800622 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700623 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800624 */
625 if (shmem_swaplist.next != &info->swaplist)
626 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800627
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800628 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700629 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
630 * but also to hold up shmem_evict_inode(): so inode cannot be freed
631 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800632 */
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700633 error = shmem_add_to_page_cache(page, mapping, index,
634 GFP_NOWAIT, radswap);
Hugh Dickins778dd892011-05-11 15:13:37 -0700635 /* which does mem_cgroup_uncharge_cache_page on error */
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700636
Hugh Dickins48f170f2011-07-25 17:12:37 -0700637 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700638 /*
639 * Truncation and eviction use free_swap_and_cache(), which
640 * only does trylock page: if we raced, best clean up here.
641 */
Hugh Dickins73b12622008-02-04 22:28:50 -0800642 delete_from_swap_cache(page);
643 set_page_dirty(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700644 if (!error) {
645 spin_lock(&info->lock);
646 info->swapped--;
647 spin_unlock(&info->lock);
648 swap_free(swap);
649 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800650 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800652 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
654
655/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700656 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700658int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700660 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 struct shmem_inode_info *info;
662 int found = 0;
Hugh Dickins778dd892011-05-11 15:13:37 -0700663 int error;
664
665 /*
666 * Charge page using GFP_KERNEL while we can wait, before taking
667 * the shmem_swaplist_mutex which might hold up shmem_writepage().
668 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700669 */
670 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
671 if (error)
672 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700673 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800675 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700676 list_for_each_safe(this, next, &shmem_swaplist) {
677 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700678 if (info->swapped)
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700679 found = shmem_unuse_inode(info, swap, page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700680 else
681 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800682 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800683 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700684 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800686 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700687
Hugh Dickins778dd892011-05-11 15:13:37 -0700688 if (!found)
689 mem_cgroup_uncharge_cache_page(page);
690 if (found < 0)
691 error = found;
692out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800693 unlock_page(page);
694 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700695 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
698/*
699 * Move the page from the page cache to the swap cache.
700 */
701static int shmem_writepage(struct page *page, struct writeback_control *wbc)
702{
703 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700706 swp_entry_t swap;
707 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
709 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 mapping = page->mapping;
711 index = page->index;
712 inode = mapping->host;
713 info = SHMEM_I(inode);
714 if (info->flags & VM_LOCKED)
715 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800716 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 goto redirty;
718
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800719 /*
720 * shmem_backing_dev_info's capabilities prevent regular writeback or
721 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700722 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800723 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700724 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800725 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700726 if (!wbc->for_reclaim) {
727 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
728 goto redirty;
729 }
730 swap = get_swap_page();
731 if (!swap.val)
732 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800733
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700734 /*
735 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700736 * if it's not already there. Do it now before the page is
737 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700738 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700739 * we've incremented swapped, because shmem_unuse_inode() will
740 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700741 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700742 mutex_lock(&shmem_swaplist_mutex);
743 if (list_empty(&info->swaplist))
744 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700745
Hugh Dickins48f170f2011-07-25 17:12:37 -0700746 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800747 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700748 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
749
750 spin_lock(&info->lock);
751 info->swapped++;
752 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700753 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700754
755 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800756 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700757 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return 0;
759 }
760
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700761 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700762 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763redirty:
764 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800765 if (wbc->for_reclaim)
766 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
767 unlock_page(page);
768 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
771#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800772#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700773static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800774{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700775 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800776
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700777 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700778 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800779
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700780 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800781
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700782 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800783}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700784
785static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
786{
787 struct mempolicy *mpol = NULL;
788 if (sbinfo->mpol) {
789 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
790 mpol = sbinfo->mpol;
791 mpol_get(mpol);
792 spin_unlock(&sbinfo->stat_lock);
793 }
794 return mpol;
795}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800796#endif /* CONFIG_TMPFS */
797
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700798static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
799 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700801 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 struct vm_area_struct pvma;
803
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700804 spol = mpol_cond_copy(&mpol,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700805 mpol_shared_policy_lookup(&info->policy, index));
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800808 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700809 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800810 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700811 pvma.vm_policy = spol;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700812 return swapin_readahead(swap, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
814
Hugh Dickins02098fe2008-02-04 22:28:42 -0800815static struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700816 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
818 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800820 /* Create a pseudo vma that just contains the policy */
821 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700822 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800823 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700824 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700825
826 /*
827 * alloc_page_vma() will drop the shared policy reference
828 */
829 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800831#else /* !CONFIG_NUMA */
832#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700833static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800834{
835}
836#endif /* CONFIG_TMPFS */
837
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700838static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
839 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700841 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
Hugh Dickins02098fe2008-02-04 22:28:42 -0800844static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700845 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000847 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800849#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700851#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
852static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
853{
854 return NULL;
855}
856#endif
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858/*
Hugh Dickins68da9f02011-07-25 17:12:34 -0700859 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 *
861 * If we allocate a new one we do not mark it dirty. That's up to the
862 * vm. If we swap it in we mark it dirty since we also free the swap
863 * entry since a page cannot live in both the swap and page cache
864 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700865static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -0700866 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867{
868 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af6042011-08-03 16:21:24 -0700869 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -0700871 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 swp_entry_t swap;
873 int error;
Hugh Dickins54af6042011-08-03 16:21:24 -0700874 int once = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700876 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878repeat:
Hugh Dickins54af6042011-08-03 16:21:24 -0700879 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700880 page = find_lock_page(mapping, index);
Hugh Dickins54af6042011-08-03 16:21:24 -0700881 if (radix_tree_exceptional_entry(page)) {
882 swap = radix_to_swp_entry(page);
883 page = NULL;
884 }
885
886 if (sgp != SGP_WRITE &&
887 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
888 error = -EINVAL;
889 goto failed;
890 }
891
892 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800893 /*
Hugh Dickins27ab7002011-07-25 17:12:36 -0700894 * Once we can get the page lock, it must be uptodate:
895 * if there were an error in reading back from swap,
896 * the page would not be inserted into the filecache.
