blob: 32f6763f16fb82ad7068c4d73479a96583f4ef15 [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>
31#include <linux/module.h>
32#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) ?
130 0 : security_vm_enough_memory_kern(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) ?
148 security_vm_enough_memory_kern(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 */
382static void shmem_pagevec_release(struct pagevec *pvec)
383{
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;
392 pagevec_release(pvec);
393}
394
395/*
396 * Remove range of pages and swap entries from radix tree, and free them.
397 */
Hugh Dickins285b2c42011-08-03 16:21:20 -0700398void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700400 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700402 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700403 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700404 pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700405 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700406 pgoff_t indices[PAGEVEC_SIZE];
407 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700408 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700409 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700411 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
412
413 pagevec_init(&pvec, 0);
414 index = start;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700415 while (index <= end) {
416 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
417 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
418 pvec.pages, indices);
419 if (!pvec.nr)
420 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700421 mem_cgroup_uncharge_start();
422 for (i = 0; i < pagevec_count(&pvec); i++) {
423 struct page *page = pvec.pages[i];
424
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700425 index = indices[i];
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700426 if (index > end)
427 break;
428
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700429 if (radix_tree_exceptional_entry(page)) {
430 nr_swaps_freed += !shmem_free_swap(mapping,
431 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700432 continue;
433 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700434
435 if (!trylock_page(page))
436 continue;
437 if (page->mapping == mapping) {
438 VM_BUG_ON(PageWriteback(page));
439 truncate_inode_page(mapping, page);
440 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700441 unlock_page(page);
442 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700443 shmem_pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700444 mem_cgroup_uncharge_end();
445 cond_resched();
446 index++;
447 }
448
449 if (partial) {
450 struct page *page = NULL;
451 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
452 if (page) {
453 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
454 set_page_dirty(page);
455 unlock_page(page);
456 page_cache_release(page);
457 }
458 }
459
460 index = start;
461 for ( ; ; ) {
462 cond_resched();
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700463 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
464 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
465 pvec.pages, indices);
466 if (!pvec.nr) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700467 if (index == start)
468 break;
469 index = start;
470 continue;
471 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700472 if (index == start && indices[0] > end) {
473 shmem_pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700474 break;
475 }
476 mem_cgroup_uncharge_start();
477 for (i = 0; i < pagevec_count(&pvec); i++) {
478 struct page *page = pvec.pages[i];
479
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700480 index = indices[i];
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700481 if (index > end)
482 break;
483
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700484 if (radix_tree_exceptional_entry(page)) {
485 nr_swaps_freed += !shmem_free_swap(mapping,
486 index, page);
487 continue;
488 }
489
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700490 lock_page(page);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700491 if (page->mapping == mapping) {
492 VM_BUG_ON(PageWriteback(page));
493 truncate_inode_page(mapping, page);
494 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700495 unlock_page(page);
496 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700497 shmem_pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700498 mem_cgroup_uncharge_end();
499 index++;
500 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700503 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 shmem_recalc_inode(inode);
505 spin_unlock(&info->lock);
506
Hugh Dickins285b2c42011-08-03 16:21:20 -0700507 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700509EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Hugh Dickins94c1e622011-06-27 16:18:03 -0700511static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512{
513 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 int error;
515
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200516 error = inode_change_ok(inode, attr);
517 if (error)
518 return error;
519
Hugh Dickins94c1e622011-06-27 16:18:03 -0700520 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
521 loff_t oldsize = inode->i_size;
522 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000523
Hugh Dickins94c1e622011-06-27 16:18:03 -0700524 if (newsize != oldsize) {
525 i_size_write(inode, newsize);
526 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
527 }
528 if (newsize < oldsize) {
529 loff_t holebegin = round_up(newsize, PAGE_SIZE);
530 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
531 shmem_truncate_range(inode, newsize, (loff_t)-1);
532 /* unmap again to remove racily COWed private pages */
533 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200537 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700538#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200539 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100540 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700541#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 return error;
543}
544
Al Viro1f895f72010-06-05 19:10:41 -0400545static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 struct shmem_inode_info *info = SHMEM_I(inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700548 struct shmem_xattr *xattr, *nxattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000550 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 shmem_unacct_size(info->flags, inode->i_size);
552 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000553 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800555 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800557 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700559 } else
560 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700561
562 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
563 kfree(xattr->name);
564 kfree(xattr);
565 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700566 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800567 shmem_free_inode(inode->i_sb);
Al Viro1f895f72010-06-05 19:10:41 -0400568 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569}
570
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700571/*
572 * If swap found in inode, free it and move page from swapcache to filecache.
573 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700574static int shmem_unuse_inode(struct shmem_inode_info *info,
575 swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700577 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700578 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700579 pgoff_t index;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800580 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700582 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700583 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700584 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700585 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800586
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800587 /*
588 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400589 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800590 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700591 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800592 */
593 if (shmem_swaplist.next != &info->swaplist)
594 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800595
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800596 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700597 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
598 * but also to hold up shmem_evict_inode(): so inode cannot be freed
599 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800600 */
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700601 error = shmem_add_to_page_cache(page, mapping, index,
602 GFP_NOWAIT, radswap);
Hugh Dickins778dd892011-05-11 15:13:37 -0700603 /* which does mem_cgroup_uncharge_cache_page on error */
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700604
Hugh Dickins48f170f2011-07-25 17:12:37 -0700605 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700606 /*
607 * Truncation and eviction use free_swap_and_cache(), which
608 * only does trylock page: if we raced, best clean up here.
609 */
Hugh Dickins73b12622008-02-04 22:28:50 -0800610 delete_from_swap_cache(page);
611 set_page_dirty(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700612 if (!error) {
613 spin_lock(&info->lock);
614 info->swapped--;
615 spin_unlock(&info->lock);
616 swap_free(swap);
617 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800618 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800620 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622
623/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700624 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700626int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700628 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 struct shmem_inode_info *info;
630 int found = 0;
Hugh Dickins778dd892011-05-11 15:13:37 -0700631 int error;
632
633 /*
634 * Charge page using GFP_KERNEL while we can wait, before taking
635 * the shmem_swaplist_mutex which might hold up shmem_writepage().
636 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700637 */
638 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
639 if (error)
640 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700641 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800643 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700644 list_for_each_safe(this, next, &shmem_swaplist) {
645 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700646 if (info->swapped)
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700647 found = shmem_unuse_inode(info, swap, page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700648 else
649 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800650 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800651 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700652 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800654 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700655
Hugh Dickins778dd892011-05-11 15:13:37 -0700656 if (!found)
657 mem_cgroup_uncharge_cache_page(page);
658 if (found < 0)
659 error = found;
660out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800661 unlock_page(page);
662 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700663 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
666/*
667 * Move the page from the page cache to the swap cache.
