blob: 89341b658bd06b05d908f8becfef7fe0d904b808 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins6922c0c2011-08-03 16:21:25 -07009 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -070011 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 *
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 *
Matt Mackall853ac432009-01-06 14:40:20 -080018 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * This file is released under the GPL.
22 */
23
Matt Mackall853ac432009-01-06 14:40:20 -080024#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070028#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080029#include <linux/file.h>
30#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040031#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080032#include <linux/swap.h>
33
34static struct vfsmount *shm_mnt;
35
36#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
41 */
42
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070043#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070044#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010045#include <linux/posix_acl.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070046#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/string.h>
49#include <linux/slab.h>
50#include <linux/backing-dev.h>
51#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070054#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070055#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070056#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070057#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/security.h>
59#include <linux/swapops.h>
60#include <linux/mempolicy.h>
61#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000062#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070063#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070064#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080065#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040066#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070067
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <asm/pgtable.h>
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074/* Pretend that each entry is of this size in directory's i_size */
75#define BOGO_DIRENT_SIZE 20
76
Hugh Dickins69f07ec2011-08-03 16:21:26 -070077/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78#define SHORT_SYMLINK_LEN 128
79
Hugh Dickins1aac1402012-05-29 15:06:42 -070080/*
81 * shmem_fallocate and shmem_writepage communicate via inode->i_private
82 * (with i_mutex making sure that it has only one user at a time):
83 * we would prefer not to enlarge the shmem inode just for that.
84 */
85struct shmem_falloc {
86 pgoff_t start; /* start of range currently being fallocated */
87 pgoff_t next; /* the next page offset to be fallocated */
88 pgoff_t nr_falloced; /* how many new pages have been fallocated */
89 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
90};
91
Hugh Dickins285b2c42011-08-03 16:21:20 -070092/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -070093enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 SGP_READ, /* don't exceed i_size, don't allocate page */
95 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080096 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -070097 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
98 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099};
100
Andrew Mortonb76db732008-02-08 04:21:49 -0800101#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800102static unsigned long shmem_default_max_blocks(void)
103{
104 return totalram_pages / 2;
105}
106
107static unsigned long shmem_default_max_inodes(void)
108{
109 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
110}
Andrew Mortonb76db732008-02-08 04:21:49 -0800111#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800112
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700113static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700116static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
117 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
118
119static inline int shmem_getpage(struct inode *inode, pgoff_t index,
120 struct page **pagep, enum sgp_type sgp, int *fault_type)
121{
122 return shmem_getpage_gfp(inode, index, pagep, sgp,
123 mapping_gfp_mask(inode->i_mapping), fault_type);
124}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
127{
128 return sb->s_fs_info;
129}
130
131/*
132 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
133 * for shared memory and for shared anonymous (/dev/zero) mappings
134 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
135 * consistent with the pre-accounting of private mappings ...
136 */
137static inline int shmem_acct_size(unsigned long flags, loff_t size)
138{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000139 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000140 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
143static inline void shmem_unacct_size(unsigned long flags, loff_t size)
144{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000145 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 vm_unacct_memory(VM_ACCT(size));
147}
148
149/*
150 * ... whereas tmpfs objects are accounted incrementally as
151 * pages are allocated, in order to allow huge sparse files.
152 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
153 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
154 */
155static inline int shmem_acct_block(unsigned long flags)
156{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000157 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000158 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160
161static inline void shmem_unacct_blocks(unsigned long flags, long pages)
162{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000163 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
165}
166
Hugh Dickins759b9772007-03-05 00:30:28 -0800167static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700168static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800169static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800170static const struct inode_operations shmem_inode_operations;
171static const struct inode_operations shmem_dir_inode_operations;
172static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400173static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700175static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700177 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178};
179
180static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800181static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800183static int shmem_reserve_inode(struct super_block *sb)
184{
185 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
186 if (sbinfo->max_inodes) {
187 spin_lock(&sbinfo->stat_lock);
188 if (!sbinfo->free_inodes) {
189 spin_unlock(&sbinfo->stat_lock);
190 return -ENOSPC;
191 }
192 sbinfo->free_inodes--;
193 spin_unlock(&sbinfo->stat_lock);
194 }
195 return 0;
196}
197
198static void shmem_free_inode(struct super_block *sb)
199{
200 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
201 if (sbinfo->max_inodes) {
202 spin_lock(&sbinfo->stat_lock);
203 sbinfo->free_inodes++;
204 spin_unlock(&sbinfo->stat_lock);
205 }
206}
207
Randy Dunlap46711812008-03-19 17:00:41 -0700208/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700209 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 * @inode: inode to recalc
211 *
212 * We have to calculate the free blocks since the mm can drop
213 * undirtied hole pages behind our back.
214 *
215 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
216 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
217 *
218 * It has to be called with the spinlock held.
219 */
220static void shmem_recalc_inode(struct inode *inode)
221{
222 struct shmem_inode_info *info = SHMEM_I(inode);
223 long freed;
224
225 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
226 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700227 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
228 if (sbinfo->max_blocks)
229 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700231 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 }
234}
235
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700236/*
237 * Replace item expected in radix tree by a new item, while holding tree lock.
238 */
239static int shmem_radix_tree_replace(struct address_space *mapping,
240 pgoff_t index, void *expected, void *replacement)
241{
242 void **pslot;
243 void *item = NULL;
244
245 VM_BUG_ON(!expected);
246 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
247 if (pslot)
248 item = radix_tree_deref_slot_protected(pslot,
249 &mapping->tree_lock);
250 if (item != expected)
251 return -ENOENT;
252 if (replacement)
253 radix_tree_replace_slot(pslot, replacement);
254 else
255 radix_tree_delete(&mapping->page_tree, index);
256 return 0;
257}
258
259/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700260 * Sometimes, before we decide whether to proceed or to fail, we must check
261 * that an entry was not already brought back from swap by a racing thread.
262 *
263 * Checking page is not enough: by the time a SwapCache page is locked, it
264 * might be reused, and again be SwapCache, using the same swap as before.
265 */
266static bool shmem_confirm_swap(struct address_space *mapping,
267 pgoff_t index, swp_entry_t swap)
268{
269 void *item;
270
271 rcu_read_lock();
272 item = radix_tree_lookup(&mapping->page_tree, index);
273 rcu_read_unlock();
274 return item == swp_to_radix_entry(swap);
275}
276
277/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700278 * Like add_to_page_cache_locked, but error if expected item has gone.
279 */
280static int shmem_add_to_page_cache(struct page *page,
281 struct address_space *mapping,
282 pgoff_t index, gfp_t gfp, void *expected)
283{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700284 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700285
286 VM_BUG_ON(!PageLocked(page));
287 VM_BUG_ON(!PageSwapBacked(page));
288
Hugh Dickinsb065b432012-07-11 14:02:48 -0700289 page_cache_get(page);
290 page->mapping = mapping;
291 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700292
Hugh Dickinsb065b432012-07-11 14:02:48 -0700293 spin_lock_irq(&mapping->tree_lock);
294 if (!expected)
295 error = radix_tree_insert(&mapping->page_tree, index, page);
296 else
297 error = shmem_radix_tree_replace(mapping, index, expected,
298 page);
299 if (!error) {
300 mapping->nrpages++;
301 __inc_zone_page_state(page, NR_FILE_PAGES);
302 __inc_zone_page_state(page, NR_SHMEM);
303 spin_unlock_irq(&mapping->tree_lock);
304 } else {
305 page->mapping = NULL;
306 spin_unlock_irq(&mapping->tree_lock);
307 page_cache_release(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700308 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700309 return error;
310}
311
312/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700313 * Like delete_from_page_cache, but substitutes swap for page.