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800897 */
Hugh Dickins54af6042011-08-03 16:21:24 -0700898 BUG_ON(page && !PageUptodate(page));
899 *pagep = page;
900 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -0700901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 /*
Hugh Dickins54af6042011-08-03 16:21:24 -0700904 * Fast cache lookup did not find it:
905 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 */
Hugh Dickins54af6042011-08-03 16:21:24 -0700907 info = SHMEM_I(inode);
908 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (swap.val) {
911 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -0700912 page = lookup_swap_cache(swap);
913 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -0700914 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -0700915 if (fault_type)
916 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700917 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700918 if (!page) {
Hugh Dickins54af6042011-08-03 16:21:24 -0700919 error = -ENOMEM;
920 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
923
924 /* We have to do this with page locked to prevent races */
Hugh Dickins54af6042011-08-03 16:21:24 -0700925 lock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700926 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 error = -EIO;
Hugh Dickins54af6042011-08-03 16:21:24 -0700928 goto failed;
929 }
930 wait_on_page_writeback(page);
931
932 /* Someone may have already done it for us */
933 if (page->mapping) {
934 if (page->mapping == mapping &&
935 page->index == index)
936 goto done;
937 error = -EEXIST;
938 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940
Hugh Dickinsaa3b1892011-08-03 16:21:24 -0700941 error = mem_cgroup_cache_charge(page, current->mm,
942 gfp & GFP_RECLAIM_MASK);
943 if (!error)
944 error = shmem_add_to_page_cache(page, mapping, index,
945 gfp, swp_to_radix_entry(swap));
Hugh Dickins54af6042011-08-03 16:21:24 -0700946 if (error)
947 goto failed;
948
949 spin_lock(&info->lock);
950 info->swapped--;
951 shmem_recalc_inode(inode);
952 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700953
Hugh Dickins27ab7002011-07-25 17:12:36 -0700954 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700955 set_page_dirty(page);
956 swap_free(swap);
957
Hugh Dickins54af6042011-08-03 16:21:24 -0700958 } else {
959 if (shmem_acct_block(info->flags)) {
960 error = -ENOSPC;
961 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700963 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -0700964 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af6042011-08-03 16:21:24 -0700965 sbinfo->max_blocks) >= 0) {
966 error = -ENOSPC;
967 goto unacct;
968 }
Tim Chen7e496292010-08-09 17:19:05 -0700969 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -0700970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Hugh Dickins54af6042011-08-03 16:21:24 -0700972 page = shmem_alloc_page(gfp, info, index);
973 if (!page) {
974 error = -ENOMEM;
975 goto decused;
976 }
977
978 SetPageSwapBacked(page);
979 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -0700980 error = mem_cgroup_cache_charge(page, current->mm,
981 gfp & GFP_RECLAIM_MASK);
982 if (!error)
983 error = shmem_add_to_page_cache(page, mapping, index,
984 gfp, NULL);
Hugh Dickins54af6042011-08-03 16:21:24 -0700985 if (error)
986 goto decused;
987 lru_cache_add_anon(page);
988
989 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 info->alloced++;
Hugh Dickins54af6042011-08-03 16:21:24 -0700991 inode->i_blocks += BLOCKS_PER_PAGE;
992 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 spin_unlock(&info->lock);
Hugh Dickins54af6042011-08-03 16:21:24 -0700994
Hugh Dickins27ab7002011-07-25 17:12:36 -0700995 clear_highpage(page);
996 flush_dcache_page(page);
997 SetPageUptodate(page);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800998 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -0700999 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001done:
Hugh Dickins54af6042011-08-03 16:21:24 -07001002 /* Perhaps the file has been truncated since we checked */
1003 if (sgp != SGP_WRITE &&
1004 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1005 error = -EINVAL;
1006 goto trunc;
Shaohua Liff36b802010-08-09 17:19:06 -07001007 }
Hugh Dickins54af6042011-08-03 16:21:24 -07001008 *pagep = page;
1009 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001010
Nick Piggind0217ac2007-07-19 01:47:03 -07001011 /*
Hugh Dickins54af6042011-08-03 16:21:24 -07001012 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 */
Hugh Dickins54af6042011-08-03 16:21:24 -07001014trunc:
1015 ClearPageDirty(page);
1016 delete_from_page_cache(page);
1017 spin_lock(&info->lock);
1018 info->alloced--;
1019 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 spin_unlock(&info->lock);
Hugh Dickins54af6042011-08-03 16:21:24 -07001021decused:
1022 if (sbinfo->max_blocks)
1023 percpu_counter_add(&sbinfo->used_blocks, -1);
1024unacct:
1025 shmem_unacct_blocks(info->flags, 1);
1026failed:
1027 if (swap.val && error != -EINVAL) {
1028 struct page *test = find_get_page(mapping, index);
1029 if (test && !radix_tree_exceptional_entry(test))
1030 page_cache_release(test);
1031 /* Have another try if the entry has changed */
1032 if (test != swp_to_radix_entry(swap))
1033 error = -EEXIST;
1034 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001035 if (page) {
Hugh Dickins54af6042011-08-03 16:21:24 -07001036 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001037 page_cache_release(page);
Hugh Dickins54af6042011-08-03 16:21:24 -07001038 }
1039 if (error == -ENOSPC && !once++) {
1040 info = SHMEM_I(inode);
1041 spin_lock(&info->lock);
1042 shmem_recalc_inode(inode);
1043 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001044 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001045 }
Hugh Dickins54af6042011-08-03 16:21:24 -07001046 if (error == -EEXIST)
1047 goto repeat;
1048 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049}
1050
1051static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1052{
1053 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1054 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001055 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1058 if (error)
1059 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001060
Ying Han456f9982011-05-26 16:25:38 -07001061 if (ret & VM_FAULT_MAJOR) {
1062 count_vm_event(PGMAJFAULT);
1063 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1064 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001065 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066}
1067
1068#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001069static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001071 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1072 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001075static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1076 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001078 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1079 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001081 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1082 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083}
1084#endif
1085
1086int shmem_lock(struct file *file, int lock, struct user_struct *user)
1087{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001088 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 struct shmem_inode_info *info = SHMEM_I(inode);
1090 int retval = -ENOMEM;
1091
1092 spin_lock(&info->lock);
1093 if (lock && !(info->flags & VM_LOCKED)) {
1094 if (!user_shm_lock(inode->i_size, user))
1095 goto out_nomem;
1096 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001097 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 }
1099 if (!lock && (info->flags & VM_LOCKED) && user) {
1100 user_shm_unlock(inode->i_size, user);
1101 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001102 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
1104 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106out_nomem:
1107 spin_unlock(&info->lock);
1108 return retval;
1109}
1110
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001111static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
1113 file_accessed(file);
1114 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001115 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 return 0;
1117}
1118
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001119static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001120 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
1122 struct inode *inode;
1123 struct shmem_inode_info *info;
1124 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1125
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001126 if (shmem_reserve_inode(sb))
1127 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
1129 inode = new_inode(sb);
1130 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001131 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001132 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1135 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001136 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 info = SHMEM_I(inode);
1138 memset(info, 0, (char *)inode - (char *)info);
1139 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001140 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 INIT_LIST_HEAD(&info->swaplist);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001142 INIT_LIST_HEAD(&info->xattr_list);
Al Viro72c04902009-06-24 16:58:48 -04001143 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145 switch (mode & S_IFMT) {
1146 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001147 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 init_special_inode(inode, mode, dev);
1149 break;
1150 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001151 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 inode->i_op = &shmem_inode_operations;
1153 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001154 mpol_shared_policy_init(&info->policy,
1155 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 break;
1157 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001158 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 /* Some things misbehave if size == 0 on a directory */
1160 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1161 inode->i_op = &shmem_dir_inode_operations;
1162 inode->i_fop = &simple_dir_operations;
1163 break;
1164 case S_IFLNK:
1165 /*
1166 * Must not load anything in the rbtree,
1167 * mpol_free_shared_policy will not be called.