668 */
669static int shmem_writepage(struct page *page, struct writeback_control *wbc)
670{
671 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700674 swp_entry_t swap;
675 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 mapping = page->mapping;
679 index = page->index;
680 inode = mapping->host;
681 info = SHMEM_I(inode);
682 if (info->flags & VM_LOCKED)
683 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800684 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 goto redirty;
686
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800687 /*
688 * shmem_backing_dev_info's capabilities prevent regular writeback or
689 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700690 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800691 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700692 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800693 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700694 if (!wbc->for_reclaim) {
695 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
696 goto redirty;
697 }
698 swap = get_swap_page();
699 if (!swap.val)
700 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800701
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700702 /*
703 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700704 * if it's not already there. Do it now before the page is
705 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700706 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700707 * we've incremented swapped, because shmem_unuse_inode() will
708 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700709 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700710 mutex_lock(&shmem_swaplist_mutex);
711 if (list_empty(&info->swaplist))
712 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700713
Hugh Dickins48f170f2011-07-25 17:12:37 -0700714 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800715 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700716 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
717
718 spin_lock(&info->lock);
719 info->swapped++;
720 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700721 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700722
723 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800724 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700725 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 return 0;
727 }
728
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700729 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700730 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731redirty:
732 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800733 if (wbc->for_reclaim)
734 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
735 unlock_page(page);
736 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800740#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700741static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800742{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700743 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800744
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700745 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700746 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800747
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700748 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800749
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700750 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800751}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700752
753static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
754{
755 struct mempolicy *mpol = NULL;
756 if (sbinfo->mpol) {
757 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
758 mpol = sbinfo->mpol;
759 mpol_get(mpol);
760 spin_unlock(&sbinfo->stat_lock);
761 }
762 return mpol;
763}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800764#endif /* CONFIG_TMPFS */
765
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700766static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
767 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700769 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 struct vm_area_struct pvma;
771
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700772 spol = mpol_cond_copy(&mpol,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700773 mpol_shared_policy_lookup(&info->policy, index));
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800776 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700777 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800778 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700779 pvma.vm_policy = spol;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700780 return swapin_readahead(swap, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
Hugh Dickins02098fe2008-02-04 22:28:42 -0800783static struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700784 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
786 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800788 /* Create a pseudo vma that just contains the policy */
789 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700790 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800791 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700792 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700793
794 /*
795 * alloc_page_vma() will drop the shared policy reference
796 */
797 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800799#else /* !CONFIG_NUMA */
800#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700801static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800802{
803}
804#endif /* CONFIG_TMPFS */
805
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700806static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
807 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700809 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
Hugh Dickins02098fe2008-02-04 22:28:42 -0800812static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700813 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000815 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800817#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700819#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
820static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
821{
822 return NULL;
823}
824#endif
825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826/*
Hugh Dickins68da9f02011-07-25 17:12:34 -0700827 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 *
829 * If we allocate a new one we do not mark it dirty. That's up to the
830 * vm. If we swap it in we mark it dirty since we also free the swap
831 * entry since a page cannot live in both the swap and page cache
832 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700833static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -0700834 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
836 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af6042011-08-03 16:21:24 -0700837 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -0700839 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 swp_entry_t swap;
841 int error;
Hugh Dickins54af6042011-08-03 16:21:24 -0700842 int once = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700844 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846repeat:
Hugh Dickins54af6042011-08-03 16:21:24 -0700847 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700848 page = find_lock_page(mapping, index);
Hugh Dickins54af6042011-08-03 16:21:24 -0700849 if (radix_tree_exceptional_entry(page)) {
850 swap = radix_to_swp_entry(page);
851 page = NULL;
852 }
853
854 if (sgp != SGP_WRITE &&
855 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
856 error = -EINVAL;
857 goto failed;
858 }
859
860 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800861 /*
Hugh Dickins27ab7002011-07-25 17:12:36 -0700862 * Once we can get the page lock, it must be uptodate:
863 * if there were an error in reading back from swap,
864 * the page would not be inserted into the filecache.
Hugh Dickinsb409f9f2008-02-04 22:28:54 -0800865 */
Hugh Dickins54af6042011-08-03 16:21:24 -0700866 BUG_ON(page && !PageUptodate(page));
867 *pagep = page;
868 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -0700869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 /*
Hugh Dickins54af6042011-08-03 16:21:24 -0700872 * Fast cache lookup did not find it:
873 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
Hugh Dickins54af6042011-08-03 16:21:24 -0700875 info = SHMEM_I(inode);
876 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (swap.val) {
879 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -0700880 page = lookup_swap_cache(swap);
881 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -0700882 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -0700883 if (fault_type)
884 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700885 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700886 if (!page) {
Hugh Dickins54af6042011-08-03 16:21:24 -0700887 error = -ENOMEM;
888 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 }
891
892 /* We have to do this with page locked to prevent races */
Hugh Dickins54af6042011-08-03 16:21:24 -0700893 lock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700894 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 error = -EIO;
Hugh Dickins54af6042011-08-03 16:21:24 -0700896 goto failed;
897 }
898 wait_on_page_writeback(page);
899
900 /* Someone may have already done it for us */
901 if (page->mapping) {
902 if (page->mapping == mapping &&
903 page->index == index)
904 goto done;
905 error = -EEXIST;
906 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908
Hugh Dickinsaa3b1892011-08-03 16:21:24 -0700909 error = mem_cgroup_cache_charge(page, current->mm,
910 gfp & GFP_RECLAIM_MASK);
911 if (!error)
912 error = shmem_add_to_page_cache(page, mapping, index,
913 gfp, swp_to_radix_entry(swap));
Hugh Dickins54af6042011-08-03 16:21:24 -0700914 if (error)
915 goto failed;
916
917 spin_lock(&info->lock);
918 info->swapped--;
919 shmem_recalc_inode(inode);
920 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700921
Hugh Dickins27ab7002011-07-25 17:12:36 -0700922 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -0700923 set_page_dirty(page);
924 swap_free(swap);
925
Hugh Dickins54af6042011-08-03 16:21:24 -0700926 } else {
927 if (shmem_acct_block(info->flags)) {
928 error = -ENOSPC;
929 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700931 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -0700932 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af6042011-08-03 16:21:24 -0700933 sbinfo->max_blocks) >= 0) {
934 error = -ENOSPC;
935 goto unacct;
936 }
Tim Chen7e496292010-08-09 17:19:05 -0700937 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -0700938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Hugh Dickins54af6042011-08-03 16:21:24 -0700940 page = shmem_alloc_page(gfp, info, index);
941 if (!page) {
942 error = -ENOMEM;
943 goto decused;
944 }
945
946 SetPageSwapBacked(page);
947 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -0700948 error = mem_cgroup_cache_charge(page, current->mm,