314 */
315static void shmem_delete_from_page_cache(struct page *page, void *radswap)
316{
317 struct address_space *mapping = page->mapping;
318 int error;
319
320 spin_lock_irq(&mapping->tree_lock);
321 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
322 page->mapping = NULL;
323 mapping->nrpages--;
324 __dec_zone_page_state(page, NR_FILE_PAGES);
325 __dec_zone_page_state(page, NR_SHMEM);
326 spin_unlock_irq(&mapping->tree_lock);
327 page_cache_release(page);
328 BUG_ON(error);
329}
330
331/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700332 * Like find_get_pages, but collecting swap entries as well as pages.
333 */
334static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
335 pgoff_t start, unsigned int nr_pages,
336 struct page **pages, pgoff_t *indices)
337{
338 unsigned int i;
339 unsigned int ret;
340 unsigned int nr_found;
341
342 rcu_read_lock();
343restart:
344 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
345 (void ***)pages, indices, start, nr_pages);
346 ret = 0;
347 for (i = 0; i < nr_found; i++) {
348 struct page *page;
349repeat:
350 page = radix_tree_deref_slot((void **)pages[i]);
351 if (unlikely(!page))
352 continue;
353 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700354 if (radix_tree_deref_retry(page))
355 goto restart;
356 /*
357 * Otherwise, we must be storing a swap entry
358 * here as an exceptional entry: so return it
359 * without attempting to raise page count.
360 */
361 goto export;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700362 }
363 if (!page_cache_get_speculative(page))
364 goto repeat;
365
366 /* Has the page moved? */
367 if (unlikely(page != *((void **)pages[i]))) {
368 page_cache_release(page);
369 goto repeat;
370 }
371export:
372 indices[ret] = indices[i];
373 pages[ret] = page;
374 ret++;
375 }
376 if (unlikely(!ret && nr_found))
377 goto restart;
378 rcu_read_unlock();
379 return ret;
380}
381
382/*
383 * Remove swap entry from radix tree, free the swap and its page cache.
384 */
385static int shmem_free_swap(struct address_space *mapping,
386 pgoff_t index, void *radswap)
387{
388 int error;
389
390 spin_lock_irq(&mapping->tree_lock);
391 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
392 spin_unlock_irq(&mapping->tree_lock);
393 if (!error)
394 free_swap_and_cache(radix_to_swp_entry(radswap));
395 return error;
396}
397
398/*
399 * Pagevec may contain swap entries, so shuffle up pages before releasing.
400 */
Hugh Dickins24513262012-01-20 14:34:21 -0800401static void shmem_deswap_pagevec(struct pagevec *pvec)
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700402{
403 int i, j;
404
405 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
406 struct page *page = pvec->pages[i];
407 if (!radix_tree_exceptional_entry(page))
408 pvec->pages[j++] = page;
409 }
410 pvec->nr = j;
Hugh Dickins24513262012-01-20 14:34:21 -0800411}
412
413/*
414 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
415 */
416void shmem_unlock_mapping(struct address_space *mapping)
417{
418 struct pagevec pvec;
419 pgoff_t indices[PAGEVEC_SIZE];
420 pgoff_t index = 0;
421
422 pagevec_init(&pvec, 0);
423 /*
424 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
425 */
426 while (!mapping_unevictable(mapping)) {
427 /*
428 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
429 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
430 */
431 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
432 PAGEVEC_SIZE, pvec.pages, indices);
433 if (!pvec.nr)
434 break;
435 index = indices[pvec.nr - 1] + 1;
436 shmem_deswap_pagevec(&pvec);
437 check_move_unevictable_pages(pvec.pages, pvec.nr);
438 pagevec_release(&pvec);
439 cond_resched();
440 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700441}
442
443/*
444 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700445 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700446 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700447static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
448 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700450 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700452 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700453 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
454 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
455 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700456 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700457 pgoff_t indices[PAGEVEC_SIZE];
458 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700459 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700460 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700462 if (lend == -1)
463 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700464
465 pagevec_init(&pvec, 0);
466 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700467 while (index < end) {
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700468 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700469 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700470 pvec.pages, indices);
471 if (!pvec.nr)
472 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700473 mem_cgroup_uncharge_start();
474 for (i = 0; i < pagevec_count(&pvec); i++) {
475 struct page *page = pvec.pages[i];
476
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700477 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700478 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700479 break;
480
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700481 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700482 if (unfalloc)
483 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700484 nr_swaps_freed += !shmem_free_swap(mapping,
485 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700486 continue;
487 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700488
489 if (!trylock_page(page))
490 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700491 if (!unfalloc || !PageUptodate(page)) {
492 if (page->mapping == mapping) {
493 VM_BUG_ON(PageWriteback(page));
494 truncate_inode_page(mapping, page);
495 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700496 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700497 unlock_page(page);
498 }
Hugh Dickins24513262012-01-20 14:34:21 -0800499 shmem_deswap_pagevec(&pvec);
500 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700501 mem_cgroup_uncharge_end();
502 cond_resched();
503 index++;
504 }
505
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700506 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700507 struct page *page = NULL;
508 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
509 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700510 unsigned int top = PAGE_CACHE_SIZE;
511 if (start > end) {
512 top = partial_end;
513 partial_end = 0;
514 }
515 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700516 set_page_dirty(page);
517 unlock_page(page);
518 page_cache_release(page);
519 }
520 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700521 if (partial_end) {
522 struct page *page = NULL;
523 shmem_getpage(inode, end, &page, SGP_READ, NULL);
524 if (page) {
525 zero_user_segment(page, 0, partial_end);
526 set_page_dirty(page);
527 unlock_page(page);
528 page_cache_release(page);
529 }
530 }
531 if (start >= end)
532 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700533
534 index = start;
535 for ( ; ; ) {
536 cond_resched();
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700537 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700538 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700539 pvec.pages, indices);
540 if (!pvec.nr) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700541 if (index == start || unfalloc)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700542 break;
543 index = start;
544 continue;
545 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700546 if ((index == start || unfalloc) && indices[0] >= end) {
Hugh Dickins24513262012-01-20 14:34:21 -0800547 shmem_deswap_pagevec(&pvec);
548 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700549 break;
550 }
551 mem_cgroup_uncharge_start();
552 for (i = 0; i < pagevec_count(&pvec); i++) {
553 struct page *page = pvec.pages[i];
554
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700555 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700556 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700557 break;
558
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700559 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700560 if (unfalloc)
561 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700562 nr_swaps_freed += !shmem_free_swap(mapping,
563 index, page);
564 continue;
565 }
566
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700567 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700568 if (!unfalloc || !PageUptodate(page)) {
569 if (page->mapping == mapping) {
570 VM_BUG_ON(PageWriteback(page));
571 truncate_inode_page(mapping, page);
572 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700573 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700574 unlock_page(page);
575 }
Hugh Dickins24513262012-01-20 14:34:21 -0800576 shmem_deswap_pagevec(&pvec);
577 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700578 mem_cgroup_uncharge_end();
579 index++;
580 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700583 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 shmem_recalc_inode(inode);
585 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700586}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700588void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
589{
590 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700591 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700593EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Hugh Dickins94c1e622011-06-27 16:18:03 -0700595static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
597 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 int error;
599
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200600 error = inode_change_ok(inode, attr);
601 if (error)
602 return error;
603
Hugh Dickins94c1e622011-06-27 16:18:03 -0700604 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
605 loff_t oldsize = inode->i_size;
606 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000607
Hugh Dickins94c1e622011-06-27 16:18:03 -0700608 if (newsize != oldsize) {
609 i_size_write(inode, newsize);
610 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
611 }
612 if (newsize < oldsize) {
613 loff_t holebegin = round_up(newsize, PAGE_SIZE);
614 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
615 shmem_truncate_range(inode, newsize, (loff_t)-1);
616 /* unmap again to remove racily COWed private pages */
617 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
618 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
620
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200621 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700622#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200623 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100624 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700625#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 return error;
627}
628
Al Viro1f895f72010-06-05 19:10:41 -0400629static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 struct shmem_inode_info *info = SHMEM_I(inode);
632
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000633 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 shmem_unacct_size(info->flags, inode->i_size);
635 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000636 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800638 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800640 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700642 } else
643 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700644
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400645 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800646 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800647 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200648 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700651/*
652 * If swap found in inode, free it and move page from swapcache to filecache.