1168 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001169 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 break;
1171 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001172 } else
1173 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 return inode;
1175}
1176
1177#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001178static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001179static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001181#ifdef CONFIG_TMPFS_XATTR
1182static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1183#else
1184#define shmem_initxattrs NULL
1185#endif
1186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001188shmem_write_begin(struct file *file, struct address_space *mapping,
1189 loff_t pos, unsigned len, unsigned flags,
1190 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191{
Nick Piggin800d15a2007-10-16 01:25:03 -07001192 struct inode *inode = mapping->host;
1193 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001194 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1195}
1196
1197static int
1198shmem_write_end(struct file *file, struct address_space *mapping,
1199 loff_t pos, unsigned len, unsigned copied,
1200 struct page *page, void *fsdata)
1201{
1202 struct inode *inode = mapping->host;
1203
Hugh Dickinsd3602442008-02-04 22:28:44 -08001204 if (pos + copied > inode->i_size)
1205 i_size_write(inode, pos + copied);
1206
Nick Piggin800d15a2007-10-16 01:25:03 -07001207 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001208 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001209 page_cache_release(page);
1210
Nick Piggin800d15a2007-10-16 01:25:03 -07001211 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1215{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001216 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001218 pgoff_t index;
1219 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001220 enum sgp_type sgp = SGP_READ;
1221
1222 /*
1223 * Might this read be for a stacking filesystem? Then when reading
1224 * holes of a sparse file, we actually need to allocate those pages,
1225 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1226 */
1227 if (segment_eq(get_fs(), KERNEL_DS))
1228 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 index = *ppos >> PAGE_CACHE_SHIFT;
1231 offset = *ppos & ~PAGE_CACHE_MASK;
1232
1233 for (;;) {
1234 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001235 pgoff_t end_index;
1236 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 loff_t i_size = i_size_read(inode);
1238
1239 end_index = i_size >> PAGE_CACHE_SHIFT;
1240 if (index > end_index)
1241 break;
1242 if (index == end_index) {
1243 nr = i_size & ~PAGE_CACHE_MASK;
1244 if (nr <= offset)
1245 break;
1246 }
1247
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001248 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 if (desc->error) {
1250 if (desc->error == -EINVAL)
1251 desc->error = 0;
1252 break;
1253 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001254 if (page)
1255 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 /*
1258 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001259 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 */
1261 nr = PAGE_CACHE_SIZE;
1262 i_size = i_size_read(inode);
1263 end_index = i_size >> PAGE_CACHE_SHIFT;
1264 if (index == end_index) {
1265 nr = i_size & ~PAGE_CACHE_MASK;
1266 if (nr <= offset) {
1267 if (page)
1268 page_cache_release(page);
1269 break;
1270 }
1271 }
1272 nr -= offset;
1273
1274 if (page) {
1275 /*
1276 * If users can be writing to this page using arbitrary
1277 * virtual addresses, take care about potential aliasing
1278 * before reading the page on the kernel side.
1279 */
1280 if (mapping_writably_mapped(mapping))
1281 flush_dcache_page(page);
1282 /*
1283 * Mark the page accessed if we read the beginning.
1284 */
1285 if (!offset)
1286 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001287 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001289 page_cache_get(page);
1290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
1292 /*
1293 * Ok, we have the page, and it's up-to-date, so
1294 * now we can copy it to user space...
1295 *
1296 * The actor routine returns how many bytes were actually used..
1297 * NOTE! This may not be the same as how much of a user buffer
1298 * we filled up (we may be padding etc), so we can only update
1299 * "pos" here (the actor routine has to update the user buffer
1300 * pointers and the remaining count).
1301 */
1302 ret = actor(desc, page, offset, nr);
1303 offset += ret;
1304 index += offset >> PAGE_CACHE_SHIFT;
1305 offset &= ~PAGE_CACHE_MASK;
1306
1307 page_cache_release(page);
1308 if (ret != nr || !desc->count)
1309 break;
1310
1311 cond_resched();
1312 }
1313
1314 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1315 file_accessed(filp);
1316}
1317
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001318static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1319 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001321 struct file *filp = iocb->ki_filp;
1322 ssize_t retval;
1323 unsigned long seg;
1324 size_t count;
1325 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001327 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1328 if (retval)
1329 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001331 for (seg = 0; seg < nr_segs; seg++) {
1332 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001334 desc.written = 0;
1335 desc.arg.buf = iov[seg].iov_base;
1336 desc.count = iov[seg].iov_len;
1337 if (desc.count == 0)
1338 continue;
1339 desc.error = 0;
1340 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1341 retval += desc.written;
1342 if (desc.error) {
1343 retval = retval ?: desc.error;
1344 break;
1345 }
1346 if (desc.count > 0)
1347 break;
1348 }
1349 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350}
1351
Hugh Dickins708e3502011-07-25 17:12:32 -07001352static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1353 struct pipe_inode_info *pipe, size_t len,
1354 unsigned int flags)
1355{
1356 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001357 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001358 unsigned int loff, nr_pages, req_pages;
1359 struct page *pages[PIPE_DEF_BUFFERS];
1360 struct partial_page partial[PIPE_DEF_BUFFERS];
1361 struct page *page;
1362 pgoff_t index, end_index;
1363 loff_t isize, left;
1364 int error, page_nr;
1365 struct splice_pipe_desc spd = {
1366 .pages = pages,
1367 .partial = partial,
1368 .flags = flags,
1369 .ops = &page_cache_pipe_buf_ops,
1370 .spd_release = spd_release_page,
1371 };
1372
Hugh Dickins71f0e072011-07-25 17:12:33 -07001373 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001374 if (unlikely(*ppos >= isize))
1375 return 0;
1376
1377 left = isize - *ppos;
1378 if (unlikely(left < len))
1379 len = left;
1380
1381 if (splice_grow_spd(pipe, &spd))
1382 return -ENOMEM;
1383
1384 index = *ppos >> PAGE_CACHE_SHIFT;
1385 loff = *ppos & ~PAGE_CACHE_MASK;
1386 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1387 nr_pages = min(req_pages, pipe->buffers);
1388
Hugh Dickins708e3502011-07-25 17:12:32 -07001389 spd.nr_pages = find_get_pages_contig(mapping, index,
1390 nr_pages, spd.pages);