949 gfp & GFP_RECLAIM_MASK);
950 if (!error)
951 error = shmem_add_to_page_cache(page, mapping, index,
952 gfp, NULL);
Hugh Dickins54af6042011-08-03 16:21:24 -0700953 if (error)
954 goto decused;
955 lru_cache_add_anon(page);
956
957 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 info->alloced++;
Hugh Dickins54af6042011-08-03 16:21:24 -0700959 inode->i_blocks += BLOCKS_PER_PAGE;
960 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 spin_unlock(&info->lock);
Hugh Dickins54af6042011-08-03 16:21:24 -0700962
Hugh Dickins27ab7002011-07-25 17:12:36 -0700963 clear_highpage(page);
964 flush_dcache_page(page);
965 SetPageUptodate(page);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800966 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -0700967 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -0700968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969done:
Hugh Dickins54af6042011-08-03 16:21:24 -0700970 /* Perhaps the file has been truncated since we checked */
971 if (sgp != SGP_WRITE &&
972 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
973 error = -EINVAL;
974 goto trunc;
Shaohua Liff36b802010-08-09 17:19:06 -0700975 }
Hugh Dickins54af6042011-08-03 16:21:24 -0700976 *pagep = page;
977 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -0700978
Nick Piggind0217ac2007-07-19 01:47:03 -0700979 /*
Hugh Dickins54af6042011-08-03 16:21:24 -0700980 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 */
Hugh Dickins54af6042011-08-03 16:21:24 -0700982trunc:
983 ClearPageDirty(page);
984 delete_from_page_cache(page);
985 spin_lock(&info->lock);
986 info->alloced--;
987 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 spin_unlock(&info->lock);
Hugh Dickins54af6042011-08-03 16:21:24 -0700989decused:
990 if (sbinfo->max_blocks)
991 percpu_counter_add(&sbinfo->used_blocks, -1);
992unacct:
993 shmem_unacct_blocks(info->flags, 1);
994failed:
995 if (swap.val && error != -EINVAL) {
996 struct page *test = find_get_page(mapping, index);
997 if (test && !radix_tree_exceptional_entry(test))
998 page_cache_release(test);
999 /* Have another try if the entry has changed */
1000 if (test != swp_to_radix_entry(swap))
1001 error = -EEXIST;
1002 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001003 if (page) {
Hugh Dickins54af6042011-08-03 16:21:24 -07001004 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001005 page_cache_release(page);
Hugh Dickins54af6042011-08-03 16:21:24 -07001006 }
1007 if (error == -ENOSPC && !once++) {
1008 info = SHMEM_I(inode);
1009 spin_lock(&info->lock);
1010 shmem_recalc_inode(inode);
1011 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001012 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001013 }
Hugh Dickins54af6042011-08-03 16:21:24 -07001014 if (error == -EEXIST)
1015 goto repeat;
1016 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
1019static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1020{
1021 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1022 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001023 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1026 if (error)
1027 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001028
Ying Han456f9982011-05-26 16:25:38 -07001029 if (ret & VM_FAULT_MAJOR) {
1030 count_vm_event(PGMAJFAULT);
1031 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1032 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001033 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034}
1035
1036#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001037static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001039 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1040 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041}
1042
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001043static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1044 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001046 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1047 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001049 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1050 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052#endif
1053
1054int shmem_lock(struct file *file, int lock, struct user_struct *user)
1055{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001056 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 struct shmem_inode_info *info = SHMEM_I(inode);
1058 int retval = -ENOMEM;
1059
1060 spin_lock(&info->lock);
1061 if (lock && !(info->flags & VM_LOCKED)) {
1062 if (!user_shm_lock(inode->i_size, user))
1063 goto out_nomem;
1064 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001065 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 }
1067 if (!lock && (info->flags & VM_LOCKED) && user) {
1068 user_shm_unlock(inode->i_size, user);
1069 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001070 mapping_clear_unevictable(file->f_mapping);
1071 scan_mapping_unevictable_pages(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001074
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075out_nomem:
1076 spin_unlock(&info->lock);
1077 return retval;
1078}
1079
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001080static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
1082 file_accessed(file);
1083 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001084 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 return 0;
1086}
1087
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001088static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1089 int mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
1091 struct inode *inode;
1092 struct shmem_inode_info *info;
1093 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1094
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001095 if (shmem_reserve_inode(sb))
1096 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 inode = new_inode(sb);
1099 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001100 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001101 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1104 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001105 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 info = SHMEM_I(inode);
1107 memset(info, 0, (char *)inode - (char *)info);
1108 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001109 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 INIT_LIST_HEAD(&info->swaplist);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001111 INIT_LIST_HEAD(&info->xattr_list);
Al Viro72c04902009-06-24 16:58:48 -04001112 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 switch (mode & S_IFMT) {
1115 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001116 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 init_special_inode(inode, mode, dev);
1118 break;
1119 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001120 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 inode->i_op = &shmem_inode_operations;
1122 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001123 mpol_shared_policy_init(&info->policy,
1124 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 break;
1126 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001127 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 /* Some things misbehave if size == 0 on a directory */
1129 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1130 inode->i_op = &shmem_dir_inode_operations;
1131 inode->i_fop = &simple_dir_operations;
1132 break;
1133 case S_IFLNK:
1134 /*
1135 * Must not load anything in the rbtree,
1136 * mpol_free_shared_policy will not be called.
1137 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001138 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 break;
1140 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001141 } else
1142 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 return inode;
1144}
1145
1146#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001147static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001148static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001151shmem_write_begin(struct file *file, struct address_space *mapping,
1152 loff_t pos, unsigned len, unsigned flags,
1153 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154{
Nick Piggin800d15a2007-10-16 01:25:03 -07001155 struct inode *inode = mapping->host;
1156 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001157 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1158}
1159
1160static int
1161shmem_write_end(struct file *file, struct address_space *mapping,
1162 loff_t pos, unsigned len, unsigned copied,
1163 struct page *page, void *fsdata)
1164{
1165 struct inode *inode = mapping->host;
1166
Hugh Dickinsd3602442008-02-04 22:28:44 -08001167 if (pos + copied > inode->i_size)
1168 i_size_write(inode, pos + copied);
1169
Nick Piggin800d15a2007-10-16 01:25:03 -07001170 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001171 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001172 page_cache_release(page);
1173
Nick Piggin800d15a2007-10-16 01:25:03 -07001174 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
1176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1178{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001179 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001181 pgoff_t index;
1182 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001183 enum sgp_type sgp = SGP_READ;
1184
1185 /*
1186 * Might this read be for a stacking filesystem? Then when reading
1187 * holes of a sparse file, we actually need to allocate those pages,
1188 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1189 */
1190 if (segment_eq(get_fs(), KERNEL_DS))
1191 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193 index = *ppos >> PAGE_CACHE_SHIFT;
1194 offset = *ppos & ~PAGE_CACHE_MASK;
1195
1196 for (;;) {
1197 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001198 pgoff_t end_index;
1199 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 loff_t i_size = i_size_read(inode);
1201
1202 end_index = i_size >> PAGE_CACHE_SHIFT;
1203 if (index > end_index)
1204 break;
1205 if (index == end_index) {
1206 nr = i_size & ~PAGE_CACHE_MASK;
1207 if (nr <= offset)
1208 break;
1209 }
1210
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001211 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (desc->error) {
1213 if (desc->error == -EINVAL)
1214 desc->error = 0;
1215 break;
1216 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001217 if (page)
1218 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 /*
1221 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001222 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 */
1224 nr = PAGE_CACHE_SIZE;
1225 i_size = i_size_read(inode);
1226 end_index = i_size >> PAGE_CACHE_SHIFT;
1227 if (index == end_index) {
1228 nr = i_size & ~PAGE_CACHE_MASK;
1229 if (nr <= offset) {
1230 if (page)
1231 page_cache_release(page);
1232 break;
1233 }
1234 }
1235 nr -= offset;
1236
1237 if (page) {
1238 /*
1239 * If users can be writing to this page using arbitrary
1240 * virtual addresses, take care about potential aliasing
1241 * before reading the page on the kernel side.