653 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700654static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700655 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700657 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700658 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700659 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700660 gfp_t gfp;
661 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700663 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700664 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700665 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700666 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800667
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800668 /*
669 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400670 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800671 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700672 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800673 */
674 if (shmem_swaplist.next != &info->swaplist)
675 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800676
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700677 gfp = mapping_gfp_mask(mapping);
678 if (shmem_should_replace_page(*pagep, gfp)) {
679 mutex_unlock(&shmem_swaplist_mutex);
680 error = shmem_replace_page(pagep, gfp, info, index);
681 mutex_lock(&shmem_swaplist_mutex);
682 /*
683 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700684 * allocation, but the inode might have been freed while we
685 * dropped it: although a racing shmem_evict_inode() cannot
686 * complete without emptying the radix_tree, our page lock
687 * on this swapcache page is not enough to prevent that -
688 * free_swap_and_cache() of our swap entry will only
689 * trylock_page(), removing swap from radix_tree whatever.
690 *
691 * We must not proceed to shmem_add_to_page_cache() if the
692 * inode has been freed, but of course we cannot rely on
693 * inode or mapping or info to check that. However, we can
694 * safely check if our swap entry is still in use (and here
695 * it can't have got reused for another page): if it's still
696 * in use, then the inode cannot have been freed yet, and we
697 * can safely proceed (if it's no longer in use, that tells
698 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700699 */
700 if (!page_swapcount(*pagep))
701 error = -ENOENT;
702 }
703
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800704 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700705 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
706 * but also to hold up shmem_evict_inode(): so inode cannot be freed
707 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800708 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700709 if (!error)
710 error = shmem_add_to_page_cache(*pagep, mapping, index,
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700711 GFP_NOWAIT, radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700712 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700713 /*
714 * Truncation and eviction use free_swap_and_cache(), which
715 * only does trylock page: if we raced, best clean up here.
716 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700717 delete_from_swap_cache(*pagep);
718 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700719 if (!error) {
720 spin_lock(&info->lock);
721 info->swapped--;
722 spin_unlock(&info->lock);
723 swap_free(swap);
724 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800725 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800727 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728}
729
730/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700731 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700733int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700735 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 struct shmem_inode_info *info;
737 int found = 0;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700738 int error = 0;
739
740 /*
741 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700742 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700743 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700744 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700745 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700746
747 /*
748 * Charge page using GFP_KERNEL while we can wait, before taking
749 * the shmem_swaplist_mutex which might hold up shmem_writepage().
750 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700751 */
752 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
753 if (error)
754 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700755 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800757 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700758 list_for_each_safe(this, next, &shmem_swaplist) {
759 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700760 if (info->swapped)
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700761 found = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700762 else
763 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800764 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800765 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700766 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800768 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700769
Hugh Dickins778dd892011-05-11 15:13:37 -0700770 if (found < 0)
771 error = found;
772out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800773 unlock_page(page);
774 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700775 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
778/*
779 * Move the page from the page cache to the swap cache.
780 */
781static int shmem_writepage(struct page *page, struct writeback_control *wbc)
782{
783 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700786 swp_entry_t swap;
787 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 mapping = page->mapping;
791 index = page->index;
792 inode = mapping->host;
793 info = SHMEM_I(inode);
794 if (info->flags & VM_LOCKED)
795 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800796 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 goto redirty;
798
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800799 /*
800 * shmem_backing_dev_info's capabilities prevent regular writeback or
801 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700802 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800803 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700804 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800805 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700806 if (!wbc->for_reclaim) {
807 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
808 goto redirty;
809 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700810
811 /*
812 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
813 * value into swapfile.c, the only way we can correctly account for a
814 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700815 *
816 * That's okay for a page already fallocated earlier, but if we have
817 * not yet completed the fallocation, then (a) we want to keep track
818 * of this page in case we have to undo it, and (b) it may not be a
819 * good idea to continue anyway, once we're pushing into swap. So
820 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700821 */
822 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700823 if (inode->i_private) {
824 struct shmem_falloc *shmem_falloc;
825 spin_lock(&inode->i_lock);
826 shmem_falloc = inode->i_private;
827 if (shmem_falloc &&
828 index >= shmem_falloc->start &&
829 index < shmem_falloc->next)
830 shmem_falloc->nr_unswapped++;
831 else
832 shmem_falloc = NULL;
833 spin_unlock(&inode->i_lock);
834 if (shmem_falloc)
835 goto redirty;
836 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700837 clear_highpage(page);
838 flush_dcache_page(page);
839 SetPageUptodate(page);
840 }
841
Hugh Dickins48f170f2011-07-25 17:12:37 -0700842 swap = get_swap_page();
843 if (!swap.val)
844 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800845
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700846 /*
847 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700848 * if it's not already there. Do it now before the page is
849 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700850 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700851 * we've incremented swapped, because shmem_unuse_inode() will
852 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700853 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700854 mutex_lock(&shmem_swaplist_mutex);
855 if (list_empty(&info->swaplist))
856 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700857
Hugh Dickins48f170f2011-07-25 17:12:37 -0700858 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800859 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700860 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
861
862 spin_lock(&info->lock);
863 info->swapped++;
864 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700865 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700866
867 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800868 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700869 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return 0;
871 }
872
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700873 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700874 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875redirty:
876 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800877 if (wbc->for_reclaim)
878 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
879 unlock_page(page);
880 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
882
883#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800884#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700885static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800886{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700887 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800888
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700889 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700890 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800891
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700892 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800893
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700894 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800895}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700896
897static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
898{
899 struct mempolicy *mpol = NULL;
900 if (sbinfo->mpol) {
901 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
902 mpol = sbinfo->mpol;
903 mpol_get(mpol);
904 spin_unlock(&sbinfo->stat_lock);
905 }
906 return mpol;
907}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800908#endif /* CONFIG_TMPFS */
909
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700910static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
911 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700913 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 struct vm_area_struct pvma;
915
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700916 spol = mpol_cond_copy(&mpol,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700917 mpol_shared_policy_lookup(&info->policy, index));
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800920 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700921 /* Bias interleave by inode number to distribute better across nodes */
922 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800923 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700924 pvma.vm_policy = spol;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700925 return swapin_readahead(swap, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
927
Hugh Dickins02098fe2008-02-04 22:28:42 -0800928static struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700929 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
931 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800933 /* Create a pseudo vma that just contains the policy */
934 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700935 /* Bias interleave by inode number to distribute better across nodes */
936 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800937 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700938 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700939
940 /*
941 * alloc_page_vma() will drop the shared policy reference
942 */
943 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800945#else /* !CONFIG_NUMA */
946#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700947static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800948{
949}
950#endif /* CONFIG_TMPFS */
951
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700952static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
953 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700955 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956}
957
Hugh Dickins02098fe2008-02-04 22:28:42 -0800958static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700959 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000961 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800963#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700965#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
966static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
967{
968 return NULL;
969}
970#endif
971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700973 * When a page is moved from swapcache to shmem filecache (either by the
974 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
975 * shmem_unuse_inode()), it may have been read in earlier from swap, in
976 * ignorance of the mapping it belongs to. If that mapping has special
977 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
978 * we may need to copy to a suitable page before moving to filecache.