1391 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001392 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001393
Hugh Dickins708e3502011-07-25 17:12:32 -07001394 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001395 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1396 if (error)
1397 break;
1398 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001399 spd.pages[spd.nr_pages++] = page;
1400 index++;
1401 }
1402
Hugh Dickins708e3502011-07-25 17:12:32 -07001403 index = *ppos >> PAGE_CACHE_SHIFT;
1404 nr_pages = spd.nr_pages;
1405 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001406
Hugh Dickins708e3502011-07-25 17:12:32 -07001407 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1408 unsigned int this_len;
1409
1410 if (!len)
1411 break;
1412
Hugh Dickins708e3502011-07-25 17:12:32 -07001413 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1414 page = spd.pages[page_nr];
1415
Hugh Dickins71f0e072011-07-25 17:12:33 -07001416 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001417 error = shmem_getpage(inode, index, &page,
1418 SGP_CACHE, NULL);
1419 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001420 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001421 unlock_page(page);
1422 page_cache_release(spd.pages[page_nr]);
1423 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001424 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001425
1426 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001427 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1428 if (unlikely(!isize || index > end_index))
1429 break;
1430
Hugh Dickins708e3502011-07-25 17:12:32 -07001431 if (end_index == index) {
1432 unsigned int plen;
1433
Hugh Dickins708e3502011-07-25 17:12:32 -07001434 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1435 if (plen <= loff)
1436 break;
1437
Hugh Dickins708e3502011-07-25 17:12:32 -07001438 this_len = min(this_len, plen - loff);
1439 len = this_len;
1440 }
1441
1442 spd.partial[page_nr].offset = loff;
1443 spd.partial[page_nr].len = this_len;
1444 len -= this_len;
1445 loff = 0;
1446 spd.nr_pages++;
1447 index++;
1448 }
1449
Hugh Dickins708e3502011-07-25 17:12:32 -07001450 while (page_nr < nr_pages)
1451 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001452
1453 if (spd.nr_pages)
1454 error = splice_to_pipe(pipe, &spd);
1455
1456 splice_shrink_spd(pipe, &spd);
1457
1458 if (error > 0) {
1459 *ppos += error;
1460 file_accessed(in);
1461 }
1462 return error;
1463}
1464
David Howells726c3342006-06-23 02:02:58 -07001465static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
David Howells726c3342006-06-23 02:02:58 -07001467 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 buf->f_type = TMPFS_MAGIC;
1470 buf->f_bsize = PAGE_CACHE_SIZE;
1471 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001472 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001474 buf->f_bavail =
1475 buf->f_bfree = sbinfo->max_blocks -
1476 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001477 }
1478 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 buf->f_files = sbinfo->max_inodes;
1480 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 }
1482 /* else leave those fields 0 like simple_statfs */
1483 return 0;
1484}
1485
1486/*
1487 * File creation. Allocate an inode, and we're done..
1488 */
1489static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001490shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001492 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 int error = -ENOSPC;
1494
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001495 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001497 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001498 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001499 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001500 if (error) {
1501 if (error != -EOPNOTSUPP) {
1502 iput(inode);
1503 return error;
1504 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001505 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001506#ifdef CONFIG_TMPFS_POSIX_ACL
1507 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001508 if (error) {
1509 iput(inode);
1510 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001511 }
Al Viro718deb62009-12-16 19:35:36 -05001512#else
1513 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001514#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 dir->i_size += BOGO_DIRENT_SIZE;
1516 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1517 d_instantiate(dentry, inode);
1518 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 }
1520 return error;
1521}
1522
Al Viro18bb1db2011-07-26 01:41:39 -04001523static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 int error;
1526
1527 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1528 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001529 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 return 0;
1531}
1532
Al Viro4acdaf22011-07-26 01:42:34 -04001533static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 struct nameidata *nd)
1535{
1536 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1537}
1538
1539/*
1540 * Link a file..
1541 */
1542static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1543{
1544 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001545 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
1547 /*
1548 * No ordinary (disk based) filesystem counts links as inodes;
1549 * but each new link needs a new dentry, pinning lowmem, and
1550 * tmpfs dentries cannot be pruned until they are unlinked.
1551 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001552 ret = shmem_reserve_inode(inode->i_sb);
1553 if (ret)
1554 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 dir->i_size += BOGO_DIRENT_SIZE;
1557 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001558 inc_nlink(inode);
Al Viro7de9c6e2010-10-23 11:11:40 -04001559 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 dget(dentry); /* Extra pinning count for the created dentry */
1561 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001562out:
1563 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
1566static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1567{
1568 struct inode *inode = dentry->d_inode;
1569
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001570 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1571 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
1573 dir->i_size -= BOGO_DIRENT_SIZE;
1574 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001575 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 dput(dentry); /* Undo the count from "create" - this does all the work */
1577 return 0;
1578}
1579
1580static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1581{
1582 if (!simple_empty(dentry))
1583 return -ENOTEMPTY;
1584
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001585 drop_nlink(dentry->d_inode);
1586 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 return shmem_unlink(dir, dentry);
1588}
1589
1590/*
1591 * The VFS layer already does all the dentry stuff for rename,
1592 * we just have to decrement the usage count for the target if
1593 * it exists so that the VFS layer correctly free's it when it
1594 * gets overwritten.