1242 */
1243 if (mapping_writably_mapped(mapping))
1244 flush_dcache_page(page);
1245 /*
1246 * Mark the page accessed if we read the beginning.
1247 */
1248 if (!offset)
1249 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001250 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001252 page_cache_get(page);
1253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255 /*
1256 * Ok, we have the page, and it's up-to-date, so
1257 * now we can copy it to user space...
1258 *
1259 * The actor routine returns how many bytes were actually used..
1260 * NOTE! This may not be the same as how much of a user buffer
1261 * we filled up (we may be padding etc), so we can only update
1262 * "pos" here (the actor routine has to update the user buffer
1263 * pointers and the remaining count).
1264 */
1265 ret = actor(desc, page, offset, nr);
1266 offset += ret;
1267 index += offset >> PAGE_CACHE_SHIFT;
1268 offset &= ~PAGE_CACHE_MASK;
1269
1270 page_cache_release(page);
1271 if (ret != nr || !desc->count)
1272 break;
1273
1274 cond_resched();
1275 }
1276
1277 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1278 file_accessed(filp);
1279}
1280
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001281static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1282 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001284 struct file *filp = iocb->ki_filp;
1285 ssize_t retval;
1286 unsigned long seg;
1287 size_t count;
1288 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001290 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1291 if (retval)
1292 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001294 for (seg = 0; seg < nr_segs; seg++) {
1295 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001297 desc.written = 0;
1298 desc.arg.buf = iov[seg].iov_base;
1299 desc.count = iov[seg].iov_len;
1300 if (desc.count == 0)
1301 continue;
1302 desc.error = 0;
1303 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1304 retval += desc.written;
1305 if (desc.error) {
1306 retval = retval ?: desc.error;
1307 break;
1308 }
1309 if (desc.count > 0)
1310 break;
1311 }
1312 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
Hugh Dickins708e3502011-07-25 17:12:32 -07001315static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1316 struct pipe_inode_info *pipe, size_t len,
1317 unsigned int flags)
1318{
1319 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001320 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001321 unsigned int loff, nr_pages, req_pages;
1322 struct page *pages[PIPE_DEF_BUFFERS];
1323 struct partial_page partial[PIPE_DEF_BUFFERS];
1324 struct page *page;
1325 pgoff_t index, end_index;
1326 loff_t isize, left;
1327 int error, page_nr;
1328 struct splice_pipe_desc spd = {
1329 .pages = pages,
1330 .partial = partial,
1331 .flags = flags,
1332 .ops = &page_cache_pipe_buf_ops,
1333 .spd_release = spd_release_page,
1334 };
1335
Hugh Dickins71f0e072011-07-25 17:12:33 -07001336 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001337 if (unlikely(*ppos >= isize))
1338 return 0;
1339
1340 left = isize - *ppos;
1341 if (unlikely(left < len))
1342 len = left;
1343
1344 if (splice_grow_spd(pipe, &spd))
1345 return -ENOMEM;
1346
1347 index = *ppos >> PAGE_CACHE_SHIFT;
1348 loff = *ppos & ~PAGE_CACHE_MASK;
1349 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1350 nr_pages = min(req_pages, pipe->buffers);
1351
Hugh Dickins708e3502011-07-25 17:12:32 -07001352 spd.nr_pages = find_get_pages_contig(mapping, index,
1353 nr_pages, spd.pages);
1354 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001355 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001356
Hugh Dickins708e3502011-07-25 17:12:32 -07001357 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001358 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1359 if (error)
1360 break;
1361 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001362 spd.pages[spd.nr_pages++] = page;
1363 index++;
1364 }
1365
Hugh Dickins708e3502011-07-25 17:12:32 -07001366 index = *ppos >> PAGE_CACHE_SHIFT;
1367 nr_pages = spd.nr_pages;
1368 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001369
Hugh Dickins708e3502011-07-25 17:12:32 -07001370 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1371 unsigned int this_len;
1372
1373 if (!len)
1374 break;
1375
Hugh Dickins708e3502011-07-25 17:12:32 -07001376 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1377 page = spd.pages[page_nr];
1378
Hugh Dickins71f0e072011-07-25 17:12:33 -07001379 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001380 error = shmem_getpage(inode, index, &page,
1381 SGP_CACHE, NULL);
1382 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001383 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001384 unlock_page(page);
1385 page_cache_release(spd.pages[page_nr]);
1386 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001387 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001388
1389 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001390 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1391 if (unlikely(!isize || index > end_index))
1392 break;
1393
Hugh Dickins708e3502011-07-25 17:12:32 -07001394 if (end_index == index) {
1395 unsigned int plen;
1396
Hugh Dickins708e3502011-07-25 17:12:32 -07001397 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1398 if (plen <= loff)
1399 break;
1400
Hugh Dickins708e3502011-07-25 17:12:32 -07001401 this_len = min(this_len, plen - loff);
1402 len = this_len;
1403 }
1404
1405 spd.partial[page_nr].offset = loff;
1406 spd.partial[page_nr].len = this_len;
1407 len -= this_len;
1408 loff = 0;
1409 spd.nr_pages++;
1410 index++;
1411 }
1412
Hugh Dickins708e3502011-07-25 17:12:32 -07001413 while (page_nr < nr_pages)
1414 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001415
1416 if (spd.nr_pages)
1417 error = splice_to_pipe(pipe, &spd);
1418
1419 splice_shrink_spd(pipe, &spd);
1420
1421 if (error > 0) {
1422 *ppos += error;
1423 file_accessed(in);
1424 }
1425 return error;
1426}
1427
David Howells726c3342006-06-23 02:02:58 -07001428static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
David Howells726c3342006-06-23 02:02:58 -07001430 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
1432 buf->f_type = TMPFS_MAGIC;
1433 buf->f_bsize = PAGE_CACHE_SIZE;
1434 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001435 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001437 buf->f_bavail =
1438 buf->f_bfree = sbinfo->max_blocks -
1439 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001440 }
1441 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 buf->f_files = sbinfo->max_inodes;
1443 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 }
1445 /* else leave those fields 0 like simple_statfs */
1446 return 0;
1447}
1448
1449/*
1450 * File creation. Allocate an inode, and we're done..
1451 */
1452static int
1453shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1454{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001455 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 int error = -ENOSPC;
1457
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001458 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001460 error = security_inode_init_security(inode, dir,
1461 &dentry->d_name, NULL,
1462 NULL, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001463 if (error) {
1464 if (error != -EOPNOTSUPP) {
1465 iput(inode);
1466 return error;
1467 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001468 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001469#ifdef CONFIG_TMPFS_POSIX_ACL
1470 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001471 if (error) {
1472 iput(inode);
1473 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001474 }
Al Viro718deb62009-12-16 19:35:36 -05001475#else
1476 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001477#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 dir->i_size += BOGO_DIRENT_SIZE;
1479 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1480 d_instantiate(dentry, inode);
1481 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483 return error;
1484}
1485
1486static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1487{
1488 int error;
1489
1490 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1491 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001492 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return 0;
1494}
1495
1496static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1497 struct nameidata *nd)
1498{
1499 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1500}
1501
1502/*
1503 * Link a file..