979 *
980 * In a future release, this may well be extended to respect cpuset and
981 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
982 * but for now it is a simple matter of zone.
983 */
984static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
985{
986 return page_zonenum(page) > gfp_zone(gfp);
987}
988
989static int shmem_replace_page(struct page **pagep, gfp_t gfp,
990 struct shmem_inode_info *info, pgoff_t index)
991{
992 struct page *oldpage, *newpage;
993 struct address_space *swap_mapping;
994 pgoff_t swap_index;
995 int error;
996
997 oldpage = *pagep;
998 swap_index = page_private(oldpage);
999 swap_mapping = page_mapping(oldpage);
1000
1001 /*
1002 * We have arrived here because our zones are constrained, so don't
1003 * limit chance of success by further cpuset and node constraints.
1004 */
1005 gfp &= ~GFP_CONSTRAINT_MASK;
1006 newpage = shmem_alloc_page(gfp, info, index);
1007 if (!newpage)
1008 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001009
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001010 page_cache_get(newpage);
1011 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001012 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001013
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001014 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001015 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001016 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001017 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001018 SetPageSwapCache(newpage);
1019
1020 /*
1021 * Our caller will very soon move newpage out of swapcache, but it's
1022 * a nice clean interface for us to replace oldpage by newpage there.
1023 */
1024 spin_lock_irq(&swap_mapping->tree_lock);
1025 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1026 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001027 if (!error) {
1028 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1029 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1030 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001031 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001032
Hugh Dickins0142ef62012-06-07 14:21:09 -07001033 if (unlikely(error)) {
1034 /*
1035 * Is this possible? I think not, now that our callers check
1036 * both PageSwapCache and page_private after getting page lock;
1037 * but be defensive. Reverse old to newpage for clear and free.
1038 */
1039 oldpage = newpage;
1040 } else {
1041 mem_cgroup_replace_page_cache(oldpage, newpage);
1042 lru_cache_add_anon(newpage);
1043 *pagep = newpage;
1044 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001045
1046 ClearPageSwapCache(oldpage);
1047 set_page_private(oldpage, 0);
1048
1049 unlock_page(oldpage);
1050 page_cache_release(oldpage);
1051 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001052 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001053}
1054
1055/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001056 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 *
1058 * If we allocate a new one we do not mark it dirty. That's up to the
1059 * vm. If we swap it in we mark it dirty since we also free the swap
1060 * entry since a page cannot live in both the swap and page cache
1061 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001062static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001063 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
1065 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001066 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001068 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 swp_entry_t swap;
1070 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001071 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001072 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001074 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001077 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001078 page = find_lock_page(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001079 if (radix_tree_exceptional_entry(page)) {
1080 swap = radix_to_swp_entry(page);
1081 page = NULL;
1082 }
1083
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001084 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001085 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1086 error = -EINVAL;
1087 goto failed;
1088 }
1089
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001090 /* fallocated page? */
1091 if (page && !PageUptodate(page)) {
1092 if (sgp != SGP_READ)
1093 goto clear;
1094 unlock_page(page);
1095 page_cache_release(page);
1096 page = NULL;
1097 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001098 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001099 *pagep = page;
1100 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001104 * Fast cache lookup did not find it:
1105 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001107 info = SHMEM_I(inode);
1108 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (swap.val) {
1111 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001112 page = lookup_swap_cache(swap);
1113 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001114 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001115 if (fault_type)
1116 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001117 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001118 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001119 error = -ENOMEM;
1120 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
1123
1124 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001125 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001126 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001127 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001128 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001129 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001130 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001131 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001133 goto failed;
1134 }
1135 wait_on_page_writeback(page);
1136
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001137 if (shmem_should_replace_page(page, gfp)) {
1138 error = shmem_replace_page(&page, gfp, info, index);
1139 if (error)
1140 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
1142
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001143 error = mem_cgroup_cache_charge(page, current->mm,
1144 gfp & GFP_RECLAIM_MASK);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001145 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001146 error = shmem_add_to_page_cache(page, mapping, index,
1147 gfp, swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001148 /*
1149 * We already confirmed swap under page lock, and make
1150 * no memory allocation here, so usually no possibility
1151 * of error; but free_swap_and_cache() only trylocks a
1152 * page, so it is just possible that the entry has been
1153 * truncated or holepunched since swap was confirmed.
1154 * shmem_undo_range() will have done some of the
1155 * unaccounting, now delete_from_swap_cache() will do
1156 * the rest (including mem_cgroup_uncharge_swapcache).
1157 * Reset swap.val? No, leave it so "failed" goes back to
1158 * "repeat": reading a hole and writing should succeed.
1159 */
1160 if (error)
1161 delete_from_swap_cache(page);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001162 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001163 if (error)
1164 goto failed;
1165
1166 spin_lock(&info->lock);
1167 info->swapped--;
1168 shmem_recalc_inode(inode);
1169 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001170
Hugh Dickins27ab7002011-07-25 17:12:36 -07001171 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001172 set_page_dirty(page);
1173 swap_free(swap);
1174
Hugh Dickins54af60422011-08-03 16:21:24 -07001175 } else {
1176 if (shmem_acct_block(info->flags)) {
1177 error = -ENOSPC;
1178 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001180 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001181 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001182 sbinfo->max_blocks) >= 0) {
1183 error = -ENOSPC;
1184 goto unacct;
1185 }
Tim Chen7e496292010-08-09 17:19:05 -07001186 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Hugh Dickins54af60422011-08-03 16:21:24 -07001189 page = shmem_alloc_page(gfp, info, index);
1190 if (!page) {
1191 error = -ENOMEM;
1192 goto decused;
1193 }
1194
1195 SetPageSwapBacked(page);
1196 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001197 error = mem_cgroup_cache_charge(page, current->mm,
1198 gfp & GFP_RECLAIM_MASK);
Hugh Dickins54af60422011-08-03 16:21:24 -07001199 if (error)
1200 goto decused;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001201 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1202 if (!error) {
1203 error = shmem_add_to_page_cache(page, mapping, index,
1204 gfp, NULL);
1205 radix_tree_preload_end();
1206 }
1207 if (error) {
1208 mem_cgroup_uncharge_cache_page(page);
1209 goto decused;
1210 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001211 lru_cache_add_anon(page);
1212
1213 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001215 inode->i_blocks += BLOCKS_PER_PAGE;
1216 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001218 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001219
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001220 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001221 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1222 */
1223 if (sgp == SGP_FALLOC)
1224 sgp = SGP_WRITE;
1225clear:
1226 /*
1227 * Let SGP_WRITE caller clear ends if write does not fill page;
1228 * but SGP_FALLOC on a page fallocated earlier must initialize
1229 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001230 */
1231 if (sgp != SGP_WRITE) {
1232 clear_highpage(page);
1233 flush_dcache_page(page);
1234 SetPageUptodate(page);
1235 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001236 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001237 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001238 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001239
Hugh Dickins54af60422011-08-03 16:21:24 -07001240 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001241 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001242 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1243 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001244 if (alloced)
1245 goto trunc;
1246 else
1247 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001248 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001249 *pagep = page;
1250 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001251