1595 */
1596static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1597{
1598 struct inode *inode = old_dentry->d_inode;
1599 int they_are_dirs = S_ISDIR(inode->i_mode);
1600
1601 if (!simple_empty(new_dentry))
1602 return -ENOTEMPTY;
1603
1604 if (new_dentry->d_inode) {
1605 (void) shmem_unlink(new_dir, new_dentry);
1606 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001607 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001609 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001610 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 }
1612
1613 old_dir->i_size -= BOGO_DIRENT_SIZE;
1614 new_dir->i_size += BOGO_DIRENT_SIZE;
1615 old_dir->i_ctime = old_dir->i_mtime =
1616 new_dir->i_ctime = new_dir->i_mtime =
1617 inode->i_ctime = CURRENT_TIME;
1618 return 0;
1619}
1620
1621static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1622{
1623 int error;
1624 int len;
1625 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07001626 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 char *kaddr;
1628 struct shmem_inode_info *info;
1629
1630 len = strlen(symname) + 1;
1631 if (len > PAGE_CACHE_SIZE)
1632 return -ENAMETOOLONG;
1633
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001634 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 if (!inode)
1636 return -ENOSPC;
1637
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001638 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001639 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001640 if (error) {
1641 if (error != -EOPNOTSUPP) {
1642 iput(inode);
1643 return error;
1644 }
1645 error = 0;
1646 }
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 info = SHMEM_I(inode);
1649 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001650 if (len <= SHORT_SYMLINK_LEN) {
1651 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1652 if (!info->symlink) {
1653 iput(inode);
1654 return -ENOMEM;
1655 }
1656 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 } else {
1658 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1659 if (error) {
1660 iput(inode);
1661 return error;
1662 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07001663 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08001665 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001667 kunmap_atomic(kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001669 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 page_cache_release(page);
1671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 dir->i_size += BOGO_DIRENT_SIZE;
1673 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1674 d_instantiate(dentry, inode);
1675 dget(dentry);
1676 return 0;
1677}
1678
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001679static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001681 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001682 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683}
1684
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001685static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686{
1687 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001688 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1689 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08001690 if (page)
1691 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001692 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693}
1694
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001695static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696{
1697 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001698 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 kunmap(page);
1700 mark_page_accessed(page);
1701 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 }
1703}
1704
Eric Parisb09e0fa2011-05-24 17:12:39 -07001705#ifdef CONFIG_TMPFS_XATTR
1706/*
1707 * Superblocks without xattr inode operations may get some security.* xattr
1708 * support from the LSM "for free". As soon as we have any other xattrs
1709 * like ACLs, we also need to implement the security.* handlers at
1710 * filesystem level, though.
1711 */
1712
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001713/*
1714 * Allocate new xattr and copy in the value; but leave the name to callers.
1715 */
1716static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
1717{
1718 struct shmem_xattr *new_xattr;
1719 size_t len;
1720
1721 /* wrap around? */
1722 len = sizeof(*new_xattr) + size;
1723 if (len <= sizeof(*new_xattr))
1724 return NULL;
1725
1726 new_xattr = kmalloc(len, GFP_KERNEL);
1727 if (!new_xattr)
1728 return NULL;
1729
1730 new_xattr->size = size;
1731 memcpy(new_xattr->value, value, size);
1732 return new_xattr;
1733}
1734
1735/*
1736 * Callback for security_inode_init_security() for acquiring xattrs.
1737 */
1738static int shmem_initxattrs(struct inode *inode,
1739 const struct xattr *xattr_array,
1740 void *fs_info)
1741{
1742 struct shmem_inode_info *info = SHMEM_I(inode);
1743 const struct xattr *xattr;
1744 struct shmem_xattr *new_xattr;
1745 size_t len;
1746
1747 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1748 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
1749 if (!new_xattr)
1750 return -ENOMEM;
1751
1752 len = strlen(xattr->name) + 1;
1753 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
1754 GFP_KERNEL);
1755 if (!new_xattr->name) {
1756 kfree(new_xattr);
1757 return -ENOMEM;
1758 }
1759
1760 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
1761 XATTR_SECURITY_PREFIX_LEN);
1762 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
1763 xattr->name, len);
1764
1765 spin_lock(&info->lock);
1766 list_add(&new_xattr->list, &info->xattr_list);
1767 spin_unlock(&info->lock);
1768 }
1769
1770 return 0;
1771}
1772
Eric Parisb09e0fa2011-05-24 17:12:39 -07001773static int shmem_xattr_get(struct dentry *dentry, const char *name,
1774 void *buffer, size_t size)
1775{
1776 struct shmem_inode_info *info;
1777 struct shmem_xattr *xattr;
1778 int ret = -ENODATA;
1779
1780 info = SHMEM_I(dentry->d_inode);
1781
1782 spin_lock(&info->lock);
1783 list_for_each_entry(xattr, &info->xattr_list, list) {
1784 if (strcmp(name, xattr->name))
1785 continue;
1786
1787 ret = xattr->size;
1788 if (buffer) {
1789 if (size < xattr->size)
1790 ret = -ERANGE;
1791 else
1792 memcpy(buffer, xattr->value, xattr->size);
1793 }
1794 break;
1795 }
1796 spin_unlock(&info->lock);
1797 return ret;
1798}
1799
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001800static int shmem_xattr_set(struct inode *inode, const char *name,
Eric Parisb09e0fa2011-05-24 17:12:39 -07001801 const void *value, size_t size, int flags)
1802{
Eric Parisb09e0fa2011-05-24 17:12:39 -07001803 struct shmem_inode_info *info = SHMEM_I(inode);
1804 struct shmem_xattr *xattr;
1805 struct shmem_xattr *new_xattr = NULL;
Eric Parisb09e0fa2011-05-24 17:12:39 -07001806 int err = 0;
1807
1808 /* value == NULL means remove */
1809 if (value) {
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001810 new_xattr = shmem_xattr_alloc(value, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001811 if (!new_xattr)
1812 return -ENOMEM;
1813
1814 new_xattr->name = kstrdup(name, GFP_KERNEL);
1815 if (!new_xattr->name) {
1816 kfree(new_xattr);
1817 return -ENOMEM;
1818 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001819 }
1820
1821 spin_lock(&info->lock);
1822 list_for_each_entry(xattr, &info->xattr_list, list) {
1823 if (!strcmp(name, xattr->name)) {
1824 if (flags & XATTR_CREATE) {
1825 xattr = new_xattr;
1826 err = -EEXIST;
1827 } else if (new_xattr) {
1828 list_replace(&xattr->list, &new_xattr->list);
1829 } else {
1830 list_del(&xattr->list);
1831 }
1832 goto out;
1833 }
1834 }
1835 if (flags & XATTR_REPLACE) {
1836 xattr = new_xattr;
1837 err = -ENODATA;
1838 } else {
1839 list_add(&new_xattr->list, &info->xattr_list);
1840 xattr = NULL;
1841 }
1842out:
1843 spin_unlock(&info->lock);
1844 if (xattr)
1845 kfree(xattr->name);
1846 kfree(xattr);
1847 return err;
1848}
1849
Eric Parisb09e0fa2011-05-24 17:12:39 -07001850static const struct xattr_handler *shmem_xattr_handlers[] = {
1851#ifdef CONFIG_TMPFS_POSIX_ACL
1852 &generic_acl_access_handler,
1853 &generic_acl_default_handler,
1854#endif
1855 NULL
1856};
1857
1858static int shmem_xattr_validate(const char *name)
1859{
1860 struct { const char *prefix; size_t len; } arr[] = {
1861 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1862 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1863 };
1864 int i;
1865
1866 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1867 size_t preflen = arr[i].len;
1868 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1869 if (!name[preflen])
1870 return -EINVAL;
1871 return 0;
1872 }
1873 }
1874 return -EOPNOTSUPP;
1875}
1876
1877static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1878 void *buffer, size_t size)
1879{
1880 int err;
1881
1882 /*
1883 * If this is a request for a synthetic attribute in the system.*
1884 * namespace use the generic infrastructure to resolve a handler
1885 * for it via sb->s_xattr.