1504 */
1505static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1506{
1507 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001508 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 /*
1511 * No ordinary (disk based) filesystem counts links as inodes;
1512 * but each new link needs a new dentry, pinning lowmem, and
1513 * tmpfs dentries cannot be pruned until they are unlinked.
1514 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001515 ret = shmem_reserve_inode(inode->i_sb);
1516 if (ret)
1517 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519 dir->i_size += BOGO_DIRENT_SIZE;
1520 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001521 inc_nlink(inode);
Al Viro7de9c6e2010-10-23 11:11:40 -04001522 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 dget(dentry); /* Extra pinning count for the created dentry */
1524 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001525out:
1526 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
1528
1529static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1530{
1531 struct inode *inode = dentry->d_inode;
1532
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001533 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1534 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536 dir->i_size -= BOGO_DIRENT_SIZE;
1537 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001538 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 dput(dentry); /* Undo the count from "create" - this does all the work */
1540 return 0;
1541}
1542
1543static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1544{
1545 if (!simple_empty(dentry))
1546 return -ENOTEMPTY;
1547
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001548 drop_nlink(dentry->d_inode);
1549 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return shmem_unlink(dir, dentry);
1551}
1552
1553/*
1554 * The VFS layer already does all the dentry stuff for rename,
1555 * we just have to decrement the usage count for the target if
1556 * it exists so that the VFS layer correctly free's it when it
1557 * gets overwritten.
1558 */
1559static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1560{
1561 struct inode *inode = old_dentry->d_inode;
1562 int they_are_dirs = S_ISDIR(inode->i_mode);
1563
1564 if (!simple_empty(new_dentry))
1565 return -ENOTEMPTY;
1566
1567 if (new_dentry->d_inode) {
1568 (void) shmem_unlink(new_dir, new_dentry);
1569 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001570 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001572 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001573 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 }
1575
1576 old_dir->i_size -= BOGO_DIRENT_SIZE;
1577 new_dir->i_size += BOGO_DIRENT_SIZE;
1578 old_dir->i_ctime = old_dir->i_mtime =
1579 new_dir->i_ctime = new_dir->i_mtime =
1580 inode->i_ctime = CURRENT_TIME;
1581 return 0;
1582}
1583
1584static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1585{
1586 int error;
1587 int len;
1588 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07001589 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 char *kaddr;
1591 struct shmem_inode_info *info;
1592
1593 len = strlen(symname) + 1;
1594 if (len > PAGE_CACHE_SIZE)
1595 return -ENAMETOOLONG;
1596
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001597 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (!inode)
1599 return -ENOSPC;
1600
Eric Paris2a7dba32011-02-01 11:05:39 -05001601 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
1602 NULL, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001603 if (error) {
1604 if (error != -EOPNOTSUPP) {
1605 iput(inode);
1606 return error;
1607 }
1608 error = 0;
1609 }
1610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 info = SHMEM_I(inode);
1612 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001613 if (len <= SHORT_SYMLINK_LEN) {
1614 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1615 if (!info->symlink) {
1616 iput(inode);
1617 return -ENOMEM;
1618 }
1619 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 } else {
1621 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1622 if (error) {
1623 iput(inode);
1624 return error;
1625 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07001626 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 inode->i_op = &shmem_symlink_inode_operations;
1628 kaddr = kmap_atomic(page, KM_USER0);
1629 memcpy(kaddr, symname, len);
1630 kunmap_atomic(kaddr, KM_USER0);
1631 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001632 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 page_cache_release(page);
1634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 dir->i_size += BOGO_DIRENT_SIZE;
1636 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1637 d_instantiate(dentry, inode);
1638 dget(dentry);
1639 return 0;
1640}
1641
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001642static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001644 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001645 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646}
1647
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001648static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649{
1650 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001651 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1652 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08001653 if (page)
1654 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001655 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656}
1657
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001658static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659{
1660 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001661 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 kunmap(page);
1663 mark_page_accessed(page);
1664 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 }
1666}
1667
Eric Parisb09e0fa2011-05-24 17:12:39 -07001668#ifdef CONFIG_TMPFS_XATTR
1669/*
1670 * Superblocks without xattr inode operations may get some security.* xattr
1671 * support from the LSM "for free". As soon as we have any other xattrs
1672 * like ACLs, we also need to implement the security.* handlers at
1673 * filesystem level, though.
1674 */
1675
1676static int shmem_xattr_get(struct dentry *dentry, const char *name,
1677 void *buffer, size_t size)
1678{
1679 struct shmem_inode_info *info;
1680 struct shmem_xattr *xattr;
1681 int ret = -ENODATA;
1682
1683 info = SHMEM_I(dentry->d_inode);
1684
1685 spin_lock(&info->lock);
1686 list_for_each_entry(xattr, &info->xattr_list, list) {
1687 if (strcmp(name, xattr->name))
1688 continue;
1689
1690 ret = xattr->size;
1691 if (buffer) {
1692 if (size < xattr->size)
1693 ret = -ERANGE;
1694 else
1695 memcpy(buffer, xattr->value, xattr->size);
1696 }
1697 break;
1698 }
1699 spin_unlock(&info->lock);
1700 return ret;
1701}
1702
1703static int shmem_xattr_set(struct dentry *dentry, const char *name,
1704 const void *value, size_t size, int flags)
1705{
1706 struct inode *inode = dentry->d_inode;
1707 struct shmem_inode_info *info = SHMEM_I(inode);
1708 struct shmem_xattr *xattr;
1709 struct shmem_xattr *new_xattr = NULL;
1710 size_t len;
1711 int err = 0;
1712
1713 /* value == NULL means remove */
1714 if (value) {
1715 /* wrap around? */
1716 len = sizeof(*new_xattr) + size;
1717 if (len <= sizeof(*new_xattr))
1718 return -ENOMEM;
1719
1720 new_xattr = kmalloc(len, GFP_KERNEL);
1721 if (!new_xattr)
1722 return -ENOMEM;
1723
1724 new_xattr->name = kstrdup(name, GFP_KERNEL);
1725 if (!new_xattr->name) {
1726 kfree(new_xattr);
1727 return -ENOMEM;
1728 }
1729
1730 new_xattr->size = size;
1731 memcpy(new_xattr->value, value, size);
1732 }
1733
1734 spin_lock(&info->lock);
1735 list_for_each_entry(xattr, &info->xattr_list, list) {
1736 if (!strcmp(name, xattr->name)) {
1737 if (flags & XATTR_CREATE) {
1738 xattr = new_xattr;
1739 err = -EEXIST;
1740 } else if (new_xattr) {
1741 list_replace(&xattr->list, &new_xattr->list);
1742 } else {
1743 list_del(&xattr->list);
1744 }
1745 goto out;
1746 }
1747 }
1748 if (flags & XATTR_REPLACE) {
1749 xattr = new_xattr;
1750 err = -ENODATA;
1751 } else {
1752 list_add(&new_xattr->list, &info->xattr_list);
1753 xattr = NULL;
1754 }
1755out:
1756 spin_unlock(&info->lock);
1757 if (xattr)
1758 kfree(xattr->name);
1759 kfree(xattr);
1760 return err;
1761}
1762
Eric Parisb09e0fa2011-05-24 17:12:39 -07001763static const struct xattr_handler *shmem_xattr_handlers[] = {
1764#ifdef CONFIG_TMPFS_POSIX_ACL
1765 &generic_acl_access_handler,
1766 &generic_acl_default_handler,
1767#endif
1768 NULL
1769};
1770
1771static int shmem_xattr_validate(const char *name)
1772{
1773 struct { const char *prefix; size_t len; } arr[] = {
1774 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1775 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1776 };
1777 int i;
1778
1779 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1780 size_t preflen = arr[i].len;
1781 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1782 if (!name[preflen])
1783 return -EINVAL;
1784 return 0;
1785 }
1786 }
1787 return -EOPNOTSUPP;
1788}
1789
1790static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1791 void *buffer, size_t size)
1792{
1793 int err;
1794
1795 /*
1796 * If this is a request for a synthetic attribute in the system.*
1797 * namespace use the generic infrastructure to resolve a handler
1798 * for it via sb->s_xattr.