Nick Piggind0217ac2007-07-19 01:47:03 -07001252 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001253 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001255trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001256 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001257 ClearPageDirty(page);
1258 delete_from_page_cache(page);
1259 spin_lock(&info->lock);
1260 info->alloced--;
1261 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001263decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001264 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001265 if (sbinfo->max_blocks)
1266 percpu_counter_add(&sbinfo->used_blocks, -1);
1267unacct:
1268 shmem_unacct_blocks(info->flags, 1);
1269failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001270 if (swap.val && error != -EINVAL &&
1271 !shmem_confirm_swap(mapping, index, swap))
1272 error = -EEXIST;
1273unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001274 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001275 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001276 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001277 }
1278 if (error == -ENOSPC && !once++) {
1279 info = SHMEM_I(inode);
1280 spin_lock(&info->lock);
1281 shmem_recalc_inode(inode);
1282 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001283 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001284 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001285 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001286 goto repeat;
1287 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288}
1289
1290static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1291{
1292 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1293 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001294 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1297 if (error)
1298 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001299
Ying Han456f9982011-05-26 16:25:38 -07001300 if (ret & VM_FAULT_MAJOR) {
1301 count_vm_event(PGMAJFAULT);
1302 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1303 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001304 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305}
1306
1307#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001308static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001310 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1311 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312}
1313
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001314static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1315 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001317 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1318 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001320 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1321 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322}
1323#endif
1324
1325int shmem_lock(struct file *file, int lock, struct user_struct *user)
1326{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001327 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 struct shmem_inode_info *info = SHMEM_I(inode);
1329 int retval = -ENOMEM;
1330
1331 spin_lock(&info->lock);
1332 if (lock && !(info->flags & VM_LOCKED)) {
1333 if (!user_shm_lock(inode->i_size, user))
1334 goto out_nomem;
1335 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001336 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
1338 if (!lock && (info->flags & VM_LOCKED) && user) {
1339 user_shm_unlock(inode->i_size, user);
1340 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001341 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
1343 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345out_nomem:
1346 spin_unlock(&info->lock);
1347 return retval;
1348}
1349
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001350static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
1352 file_accessed(file);
1353 vma->vm_ops = &shmem_vm_ops;
1354 return 0;
1355}
1356
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001357static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001358 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359{
1360 struct inode *inode;
1361 struct shmem_inode_info *info;
1362 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1363
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001364 if (shmem_reserve_inode(sb))
1365 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 inode = new_inode(sb);
1368 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001369 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001370 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1373 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001374 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 info = SHMEM_I(inode);
1376 memset(info, 0, (char *)inode - (char *)info);
1377 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001378 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001380 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001381 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
1383 switch (mode & S_IFMT) {
1384 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001385 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 init_special_inode(inode, mode, dev);
1387 break;
1388 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001389 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 inode->i_op = &shmem_inode_operations;
1391 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001392 mpol_shared_policy_init(&info->policy,
1393 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 break;
1395 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001396 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 /* Some things misbehave if size == 0 on a directory */
1398 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1399 inode->i_op = &shmem_dir_inode_operations;
1400 inode->i_fop = &simple_dir_operations;
1401 break;
1402 case S_IFLNK:
1403 /*
1404 * Must not load anything in the rbtree,
1405 * mpol_free_shared_policy will not be called.
1406 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001407 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 break;
1409 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001410 } else
1411 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return inode;
1413}
1414
1415#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001416static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001417static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001419#ifdef CONFIG_TMPFS_XATTR
1420static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1421#else
1422#define shmem_initxattrs NULL
1423#endif
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001426shmem_write_begin(struct file *file, struct address_space *mapping,
1427 loff_t pos, unsigned len, unsigned flags,
1428 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
Nick Piggin800d15a2007-10-16 01:25:03 -07001430 struct inode *inode = mapping->host;
1431 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001432 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1433}
1434
1435static int
1436shmem_write_end(struct file *file, struct address_space *mapping,
1437 loff_t pos, unsigned len, unsigned copied,
1438 struct page *page, void *fsdata)
1439{
1440 struct inode *inode = mapping->host;
1441
Hugh Dickinsd3602442008-02-04 22:28:44 -08001442 if (pos + copied > inode->i_size)
1443 i_size_write(inode, pos + copied);
1444
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001445 if (!PageUptodate(page)) {
1446 if (copied < PAGE_CACHE_SIZE) {
1447 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1448 zero_user_segments(page, 0, from,
1449 from + copied, PAGE_CACHE_SIZE);
1450 }
1451 SetPageUptodate(page);
1452 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001453 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001454 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001455 page_cache_release(page);
1456
Nick Piggin800d15a2007-10-16 01:25:03 -07001457 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458}
1459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1461{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001462 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001464 pgoff_t index;
1465 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001466 enum sgp_type sgp = SGP_READ;
1467
1468 /*
1469 * Might this read be for a stacking filesystem? Then when reading
1470 * holes of a sparse file, we actually need to allocate those pages,
1471 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1472 */
1473 if (segment_eq(get_fs(), KERNEL_DS))
1474 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 index = *ppos >> PAGE_CACHE_SHIFT;
1477 offset = *ppos & ~PAGE_CACHE_MASK;
1478
1479 for (;;) {
1480 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001481 pgoff_t end_index;
1482 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 loff_t i_size = i_size_read(inode);
1484
1485 end_index = i_size >> PAGE_CACHE_SHIFT;
1486 if (index > end_index)
1487 break;
1488 if (index == end_index) {
1489 nr = i_size & ~PAGE_CACHE_MASK;
1490 if (nr <= offset)
1491 break;
1492 }
1493
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001494 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (desc->error) {
1496 if (desc->error == -EINVAL)
1497 desc->error = 0;
1498 break;
1499 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001500 if (page)
1501 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 /*
1504 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001505 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 */
1507 nr = PAGE_CACHE_SIZE;
1508 i_size = i_size_read(inode);
1509 end_index = i_size >> PAGE_CACHE_SHIFT;
1510 if (index == end_index) {
1511 nr = i_size & ~PAGE_CACHE_MASK;
1512 if (nr <= offset) {
1513 if (page)
1514 page_cache_release(page);
1515 break;
1516 }
1517 }
1518 nr -= offset;
1519
1520 if (page) {
1521 /*
1522 * If users can be writing to this page using arbitrary
1523 * virtual addresses, take care about potential aliasing
1524 * before reading the page on the kernel side.
1525 */
1526 if (mapping_writably_mapped(mapping))
1527 flush_dcache_page(page);
1528 /*
1529 * Mark the page accessed if we read the beginning.
1530 */
1531 if (!offset)
1532 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001533 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001535 page_cache_get(page);
1536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
1538 /*
1539 * Ok, we have the page, and it's up-to-date, so
1540 * now we can copy it to user space...
1541 *
1542 * The actor routine returns how many bytes were actually used..
1543 * NOTE! This may not be the same as how much of a user buffer
1544 * we filled up (we may be padding etc), so we can only update
1545 * "pos" here (the actor routine has to update the user buffer
1546 * pointers and the remaining count).