1886 */
1887 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1888 return generic_getxattr(dentry, name, buffer, size);
1889
1890 err = shmem_xattr_validate(name);
1891 if (err)
1892 return err;
1893
1894 return shmem_xattr_get(dentry, name, buffer, size);
1895}
1896
1897static int shmem_setxattr(struct dentry *dentry, const char *name,
1898 const void *value, size_t size, int flags)
1899{
1900 int err;
1901
1902 /*
1903 * If this is a request for a synthetic attribute in the system.*
1904 * namespace use the generic infrastructure to resolve a handler
1905 * for it via sb->s_xattr.
1906 */
1907 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1908 return generic_setxattr(dentry, name, value, size, flags);
1909
1910 err = shmem_xattr_validate(name);
1911 if (err)
1912 return err;
1913
1914 if (size == 0)
1915 value = ""; /* empty EA, do not remove */
1916
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001917 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001918
1919}
1920
1921static int shmem_removexattr(struct dentry *dentry, const char *name)
1922{
1923 int err;
1924
1925 /*
1926 * If this is a request for a synthetic attribute in the system.*
1927 * namespace use the generic infrastructure to resolve a handler
1928 * for it via sb->s_xattr.
1929 */
1930 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1931 return generic_removexattr(dentry, name);
1932
1933 err = shmem_xattr_validate(name);
1934 if (err)
1935 return err;
1936
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001937 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001938}
1939
1940static bool xattr_is_trusted(const char *name)
1941{
1942 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1943}
1944
1945static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1946{
1947 bool trusted = capable(CAP_SYS_ADMIN);
1948 struct shmem_xattr *xattr;
1949 struct shmem_inode_info *info;
1950 size_t used = 0;
1951
1952 info = SHMEM_I(dentry->d_inode);
1953
1954 spin_lock(&info->lock);
1955 list_for_each_entry(xattr, &info->xattr_list, list) {
1956 size_t len;
1957
1958 /* skip "trusted." attributes for unprivileged callers */
1959 if (!trusted && xattr_is_trusted(xattr->name))
1960 continue;
1961
1962 len = strlen(xattr->name) + 1;
1963 used += len;
1964 if (buffer) {
1965 if (size < used) {
1966 used = -ERANGE;
1967 break;
1968 }
1969 memcpy(buffer, xattr->name, len);
1970 buffer += len;
1971 }
1972 }
1973 spin_unlock(&info->lock);
1974
1975 return used;
1976}
1977#endif /* CONFIG_TMPFS_XATTR */
1978
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001979static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001981 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07001982#ifdef CONFIG_TMPFS_XATTR
1983 .setxattr = shmem_setxattr,
1984 .getxattr = shmem_getxattr,
1985 .listxattr = shmem_listxattr,
1986 .removexattr = shmem_removexattr,
1987#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988};
1989
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001990static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 .readlink = generic_readlink,
1992 .follow_link = shmem_follow_link,
1993 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07001994#ifdef CONFIG_TMPFS_XATTR
1995 .setxattr = shmem_setxattr,
1996 .getxattr = shmem_getxattr,
1997 .listxattr = shmem_listxattr,
1998 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001999#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002000};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002001
David M. Grimes91828a42006-10-17 00:09:45 -07002002static struct dentry *shmem_get_parent(struct dentry *child)
2003{
2004 return ERR_PTR(-ESTALE);
2005}
2006
2007static int shmem_match(struct inode *ino, void *vfh)
2008{
2009 __u32 *fh = vfh;
2010 __u64 inum = fh[2];
2011 inum = (inum << 32) | fh[1];
2012 return ino->i_ino == inum && fh[0] == ino->i_generation;
2013}
2014
Christoph Hellwig480b1162007-10-21 16:42:13 -07002015static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2016 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002017{
David M. Grimes91828a42006-10-17 00:09:45 -07002018 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002019 struct dentry *dentry = NULL;
2020 u64 inum = fid->raw[2];
2021 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002022
Christoph Hellwig480b1162007-10-21 16:42:13 -07002023 if (fh_len < 3)
2024 return NULL;
2025
2026 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2027 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002028 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002029 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002030 iput(inode);
2031 }
2032
Christoph Hellwig480b1162007-10-21 16:42:13 -07002033 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002034}
2035
2036static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2037 int connectable)
2038{
2039 struct inode *inode = dentry->d_inode;
2040
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302041 if (*len < 3) {
2042 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07002043 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302044 }
David M. Grimes91828a42006-10-17 00:09:45 -07002045
Al Viro1d3382c2010-10-23 15:19:20 -04002046 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002047 /* Unfortunately insert_inode_hash is not idempotent,
2048 * so as we hash inodes here rather than at creation
2049 * time, we need a lock to ensure we only try
2050 * to do it once
2051 */
2052 static DEFINE_SPINLOCK(lock);
2053 spin_lock(&lock);
Al Viro1d3382c2010-10-23 15:19:20 -04002054 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002055 __insert_inode_hash(inode,
2056 inode->i_ino + inode->i_generation);
2057 spin_unlock(&lock);
2058 }
2059
2060 fh[0] = inode->i_generation;
2061 fh[1] = inode->i_ino;
2062 fh[2] = ((__u64)inode->i_ino) >> 32;
2063
2064 *len = 3;
2065 return 1;
2066}
2067
Christoph Hellwig39655162007-10-21 16:42:17 -07002068static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002069 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002070 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002071 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002072};
2073
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002074static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2075 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
2077 char *this_char, *value, *rest;
2078
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002079 while (options != NULL) {
2080 this_char = options;
2081 for (;;) {
2082 /*
2083 * NUL-terminate this option: unfortunately,
2084 * mount options form a comma-separated list,
2085 * but mpol's nodelist may also contain commas.