1799 */
1800 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1801 return generic_getxattr(dentry, name, buffer, size);
1802
1803 err = shmem_xattr_validate(name);
1804 if (err)
1805 return err;
1806
1807 return shmem_xattr_get(dentry, name, buffer, size);
1808}
1809
1810static int shmem_setxattr(struct dentry *dentry, const char *name,
1811 const void *value, size_t size, int flags)
1812{
1813 int err;
1814
1815 /*
1816 * If this is a request for a synthetic attribute in the system.*
1817 * namespace use the generic infrastructure to resolve a handler
1818 * for it via sb->s_xattr.
1819 */
1820 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1821 return generic_setxattr(dentry, name, value, size, flags);
1822
1823 err = shmem_xattr_validate(name);
1824 if (err)
1825 return err;
1826
1827 if (size == 0)
1828 value = ""; /* empty EA, do not remove */
1829
1830 return shmem_xattr_set(dentry, name, value, size, flags);
1831
1832}
1833
1834static int shmem_removexattr(struct dentry *dentry, const char *name)
1835{
1836 int err;
1837
1838 /*
1839 * If this is a request for a synthetic attribute in the system.*
1840 * namespace use the generic infrastructure to resolve a handler
1841 * for it via sb->s_xattr.
1842 */
1843 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1844 return generic_removexattr(dentry, name);
1845
1846 err = shmem_xattr_validate(name);
1847 if (err)
1848 return err;
1849
1850 return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1851}
1852
1853static bool xattr_is_trusted(const char *name)
1854{
1855 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1856}
1857
1858static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1859{
1860 bool trusted = capable(CAP_SYS_ADMIN);
1861 struct shmem_xattr *xattr;
1862 struct shmem_inode_info *info;
1863 size_t used = 0;
1864
1865 info = SHMEM_I(dentry->d_inode);
1866
1867 spin_lock(&info->lock);
1868 list_for_each_entry(xattr, &info->xattr_list, list) {
1869 size_t len;
1870
1871 /* skip "trusted." attributes for unprivileged callers */
1872 if (!trusted && xattr_is_trusted(xattr->name))
1873 continue;
1874
1875 len = strlen(xattr->name) + 1;
1876 used += len;
1877 if (buffer) {
1878 if (size < used) {
1879 used = -ERANGE;
1880 break;
1881 }
1882 memcpy(buffer, xattr->name, len);
1883 buffer += len;
1884 }
1885 }
1886 spin_unlock(&info->lock);
1887
1888 return used;
1889}
1890#endif /* CONFIG_TMPFS_XATTR */
1891
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001892static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001894 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07001895#ifdef CONFIG_TMPFS_XATTR
1896 .setxattr = shmem_setxattr,
1897 .getxattr = shmem_getxattr,
1898 .listxattr = shmem_listxattr,
1899 .removexattr = shmem_removexattr,
1900#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901};
1902
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001903static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 .readlink = generic_readlink,
1905 .follow_link = shmem_follow_link,
1906 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07001907#ifdef CONFIG_TMPFS_XATTR
1908 .setxattr = shmem_setxattr,
1909 .getxattr = shmem_getxattr,
1910 .listxattr = shmem_listxattr,
1911 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001912#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07001913};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001914
David M. Grimes91828a42006-10-17 00:09:45 -07001915static struct dentry *shmem_get_parent(struct dentry *child)
1916{
1917 return ERR_PTR(-ESTALE);
1918}
1919
1920static int shmem_match(struct inode *ino, void *vfh)
1921{
1922 __u32 *fh = vfh;
1923 __u64 inum = fh[2];
1924 inum = (inum << 32) | fh[1];
1925 return ino->i_ino == inum && fh[0] == ino->i_generation;
1926}
1927
Christoph Hellwig480b1162007-10-21 16:42:13 -07001928static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1929 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07001930{
David M. Grimes91828a42006-10-17 00:09:45 -07001931 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07001932 struct dentry *dentry = NULL;
1933 u64 inum = fid->raw[2];
1934 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07001935
Christoph Hellwig480b1162007-10-21 16:42:13 -07001936 if (fh_len < 3)
1937 return NULL;
1938
1939 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1940 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07001941 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07001942 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07001943 iput(inode);
1944 }
1945
Christoph Hellwig480b1162007-10-21 16:42:13 -07001946 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07001947}
1948
1949static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1950 int connectable)
1951{
1952 struct inode *inode = dentry->d_inode;
1953
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05301954 if (*len < 3) {
1955 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07001956 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05301957 }
David M. Grimes91828a42006-10-17 00:09:45 -07001958
Al Viro1d3382c2010-10-23 15:19:20 -04001959 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07001960 /* Unfortunately insert_inode_hash is not idempotent,
1961 * so as we hash inodes here rather than at creation
1962 * time, we need a lock to ensure we only try
1963 * to do it once
1964 */
1965 static DEFINE_SPINLOCK(lock);
1966 spin_lock(&lock);
Al Viro1d3382c2010-10-23 15:19:20 -04001967 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07001968 __insert_inode_hash(inode,
1969 inode->i_ino + inode->i_generation);
1970 spin_unlock(&lock);
1971 }
1972
1973 fh[0] = inode->i_generation;
1974 fh[1] = inode->i_ino;
1975 fh[2] = ((__u64)inode->i_ino) >> 32;
1976
1977 *len = 3;
1978 return 1;
1979}
1980
Christoph Hellwig39655162007-10-21 16:42:17 -07001981static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07001982 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07001983 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07001984 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07001985};
1986
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001987static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
1988 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
1990 char *this_char, *value, *rest;
1991
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001992 while (options != NULL) {
1993 this_char = options;
1994 for (;;) {
1995 /*
1996 * NUL-terminate this option: unfortunately,
1997 * mount options form a comma-separated list,
1998 * but mpol's nodelist may also contain commas.