1547 */
1548 ret = actor(desc, page, offset, nr);
1549 offset += ret;
1550 index += offset >> PAGE_CACHE_SHIFT;
1551 offset &= ~PAGE_CACHE_MASK;
1552
1553 page_cache_release(page);
1554 if (ret != nr || !desc->count)
1555 break;
1556
1557 cond_resched();
1558 }
1559
1560 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1561 file_accessed(filp);
1562}
1563
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001564static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1565 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001567 struct file *filp = iocb->ki_filp;
1568 ssize_t retval;
1569 unsigned long seg;
1570 size_t count;
1571 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001573 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1574 if (retval)
1575 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001577 for (seg = 0; seg < nr_segs; seg++) {
1578 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001580 desc.written = 0;
1581 desc.arg.buf = iov[seg].iov_base;
1582 desc.count = iov[seg].iov_len;
1583 if (desc.count == 0)
1584 continue;
1585 desc.error = 0;
1586 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1587 retval += desc.written;
1588 if (desc.error) {
1589 retval = retval ?: desc.error;
1590 break;
1591 }
1592 if (desc.count > 0)
1593 break;
1594 }
1595 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
1597
Hugh Dickins708e3502011-07-25 17:12:32 -07001598static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1599 struct pipe_inode_info *pipe, size_t len,
1600 unsigned int flags)
1601{
1602 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001603 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001604 unsigned int loff, nr_pages, req_pages;
1605 struct page *pages[PIPE_DEF_BUFFERS];
1606 struct partial_page partial[PIPE_DEF_BUFFERS];
1607 struct page *page;
1608 pgoff_t index, end_index;
1609 loff_t isize, left;
1610 int error, page_nr;
1611 struct splice_pipe_desc spd = {
1612 .pages = pages,
1613 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001614 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001615 .flags = flags,
1616 .ops = &page_cache_pipe_buf_ops,
1617 .spd_release = spd_release_page,
1618 };
1619
Hugh Dickins71f0e072011-07-25 17:12:33 -07001620 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001621 if (unlikely(*ppos >= isize))
1622 return 0;
1623
1624 left = isize - *ppos;
1625 if (unlikely(left < len))
1626 len = left;
1627
1628 if (splice_grow_spd(pipe, &spd))
1629 return -ENOMEM;
1630
1631 index = *ppos >> PAGE_CACHE_SHIFT;
1632 loff = *ppos & ~PAGE_CACHE_MASK;
1633 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1634 nr_pages = min(req_pages, pipe->buffers);
1635
Hugh Dickins708e3502011-07-25 17:12:32 -07001636 spd.nr_pages = find_get_pages_contig(mapping, index,
1637 nr_pages, spd.pages);
1638 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001639 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001640
Hugh Dickins708e3502011-07-25 17:12:32 -07001641 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001642 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1643 if (error)
1644 break;
1645 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001646 spd.pages[spd.nr_pages++] = page;
1647 index++;
1648 }
1649
Hugh Dickins708e3502011-07-25 17:12:32 -07001650 index = *ppos >> PAGE_CACHE_SHIFT;
1651 nr_pages = spd.nr_pages;
1652 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001653
Hugh Dickins708e3502011-07-25 17:12:32 -07001654 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1655 unsigned int this_len;
1656
1657 if (!len)
1658 break;
1659
Hugh Dickins708e3502011-07-25 17:12:32 -07001660 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1661 page = spd.pages[page_nr];
1662
Hugh Dickins71f0e072011-07-25 17:12:33 -07001663 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001664 error = shmem_getpage(inode, index, &page,
1665 SGP_CACHE, NULL);
1666 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001667 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001668 unlock_page(page);
1669 page_cache_release(spd.pages[page_nr]);
1670 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001671 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001672
1673 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001674 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1675 if (unlikely(!isize || index > end_index))
1676 break;
1677
Hugh Dickins708e3502011-07-25 17:12:32 -07001678 if (end_index == index) {
1679 unsigned int plen;
1680
Hugh Dickins708e3502011-07-25 17:12:32 -07001681 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1682 if (plen <= loff)
1683 break;
1684
Hugh Dickins708e3502011-07-25 17:12:32 -07001685 this_len = min(this_len, plen - loff);
1686 len = this_len;
1687 }
1688
1689 spd.partial[page_nr].offset = loff;
1690 spd.partial[page_nr].len = this_len;
1691 len -= this_len;
1692 loff = 0;
1693 spd.nr_pages++;
1694 index++;
1695 }
1696
Hugh Dickins708e3502011-07-25 17:12:32 -07001697 while (page_nr < nr_pages)
1698 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001699
1700 if (spd.nr_pages)
1701 error = splice_to_pipe(pipe, &spd);
1702
Eric Dumazet047fe362012-06-12 15:24:40 +02001703 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001704
1705 if (error > 0) {
1706 *ppos += error;
1707 file_accessed(in);
1708 }
1709 return error;
1710}
1711
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001712static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1713 loff_t len)
1714{
1715 struct inode *inode = file->f_path.dentry->d_inode;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001716 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001717 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001718 pgoff_t start, index, end;
1719 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001720
1721 mutex_lock(&inode->i_mutex);
1722
1723 if (mode & FALLOC_FL_PUNCH_HOLE) {
1724 struct address_space *mapping = file->f_mapping;
1725 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1726 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1727
1728 if ((u64)unmap_end > (u64)unmap_start)
1729 unmap_mapping_range(mapping, unmap_start,
1730 1 + unmap_end - unmap_start, 0);
1731 shmem_truncate_range(inode, offset, offset + len - 1);
1732 /* No need to unmap again: hole-punching leaves COWed pages */
1733 error = 0;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001734 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001735 }
1736
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001737 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1738 error = inode_newsize_ok(inode, offset + len);
1739 if (error)
1740 goto out;
1741
1742 start = offset >> PAGE_CACHE_SHIFT;
1743 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1744 /* Try to avoid a swapstorm if len is impossible to satisfy */
1745 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1746 error = -ENOSPC;
1747 goto out;
1748 }
1749
Hugh Dickins1aac1402012-05-29 15:06:42 -07001750 shmem_falloc.start = start;
1751 shmem_falloc.next = start;
1752 shmem_falloc.nr_falloced = 0;
1753 shmem_falloc.nr_unswapped = 0;
1754 spin_lock(&inode->i_lock);
1755 inode->i_private = &shmem_falloc;
1756 spin_unlock(&inode->i_lock);
1757
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001758 for (index = start; index < end; index++) {
1759 struct page *page;
1760
1761 /*
1762 * Good, the fallocate(2) manpage permits EINTR: we may have
1763 * been interrupted because we are using up too much memory.
1764 */
1765 if (signal_pending(current))
1766 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001767 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1768 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001769 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001770 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001771 NULL);
1772 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001773 /* Remove the !PageUptodate pages we added */
1774 shmem_undo_range(inode,
1775 (loff_t)start << PAGE_CACHE_SHIFT,
1776 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001777 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001778 }
1779
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001780 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07001781 * Inform shmem_writepage() how far we have reached.
1782 * No need for lock or barrier: we have the page lock.
1783 */
1784 shmem_falloc.next++;
1785 if (!PageUptodate(page))
1786 shmem_falloc.nr_falloced++;
1787
1788 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001789 * If !PageUptodate, leave it that way so that freeable pages
1790 * can be recognized if we need to rollback on error later.
1791 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001792 * than free the pages we are allocating (and SGP_CACHE pages
1793 * might still be clean: we now need to mark those dirty too).
1794 */
1795 set_page_dirty(page);
1796 unlock_page(page);
1797 page_cache_release(page);
1798 cond_resched();
1799 }
1800
1801 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1802 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001803 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001804undone:
1805 spin_lock(&inode->i_lock);
1806 inode->i_private = NULL;
1807 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001808out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001809 mutex_unlock(&inode->i_mutex);
1810 return error;
1811}
1812
David Howells726c3342006-06-23 02:02:58 -07001813static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814{
David Howells726c3342006-06-23 02:02:58 -07001815 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
1817 buf->f_type = TMPFS_MAGIC;
1818 buf->f_bsize = PAGE_CACHE_SIZE;
1819 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001820 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001822 buf->f_bavail =
1823 buf->f_bfree = sbinfo->max_blocks -
1824 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001825 }
1826 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 buf->f_files = sbinfo->max_inodes;
1828 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 }
1830 /* else leave those fields 0 like simple_statfs */
1831 return 0;
1832}
1833
1834/*
1835 * File creation. Allocate an inode, and we're done..