2086 */
2087 options = strchr(options, ',');
2088 if (options == NULL)
2089 break;
2090 options++;
2091 if (!isdigit(*options)) {
2092 options[-1] = '\0';
2093 break;
2094 }
2095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 if (!*this_char)
2097 continue;
2098 if ((value = strchr(this_char,'=')) != NULL) {
2099 *value++ = 0;
2100 } else {
2101 printk(KERN_ERR
2102 "tmpfs: No value for mount option '%s'\n",
2103 this_char);
2104 return 1;
2105 }
2106
2107 if (!strcmp(this_char,"size")) {
2108 unsigned long long size;
2109 size = memparse(value,&rest);
2110 if (*rest == '%') {
2111 size <<= PAGE_SHIFT;
2112 size *= totalram_pages;
2113 do_div(size, 100);
2114 rest++;
2115 }
2116 if (*rest)
2117 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002118 sbinfo->max_blocks =
2119 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002121 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 if (*rest)
2123 goto bad_val;
2124 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002125 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 if (*rest)
2127 goto bad_val;
2128 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002129 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002131 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (*rest)
2133 goto bad_val;
2134 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002135 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002137 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (*rest)
2139 goto bad_val;
2140 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002141 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002143 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (*rest)
2145 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002146 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002147 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002148 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 } else {
2150 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2151 this_char);
2152 return 1;
2153 }
2154 }
2155 return 0;
2156
2157bad_val:
2158 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2159 value, this_char);
2160 return 1;
2161
2162}
2163
2164static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2165{
2166 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002167 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002168 unsigned long inodes;
2169 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002171 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002172 return error;
2173
2174 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002175 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002176 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002177 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002178 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002179 goto out;
2180 /*
Hugh Dickins54af6042011-08-03 16:21:24 -07002181 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002182 * but we must separately disallow unlimited->limited, because
2183 * in that case we have no record of how much is already in use.
2184 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002185 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002186 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002187 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002188 goto out;
2189
2190 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002191 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002192 sbinfo->max_inodes = config.max_inodes;
2193 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002194
2195 mpol_put(sbinfo->mpol);
2196 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002197out:
2198 spin_unlock(&sbinfo->stat_lock);
2199 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002201
Al Viro34c80b12011-12-08 21:32:45 -05002202static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002203{
Al Viro34c80b12011-12-08 21:32:45 -05002204 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002205
2206 if (sbinfo->max_blocks != shmem_default_max_blocks())
2207 seq_printf(seq, ",size=%luk",
2208 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2209 if (sbinfo->max_inodes != shmem_default_max_inodes())
2210 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2211 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002212 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002213 if (sbinfo->uid != 0)
2214 seq_printf(seq, ",uid=%u", sbinfo->uid);
2215 if (sbinfo->gid != 0)
2216 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002217 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002218 return 0;
2219}
2220#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
2222static void shmem_put_super(struct super_block *sb)
2223{
Hugh Dickins602586a2010-08-17 15:23:56 -07002224 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2225
2226 percpu_counter_destroy(&sbinfo->used_blocks);
2227 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 sb->s_fs_info = NULL;
2229}
2230
Kay Sievers2b2af542009-04-30 15:23:42 +02002231int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232{
2233 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002234 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002235 int err = -ENOMEM;
2236
2237 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002238 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002239 L1_CACHE_BYTES), GFP_KERNEL);
2240 if (!sbinfo)
2241 return -ENOMEM;
2242
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002243 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002244 sbinfo->uid = current_fsuid();
2245 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002246 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002248#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 /*
2250 * Per default we only allow half of the physical ram per
2251 * tmpfs instance, limiting inodes to one per page of lowmem;
2252 * but the internal instance is left unlimited.
2253 */
2254 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002255 sbinfo->max_blocks = shmem_default_max_blocks();
2256 sbinfo->max_inodes = shmem_default_max_inodes();
2257 if (shmem_parse_options(data, sbinfo, false)) {
2258 err = -EINVAL;
2259 goto failed;
2260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 }
David M. Grimes91828a42006-10-17 00:09:45 -07002262 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263#else
2264 sb->s_flags |= MS_NOUSER;
2265#endif
2266
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002267 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002268 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2269 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002270 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002271
Hugh Dickins285b2c42011-08-03 16:21:20 -07002272 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 sb->s_blocksize = PAGE_CACHE_SIZE;
2274 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2275 sb->s_magic = TMPFS_MAGIC;
2276 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002277 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002278#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002279 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002280#endif
2281#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002282 sb->s_flags |= MS_POSIXACL;
2283#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002284
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002285 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 if (!inode)
2287 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002288 inode->i_uid = sbinfo->uid;
2289 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002290 sb->s_root = d_make_root(inode);
2291 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002292 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 return 0;
2294
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295failed:
2296 shmem_put_super(sb);
2297 return err;
2298}
2299
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002300static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
2302static struct inode *shmem_alloc_inode(struct super_block *sb)
2303{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002304 struct shmem_inode_info *info;
2305 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2306 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002308 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309}
2310
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002311static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e32011-01-07 17:49:49 +11002312{
2313 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e32011-01-07 17:49:49 +11002314 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2315}
2316
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317static void shmem_destroy_inode(struct inode *inode)
2318{
Al Viro09208d12011-07-26 03:15:03 -04002319 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002321 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322}
2323
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002324static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002326 struct shmem_inode_info *info = foo;
2327 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328}
2329
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002330static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
2332 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2333 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002334 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 return 0;
2336}
2337
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002338static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002340 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341}
2342
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002343static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002345 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002347 .write_begin = shmem_write_begin,
2348 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002350 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002351 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352};
2353
Helge Deller15ad7cd2006-12-06 20:40:36 -08002354static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 .mmap = shmem_mmap,
2356#ifdef CONFIG_TMPFS
2357 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002358 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002359 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002360 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002361 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002362 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002363 .splice_read = shmem_file_splice_read,
Hugh Dickinsae976412007-06-04 10:00:39 +02002364 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365#endif
2366};
2367
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002368static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002369 .setattr = shmem_setattr,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002370 .truncate_range = shmem_truncate_range,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002371#ifdef CONFIG_TMPFS_XATTR
2372 .setxattr = shmem_setxattr,
2373 .getxattr = shmem_getxattr,