1999 */
2000 options = strchr(options, ',');
2001 if (options == NULL)
2002 break;
2003 options++;
2004 if (!isdigit(*options)) {
2005 options[-1] = '\0';
2006 break;
2007 }
2008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 if (!*this_char)
2010 continue;
2011 if ((value = strchr(this_char,'=')) != NULL) {
2012 *value++ = 0;
2013 } else {
2014 printk(KERN_ERR
2015 "tmpfs: No value for mount option '%s'\n",
2016 this_char);
2017 return 1;
2018 }
2019
2020 if (!strcmp(this_char,"size")) {
2021 unsigned long long size;
2022 size = memparse(value,&rest);
2023 if (*rest == '%') {
2024 size <<= PAGE_SHIFT;
2025 size *= totalram_pages;
2026 do_div(size, 100);
2027 rest++;
2028 }
2029 if (*rest)
2030 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002031 sbinfo->max_blocks =
2032 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002034 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 if (*rest)
2036 goto bad_val;
2037 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002038 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 if (*rest)
2040 goto bad_val;
2041 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002042 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002044 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 if (*rest)
2046 goto bad_val;
2047 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002048 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002050 sbinfo->uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 if (*rest)
2052 goto bad_val;
2053 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002054 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002056 sbinfo->gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 if (*rest)
2058 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002059 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002060 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002061 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 } else {
2063 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2064 this_char);
2065 return 1;
2066 }
2067 }
2068 return 0;
2069
2070bad_val:
2071 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2072 value, this_char);
2073 return 1;
2074
2075}
2076
2077static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2078{
2079 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002080 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002081 unsigned long inodes;
2082 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002084 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002085 return error;
2086
2087 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002088 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002089 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002090 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002091 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002092 goto out;
2093 /*
Hugh Dickins54af6042011-08-03 16:21:24 -07002094 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002095 * but we must separately disallow unlimited->limited, because
2096 * in that case we have no record of how much is already in use.
2097 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002098 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002099 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002100 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002101 goto out;
2102
2103 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002104 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002105 sbinfo->max_inodes = config.max_inodes;
2106 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002107
2108 mpol_put(sbinfo->mpol);
2109 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002110out:
2111 spin_unlock(&sbinfo->stat_lock);
2112 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002114
2115static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2116{
2117 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2118
2119 if (sbinfo->max_blocks != shmem_default_max_blocks())
2120 seq_printf(seq, ",size=%luk",
2121 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2122 if (sbinfo->max_inodes != shmem_default_max_inodes())
2123 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2124 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2125 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2126 if (sbinfo->uid != 0)
2127 seq_printf(seq, ",uid=%u", sbinfo->uid);
2128 if (sbinfo->gid != 0)
2129 seq_printf(seq, ",gid=%u", sbinfo->gid);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002130 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002131 return 0;
2132}
2133#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
2135static void shmem_put_super(struct super_block *sb)
2136{
Hugh Dickins602586a2010-08-17 15:23:56 -07002137 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2138
2139 percpu_counter_destroy(&sbinfo->used_blocks);
2140 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 sb->s_fs_info = NULL;
2142}
2143
Kay Sievers2b2af542009-04-30 15:23:42 +02002144int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145{
2146 struct inode *inode;
2147 struct dentry *root;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002148 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002149 int err = -ENOMEM;
2150
2151 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002152 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002153 L1_CACHE_BYTES), GFP_KERNEL);
2154 if (!sbinfo)
2155 return -ENOMEM;
2156
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002157 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002158 sbinfo->uid = current_fsuid();
2159 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002160 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002162#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 /*
2164 * Per default we only allow half of the physical ram per
2165 * tmpfs instance, limiting inodes to one per page of lowmem;
2166 * but the internal instance is left unlimited.
2167 */
2168 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002169 sbinfo->max_blocks = shmem_default_max_blocks();
2170 sbinfo->max_inodes = shmem_default_max_inodes();
2171 if (shmem_parse_options(data, sbinfo, false)) {
2172 err = -EINVAL;
2173 goto failed;
2174 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 }
David M. Grimes91828a42006-10-17 00:09:45 -07002176 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177#else
2178 sb->s_flags |= MS_NOUSER;
2179#endif
2180
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002181 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002182 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2183 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002184 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002185
Hugh Dickins285b2c42011-08-03 16:21:20 -07002186 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 sb->s_blocksize = PAGE_CACHE_SIZE;
2188 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2189 sb->s_magic = TMPFS_MAGIC;
2190 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002191 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002192#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002193 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002194#endif
2195#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002196 sb->s_flags |= MS_POSIXACL;
2197#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002198
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002199 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 if (!inode)
2201 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002202 inode->i_uid = sbinfo->uid;
2203 inode->i_gid = sbinfo->gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 root = d_alloc_root(inode);
2205 if (!root)
2206 goto failed_iput;
2207 sb->s_root = root;
2208 return 0;
2209
2210failed_iput:
2211 iput(inode);
2212failed:
2213 shmem_put_super(sb);
2214 return err;
2215}
2216
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002217static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
2219static struct inode *shmem_alloc_inode(struct super_block *sb)
2220{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002221 struct shmem_inode_info *info;
2222 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2223 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002225 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226}
2227
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002228static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e32011-01-07 17:49:49 +11002229{
2230 struct inode *inode = container_of(head, struct inode, i_rcu);
2231 INIT_LIST_HEAD(&inode->i_dentry);
2232 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2233}
2234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235static void shmem_destroy_inode(struct inode *inode)
2236{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002237 if ((inode->i_mode & S_IFMT) == S_IFREG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002239 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240}
2241
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002242static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002244 struct shmem_inode_info *info = foo;
2245 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246}
2247
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002248static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249{
2250 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2251 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002252 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return 0;
2254}
2255
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002256static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002258 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259}
2260
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002261static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002263 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002265 .write_begin = shmem_write_begin,
2266 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002268 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002269 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270};
2271
Helge Deller15ad7cd2006-12-06 20:40:36 -08002272static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 .mmap = shmem_mmap,
2274#ifdef CONFIG_TMPFS
2275 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002276 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002277 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002278 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002279 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002280 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002281 .splice_read = shmem_file_splice_read,
Hugh Dickinsae976412007-06-04 10:00:39 +02002282 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283#endif
2284};
2285
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002286static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002287 .setattr = shmem_setattr,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002288 .truncate_range = shmem_truncate_range,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002289#ifdef CONFIG_TMPFS_XATTR