1836 */
1837static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001838shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001840 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 int error = -ENOSPC;
1842
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001843 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001845 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001846 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001847 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001848 if (error) {
1849 if (error != -EOPNOTSUPP) {
1850 iput(inode);
1851 return error;
1852 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001853 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001854#ifdef CONFIG_TMPFS_POSIX_ACL
1855 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001856 if (error) {
1857 iput(inode);
1858 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001859 }
Al Viro718deb62009-12-16 19:35:36 -05001860#else
1861 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001862#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 dir->i_size += BOGO_DIRENT_SIZE;
1864 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1865 d_instantiate(dentry, inode);
1866 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
1868 return error;
1869}
1870
Al Viro18bb1db2011-07-26 01:41:39 -04001871static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
1873 int error;
1874
1875 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1876 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001877 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 return 0;
1879}
1880
Al Viro4acdaf22011-07-26 01:42:34 -04001881static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04001882 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883{
1884 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1885}
1886
1887/*
1888 * Link a file..
1889 */
1890static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1891{
1892 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001893 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
1895 /*
1896 * No ordinary (disk based) filesystem counts links as inodes;
1897 * but each new link needs a new dentry, pinning lowmem, and
1898 * tmpfs dentries cannot be pruned until they are unlinked.
1899 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001900 ret = shmem_reserve_inode(inode->i_sb);
1901 if (ret)
1902 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 dir->i_size += BOGO_DIRENT_SIZE;
1905 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001906 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04001907 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 dget(dentry); /* Extra pinning count for the created dentry */
1909 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001910out:
1911 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912}
1913
1914static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1915{
1916 struct inode *inode = dentry->d_inode;
1917
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001918 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1919 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921 dir->i_size -= BOGO_DIRENT_SIZE;
1922 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001923 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 dput(dentry); /* Undo the count from "create" - this does all the work */
1925 return 0;
1926}
1927
1928static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1929{
1930 if (!simple_empty(dentry))
1931 return -ENOTEMPTY;
1932
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001933 drop_nlink(dentry->d_inode);
1934 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return shmem_unlink(dir, dentry);
1936}
1937
1938/*
1939 * The VFS layer already does all the dentry stuff for rename,
1940 * we just have to decrement the usage count for the target if
1941 * it exists so that the VFS layer correctly free's it when it
1942 * gets overwritten.
1943 */
1944static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1945{
1946 struct inode *inode = old_dentry->d_inode;
1947 int they_are_dirs = S_ISDIR(inode->i_mode);
1948
1949 if (!simple_empty(new_dentry))
1950 return -ENOTEMPTY;
1951
1952 if (new_dentry->d_inode) {
1953 (void) shmem_unlink(new_dir, new_dentry);
1954 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001955 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001957 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001958 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 }
1960
1961 old_dir->i_size -= BOGO_DIRENT_SIZE;
1962 new_dir->i_size += BOGO_DIRENT_SIZE;
1963 old_dir->i_ctime = old_dir->i_mtime =
1964 new_dir->i_ctime = new_dir->i_mtime =
1965 inode->i_ctime = CURRENT_TIME;
1966 return 0;
1967}
1968
1969static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1970{
1971 int error;
1972 int len;
1973 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07001974 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 char *kaddr;
1976 struct shmem_inode_info *info;
1977
1978 len = strlen(symname) + 1;
1979 if (len > PAGE_CACHE_SIZE)
1980 return -ENAMETOOLONG;
1981
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001982 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (!inode)
1984 return -ENOSPC;
1985
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001986 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001987 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001988 if (error) {
1989 if (error != -EOPNOTSUPP) {
1990 iput(inode);
1991 return error;
1992 }
1993 error = 0;
1994 }
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 info = SHMEM_I(inode);
1997 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001998 if (len <= SHORT_SYMLINK_LEN) {
1999 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2000 if (!info->symlink) {
2001 iput(inode);
2002 return -ENOMEM;
2003 }
2004 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 } else {
2006 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2007 if (error) {
2008 iput(inode);
2009 return error;
2010 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002011 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002013 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002015 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002016 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002018 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 page_cache_release(page);
2020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 dir->i_size += BOGO_DIRENT_SIZE;
2022 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2023 d_instantiate(dentry, inode);
2024 dget(dentry);
2025 return 0;
2026}
2027
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002028static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002030 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002031 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002034static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035{
2036 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002037 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2038 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002039 if (page)
2040 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002041 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042}
2043
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002044static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045{
2046 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002047 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 kunmap(page);
2049 mark_page_accessed(page);
2050 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
2052}
2053
Eric Parisb09e0fa2011-05-24 17:12:39 -07002054#ifdef CONFIG_TMPFS_XATTR
2055/*
2056 * Superblocks without xattr inode operations may get some security.* xattr
2057 * support from the LSM "for free". As soon as we have any other xattrs
2058 * like ACLs, we also need to implement the security.* handlers at
2059 * filesystem level, though.
2060 */
2061
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002062/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002063 * Callback for security_inode_init_security() for acquiring xattrs.
2064 */
2065static int shmem_initxattrs(struct inode *inode,
2066 const struct xattr *xattr_array,
2067 void *fs_info)
2068{
2069 struct shmem_inode_info *info = SHMEM_I(inode);
2070 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002071 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002072 size_t len;
2073
2074 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002075 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002076 if (!new_xattr)
2077 return -ENOMEM;
2078
2079 len = strlen(xattr->name) + 1;
2080 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2081 GFP_KERNEL);
2082 if (!new_xattr->name) {
2083 kfree(new_xattr);
2084 return -ENOMEM;
2085 }
2086
2087 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2088 XATTR_SECURITY_PREFIX_LEN);
2089 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2090 xattr->name, len);
2091
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002092 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002093 }
2094
2095 return 0;
2096}
2097
Eric Parisb09e0fa2011-05-24 17:12:39 -07002098static const struct xattr_handler *shmem_xattr_handlers[] = {
2099#ifdef CONFIG_TMPFS_POSIX_ACL
2100 &generic_acl_access_handler,
2101 &generic_acl_default_handler,
2102#endif
2103 NULL
2104};
2105
2106static int shmem_xattr_validate(const char *name)
2107{
2108 struct { const char *prefix; size_t len; } arr[] = {
2109 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2110 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2111 };
2112 int i;
2113
2114 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2115 size_t preflen = arr[i].len;
2116 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2117 if (!name[preflen])
2118 return -EINVAL;
2119 return 0;
2120 }
2121 }
2122 return -EOPNOTSUPP;
2123}
2124
2125static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2126 void *buffer, size_t size)
2127{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002128 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002129 int err;
2130
2131 /*
2132 * If this is a request for a synthetic attribute in the system.*
2133 * namespace use the generic infrastructure to resolve a handler
2134 * for it via sb->s_xattr.
2135 */
2136 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2137 return generic_getxattr(dentry, name, buffer, size);
2138
2139 err = shmem_xattr_validate(name);
2140 if (err)
2141 return err;
2142
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002143 return simple_xattr_get(&info->xattrs, name, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002144}
2145
2146static int shmem_setxattr(struct dentry *dentry, const char *name,
2147 const void *value, size_t size, int flags)
2148{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002149 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002150 int err;
2151
2152 /*
2153 * If this is a request for a synthetic attribute in the system.*
2154 * namespace use the generic infrastructure to resolve a handler
2155 * for it via sb->s_xattr.
2156 */
2157 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2158 return generic_setxattr(dentry, name, value, size, flags);
2159
2160 err = shmem_xattr_validate(name);
2161 if (err)
2162 return err;
2163
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002164 return simple_xattr_set(&info->xattrs, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002165}
2166
2167static int shmem_removexattr(struct dentry *dentry, const char *name)
2168{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002169 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002170 int err;
2171
2172 /*
2173 * If this is a request for a synthetic attribute in the system.*
2174 * namespace use the generic infrastructure to resolve a handler
2175 * for it via sb->s_xattr.