2374 .listxattr = shmem_listxattr,
2375 .removexattr = shmem_removexattr,
2376#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377};
2378
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002379static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380#ifdef CONFIG_TMPFS
2381 .create = shmem_create,
2382 .lookup = simple_lookup,
2383 .link = shmem_link,
2384 .unlink = shmem_unlink,
2385 .symlink = shmem_symlink,
2386 .mkdir = shmem_mkdir,
2387 .rmdir = shmem_rmdir,
2388 .mknod = shmem_mknod,
2389 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002391#ifdef CONFIG_TMPFS_XATTR
2392 .setxattr = shmem_setxattr,
2393 .getxattr = shmem_getxattr,
2394 .listxattr = shmem_listxattr,
2395 .removexattr = shmem_removexattr,
2396#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002397#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002398 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002399#endif
2400};
2401
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002402static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002403#ifdef CONFIG_TMPFS_XATTR
2404 .setxattr = shmem_setxattr,
2405 .getxattr = shmem_getxattr,
2406 .listxattr = shmem_listxattr,
2407 .removexattr = shmem_removexattr,
2408#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002409#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002410 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002411#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412};
2413
Hugh Dickins759b9772007-03-05 00:30:28 -08002414static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 .alloc_inode = shmem_alloc_inode,
2416 .destroy_inode = shmem_destroy_inode,
2417#ifdef CONFIG_TMPFS
2418 .statfs = shmem_statfs,
2419 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002420 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421#endif
Al Viro1f895f72010-06-05 19:10:41 -04002422 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 .drop_inode = generic_delete_inode,
2424 .put_super = shmem_put_super,
2425};
2426
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002427static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002428 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429#ifdef CONFIG_NUMA
2430 .set_policy = shmem_set_policy,
2431 .get_policy = shmem_get_policy,
2432#endif
2433};
2434
Al Viro3c26ff62010-07-25 11:46:36 +04002435static struct dentry *shmem_mount(struct file_system_type *fs_type,
2436 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437{
Al Viro3c26ff62010-07-25 11:46:36 +04002438 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439}
2440
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002441static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 .owner = THIS_MODULE,
2443 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002444 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 .kill_sb = kill_litter_super,
2446};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002448int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449{
2450 int error;
2451
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002452 error = bdi_init(&shmem_backing_dev_info);
2453 if (error)
2454 goto out4;
2455
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002456 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 if (error)
2458 goto out3;
2459
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002460 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (error) {
2462 printk(KERN_ERR "Could not register tmpfs\n");
2463 goto out2;
2464 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002465
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002466 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2467 shmem_fs_type.name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 if (IS_ERR(shm_mnt)) {
2469 error = PTR_ERR(shm_mnt);
2470 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2471 goto out1;
2472 }
2473 return 0;
2474
2475out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002476 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002478 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002480 bdi_destroy(&shmem_backing_dev_info);
2481out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 shm_mnt = ERR_PTR(error);
2483 return error;
2484}
Matt Mackall853ac432009-01-06 14:40:20 -08002485
2486#else /* !CONFIG_SHMEM */
2487
2488/*
2489 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2490 *
2491 * This is intended for small system where the benefits of the full
2492 * shmem code (swap-backed and resource-limited) are outweighed by
2493 * their complexity. On systems without swap this code should be
2494 * effectively equivalent, but much lighter weight.
2495 */
2496
2497#include <linux/ramfs.h>
2498
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002499static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002500 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002501 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002502 .kill_sb = kill_litter_super,
2503};
2504
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002505int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002506{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002507 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002508
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002509 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002510 BUG_ON(IS_ERR(shm_mnt));
2511
2512 return 0;
2513}
2514
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002515int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002516{
2517 return 0;
2518}
2519
Hugh Dickins3f96b792009-09-21 17:03:37 -07002520int shmem_lock(struct file *file, int lock, struct user_struct *user)
2521{
2522 return 0;
2523}
2524
Hugh Dickins24513262012-01-20 14:34:21 -08002525void shmem_unlock_mapping(struct address_space *mapping)
2526{
2527}
2528
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002529void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002530{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002531 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002532}
2533EXPORT_SYMBOL_GPL(shmem_truncate_range);
2534
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002535#define shmem_vm_ops generic_file_vm_ops
2536#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002537#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002538#define shmem_acct_size(flags, size) 0
2539#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002540
2541#endif /* CONFIG_SHMEM */
2542
2543/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
Randy Dunlap46711812008-03-19 17:00:41 -07002545/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 * @name: name for dentry (to be seen in /proc/<pid>/maps
2548 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002549 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002551struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552{
2553 int error;
2554 struct file *file;
2555 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002556 struct path path;
2557 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 struct qstr this;
2559
2560 if (IS_ERR(shm_mnt))
2561 return (void *)shm_mnt;
2562
Hugh Dickins285b2c42011-08-03 16:21:20 -07002563 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 return ERR_PTR(-EINVAL);
2565
2566 if (shmem_acct_size(flags, size))
2567 return ERR_PTR(-ENOMEM);
2568
2569 error = -ENOMEM;
2570 this.name = name;
2571 this.len = strlen(name);
2572 this.hash = 0; /* will go */
2573 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002574 path.dentry = d_alloc(root, &this);
2575 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002577 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002580 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002582 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583
Al Viro2c48b9c2009-08-09 00:52:35 +04002584 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002586 clear_nlink(inode); /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002587#ifndef CONFIG_MMU
2588 error = ramfs_nommu_expand_for_mapping(inode, size);
2589 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002590 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002591#endif
Al Viro4b42af82009-08-05 18:25:56 +04002592
2593 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002594 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002595 &shmem_file_operations);
2596 if (!file)
2597 goto put_dentry;
2598
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 return file;
2600
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002602 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603put_memory:
2604 shmem_unacct_size(flags, size);
2605 return ERR_PTR(error);
2606}
Keith Packard395e0dd2008-06-20 00:08:06 -07002607EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608
John Stultzd92aa582011-12-15 18:51:04 -08002609void shmem_set_file(struct vm_area_struct *vma, struct file *file)
2610{
2611 if (vma->vm_file)
2612 fput(vma->vm_file);
2613 vma->vm_file = file;
2614 vma->vm_ops = &shmem_vm_ops;
2615 vma->vm_flags |= VM_CAN_NONLINEAR;
2616}
2617
Randy Dunlap46711812008-03-19 17:00:41 -07002618/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2621 */
2622int shmem_zero_setup(struct vm_area_struct *vma)
2623{
2624 struct file *file;
2625 loff_t size = vma->vm_end - vma->vm_start;
2626
2627 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2628 if (IS_ERR(file))
2629 return PTR_ERR(file);
2630
John Stultzd92aa582011-12-15 18:51:04 -08002631 shmem_set_file(vma, file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 return 0;
2633}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002634
2635/**
2636 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2637 * @mapping: the page's address_space
2638 * @index: the page index
2639 * @gfp: the page allocator flags to use if allocating
2640 *
2641 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2642 * with any new page allocations done using the specified allocation flags.
2643 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2644 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2645 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2646 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07002647 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2648 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002649 */
2650struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2651 pgoff_t index, gfp_t gfp)
2652{
Hugh Dickins68da9f02011-07-25 17:12:34 -07002653#ifdef CONFIG_SHMEM
2654 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002655 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07002656 int error;
2657
2658 BUG_ON(mapping->a_ops != &shmem_aops);
2659 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2660 if (error)
2661 page = ERR_PTR(error);
2662 else
2663 unlock_page(page);
2664 return page;
2665#else
2666 /*
2667 * The tiny !SHMEM case uses ramfs without swap
2668 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002669 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002670#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002671}
2672EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);