2290 .setxattr = shmem_setxattr,
2291 .getxattr = shmem_getxattr,
2292 .listxattr = shmem_listxattr,
2293 .removexattr = shmem_removexattr,
2294#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295};
2296
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002297static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298#ifdef CONFIG_TMPFS
2299 .create = shmem_create,
2300 .lookup = simple_lookup,
2301 .link = shmem_link,
2302 .unlink = shmem_unlink,
2303 .symlink = shmem_symlink,
2304 .mkdir = shmem_mkdir,
2305 .rmdir = shmem_rmdir,
2306 .mknod = shmem_mknod,
2307 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002309#ifdef CONFIG_TMPFS_XATTR
2310 .setxattr = shmem_setxattr,
2311 .getxattr = shmem_getxattr,
2312 .listxattr = shmem_listxattr,
2313 .removexattr = shmem_removexattr,
2314#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002315#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002316 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002317#endif
2318};
2319
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002320static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002321#ifdef CONFIG_TMPFS_XATTR
2322 .setxattr = shmem_setxattr,
2323 .getxattr = shmem_getxattr,
2324 .listxattr = shmem_listxattr,
2325 .removexattr = shmem_removexattr,
2326#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002327#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002328 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002329#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330};
2331
Hugh Dickins759b9772007-03-05 00:30:28 -08002332static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 .alloc_inode = shmem_alloc_inode,
2334 .destroy_inode = shmem_destroy_inode,
2335#ifdef CONFIG_TMPFS
2336 .statfs = shmem_statfs,
2337 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002338 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339#endif
Al Viro1f895f72010-06-05 19:10:41 -04002340 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 .drop_inode = generic_delete_inode,
2342 .put_super = shmem_put_super,
2343};
2344
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002345static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002346 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347#ifdef CONFIG_NUMA
2348 .set_policy = shmem_set_policy,
2349 .get_policy = shmem_get_policy,
2350#endif
2351};
2352
Al Viro3c26ff62010-07-25 11:46:36 +04002353static struct dentry *shmem_mount(struct file_system_type *fs_type,
2354 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355{
Al Viro3c26ff62010-07-25 11:46:36 +04002356 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357}
2358
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002359static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 .owner = THIS_MODULE,
2361 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002362 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 .kill_sb = kill_litter_super,
2364};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002366int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367{
2368 int error;
2369
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002370 error = bdi_init(&shmem_backing_dev_info);
2371 if (error)
2372 goto out4;
2373
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002374 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 if (error)
2376 goto out3;
2377
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002378 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 if (error) {
2380 printk(KERN_ERR "Could not register tmpfs\n");
2381 goto out2;
2382 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002383
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002384 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2385 shmem_fs_type.name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 if (IS_ERR(shm_mnt)) {
2387 error = PTR_ERR(shm_mnt);
2388 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2389 goto out1;
2390 }
2391 return 0;
2392
2393out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002394 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002396 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002398 bdi_destroy(&shmem_backing_dev_info);
2399out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 shm_mnt = ERR_PTR(error);
2401 return error;
2402}
Matt Mackall853ac432009-01-06 14:40:20 -08002403
2404#else /* !CONFIG_SHMEM */
2405
2406/*
2407 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2408 *
2409 * This is intended for small system where the benefits of the full
2410 * shmem code (swap-backed and resource-limited) are outweighed by
2411 * their complexity. On systems without swap this code should be
2412 * effectively equivalent, but much lighter weight.
2413 */
2414
2415#include <linux/ramfs.h>
2416
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002417static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002418 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002419 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002420 .kill_sb = kill_litter_super,
2421};
2422
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002423int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002424{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002425 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002426
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002427 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002428 BUG_ON(IS_ERR(shm_mnt));
2429
2430 return 0;
2431}
2432
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002433int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002434{
2435 return 0;
2436}
2437
Hugh Dickins3f96b792009-09-21 17:03:37 -07002438int shmem_lock(struct file *file, int lock, struct user_struct *user)
2439{
2440 return 0;
2441}
2442
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002443void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002444{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002445 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002446}
2447EXPORT_SYMBOL_GPL(shmem_truncate_range);
2448
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002449#define shmem_vm_ops generic_file_vm_ops
2450#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002451#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002452#define shmem_acct_size(flags, size) 0
2453#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002454
2455#endif /* CONFIG_SHMEM */
2456
2457/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Randy Dunlap46711812008-03-19 17:00:41 -07002459/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 * @name: name for dentry (to be seen in /proc/<pid>/maps
2462 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002463 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002465struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
2467 int error;
2468 struct file *file;
2469 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002470 struct path path;
2471 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 struct qstr this;
2473
2474 if (IS_ERR(shm_mnt))
2475 return (void *)shm_mnt;
2476
Hugh Dickins285b2c42011-08-03 16:21:20 -07002477 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 return ERR_PTR(-EINVAL);
2479
2480 if (shmem_acct_size(flags, size))
2481 return ERR_PTR(-ENOMEM);
2482
2483 error = -ENOMEM;
2484 this.name = name;
2485 this.len = strlen(name);
2486 this.hash = 0; /* will go */
2487 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002488 path.dentry = d_alloc(root, &this);
2489 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002491 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002494 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002496 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497
Al Viro2c48b9c2009-08-09 00:52:35 +04002498 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 inode->i_size = size;
2500 inode->i_nlink = 0; /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002501#ifndef CONFIG_MMU
2502 error = ramfs_nommu_expand_for_mapping(inode, size);
2503 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002504 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002505#endif
Al Viro4b42af82009-08-05 18:25:56 +04002506
2507 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002508 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002509 &shmem_file_operations);
2510 if (!file)
2511 goto put_dentry;
2512
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 return file;
2514
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002516 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517put_memory:
2518 shmem_unacct_size(flags, size);
2519 return ERR_PTR(error);
2520}
Keith Packard395e0dd2008-06-20 00:08:06 -07002521EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Randy Dunlap46711812008-03-19 17:00:41 -07002523/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2526 */
2527int shmem_zero_setup(struct vm_area_struct *vma)
2528{
2529 struct file *file;
2530 loff_t size = vma->vm_end - vma->vm_start;
2531
2532 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2533 if (IS_ERR(file))
2534 return PTR_ERR(file);
2535
2536 if (vma->vm_file)
2537 fput(vma->vm_file);
2538 vma->vm_file = file;
2539 vma->vm_ops = &shmem_vm_ops;
Hugh Dickinsbee4c362011-03-22 16:33:43 -07002540 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 return 0;
2542}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002543
2544/**
2545 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2546 * @mapping: the page's address_space
2547 * @index: the page index
2548 * @gfp: the page allocator flags to use if allocating
2549 *
2550 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2551 * with any new page allocations done using the specified allocation flags.
2552 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2553 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2554 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2555 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07002556 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2557 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002558 */
2559struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2560 pgoff_t index, gfp_t gfp)
2561{
Hugh Dickins68da9f02011-07-25 17:12:34 -07002562#ifdef CONFIG_SHMEM
2563 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002564 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07002565 int error;
2566
2567 BUG_ON(mapping->a_ops != &shmem_aops);
2568 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2569 if (error)
2570 page = ERR_PTR(error);
2571 else
2572 unlock_page(page);
2573 return page;
2574#else
2575 /*
2576 * The tiny !SHMEM case uses ramfs without swap
2577 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002578 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002579#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002580}
2581EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);