2176 */
2177 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2178 return generic_removexattr(dentry, name);
2179
2180 err = shmem_xattr_validate(name);
2181 if (err)
2182 return err;
2183
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002184 return simple_xattr_remove(&info->xattrs, name);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002185}
2186
2187static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2188{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002189 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2190 return simple_xattr_list(&info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002191}
2192#endif /* CONFIG_TMPFS_XATTR */
2193
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002194static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002196 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002197#ifdef CONFIG_TMPFS_XATTR
2198 .setxattr = shmem_setxattr,
2199 .getxattr = shmem_getxattr,
2200 .listxattr = shmem_listxattr,
2201 .removexattr = shmem_removexattr,
2202#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203};
2204
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002205static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 .readlink = generic_readlink,
2207 .follow_link = shmem_follow_link,
2208 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002209#ifdef CONFIG_TMPFS_XATTR
2210 .setxattr = shmem_setxattr,
2211 .getxattr = shmem_getxattr,
2212 .listxattr = shmem_listxattr,
2213 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002214#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002215};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002216
David M. Grimes91828a42006-10-17 00:09:45 -07002217static struct dentry *shmem_get_parent(struct dentry *child)
2218{
2219 return ERR_PTR(-ESTALE);
2220}
2221
2222static int shmem_match(struct inode *ino, void *vfh)
2223{
2224 __u32 *fh = vfh;
2225 __u64 inum = fh[2];
2226 inum = (inum << 32) | fh[1];
2227 return ino->i_ino == inum && fh[0] == ino->i_generation;
2228}
2229
Christoph Hellwig480b1162007-10-21 16:42:13 -07002230static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2231 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002232{
David M. Grimes91828a42006-10-17 00:09:45 -07002233 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002234 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002235 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002236
Christoph Hellwig480b1162007-10-21 16:42:13 -07002237 if (fh_len < 3)
2238 return NULL;
2239
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002240 inum = fid->raw[2];
2241 inum = (inum << 32) | fid->raw[1];
2242
Christoph Hellwig480b1162007-10-21 16:42:13 -07002243 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2244 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002245 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002246 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002247 iput(inode);
2248 }
2249
Christoph Hellwig480b1162007-10-21 16:42:13 -07002250 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002251}
2252
Al Virob0b03822012-04-02 14:34:06 -04002253static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2254 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002255{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302256 if (*len < 3) {
2257 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07002258 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302259 }
David M. Grimes91828a42006-10-17 00:09:45 -07002260
Al Viro1d3382cb2010-10-23 15:19:20 -04002261 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002262 /* Unfortunately insert_inode_hash is not idempotent,
2263 * so as we hash inodes here rather than at creation
2264 * time, we need a lock to ensure we only try
2265 * to do it once
2266 */
2267 static DEFINE_SPINLOCK(lock);
2268 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002269 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002270 __insert_inode_hash(inode,
2271 inode->i_ino + inode->i_generation);
2272 spin_unlock(&lock);
2273 }
2274
2275 fh[0] = inode->i_generation;
2276 fh[1] = inode->i_ino;
2277 fh[2] = ((__u64)inode->i_ino) >> 32;
2278
2279 *len = 3;
2280 return 1;
2281}
2282
Christoph Hellwig39655162007-10-21 16:42:17 -07002283static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002284 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002285 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002286 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002287};
2288
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002289static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2290 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
2292 char *this_char, *value, *rest;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002293 uid_t uid;
2294 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002296 while (options != NULL) {
2297 this_char = options;
2298 for (;;) {
2299 /*
2300 * NUL-terminate this option: unfortunately,
2301 * mount options form a comma-separated list,
2302 * but mpol's nodelist may also contain commas.
2303 */
2304 options = strchr(options, ',');
2305 if (options == NULL)
2306 break;
2307 options++;
2308 if (!isdigit(*options)) {
2309 options[-1] = '\0';
2310 break;
2311 }
2312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 if (!*this_char)
2314 continue;
2315 if ((value = strchr(this_char,'=')) != NULL) {
2316 *value++ = 0;
2317 } else {
2318 printk(KERN_ERR
2319 "tmpfs: No value for mount option '%s'\n",
2320 this_char);
2321 return 1;
2322 }
2323
2324 if (!strcmp(this_char,"size")) {
2325 unsigned long long size;
2326 size = memparse(value,&rest);
2327 if (*rest == '%') {
2328 size <<= PAGE_SHIFT;
2329 size *= totalram_pages;
2330 do_div(size, 100);
2331 rest++;
2332 }
2333 if (*rest)
2334 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002335 sbinfo->max_blocks =
2336 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002338 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 if (*rest)
2340 goto bad_val;
2341 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002342 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 if (*rest)
2344 goto bad_val;
2345 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002346 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002348 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (*rest)
2350 goto bad_val;
2351 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002352 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002354 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 if (*rest)
2356 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002357 sbinfo->uid = make_kuid(current_user_ns(), uid);
2358 if (!uid_valid(sbinfo->uid))
2359 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002361 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002363 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 if (*rest)
2365 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002366 sbinfo->gid = make_kgid(current_user_ns(), gid);
2367 if (!gid_valid(sbinfo->gid))
2368 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002369 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002370 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002371 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 } else {
2373 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2374 this_char);
2375 return 1;
2376 }
2377 }
2378 return 0;
2379
2380bad_val:
2381 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2382 value, this_char);
2383 return 1;
2384
2385}
2386
2387static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2388{
2389 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002390 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002391 unsigned long inodes;
2392 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002394 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002395 return error;
2396
2397 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002398 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002399 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002400 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002401 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002402 goto out;
2403 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002404 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002405 * but we must separately disallow unlimited->limited, because
2406 * in that case we have no record of how much is already in use.
2407 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002408 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002409 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002410 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002411 goto out;
2412
2413 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002414 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002415 sbinfo->max_inodes = config.max_inodes;
2416 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002417
2418 mpol_put(sbinfo->mpol);
2419 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002420out:
2421 spin_unlock(&sbinfo->stat_lock);
2422 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002424
Al Viro34c80b12011-12-08 21:32:45 -05002425static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002426{
Al Viro34c80b12011-12-08 21:32:45 -05002427 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002428
2429 if (sbinfo->max_blocks != shmem_default_max_blocks())
2430 seq_printf(seq, ",size=%luk",
2431 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2432 if (sbinfo->max_inodes != shmem_default_max_inodes())
2433 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2434 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002435 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002436 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2437 seq_printf(seq, ",uid=%u",
2438 from_kuid_munged(&init_user_ns, sbinfo->uid));
2439 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2440 seq_printf(seq, ",gid=%u",
2441 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002442 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002443 return 0;
2444}
2445#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446
2447static void shmem_put_super(struct super_block *sb)
2448{
Hugh Dickins602586a2010-08-17 15:23:56 -07002449 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2450
2451 percpu_counter_destroy(&sbinfo->used_blocks);
2452 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 sb->s_fs_info = NULL;
2454}
2455
Kay Sievers2b2af542009-04-30 15:23:42 +02002456int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457{
2458 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002459 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002460 int err = -ENOMEM;
2461
2462 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002463 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002464 L1_CACHE_BYTES), GFP_KERNEL);
2465 if (!sbinfo)
2466 return -ENOMEM;
2467
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002468 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002469 sbinfo->uid = current_fsuid();
2470 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002471 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002473#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 /*
2475 * Per default we only allow half of the physical ram per
2476 * tmpfs instance, limiting inodes to one per page of lowmem;
2477 * but the internal instance is left unlimited.
2478 */
2479 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002480 sbinfo->max_blocks = shmem_default_max_blocks();
2481 sbinfo->max_inodes = shmem_default_max_inodes();
2482 if (shmem_parse_options(data, sbinfo, false)) {
2483 err = -EINVAL;
2484 goto failed;
2485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 }
David M. Grimes91828a42006-10-17 00:09:45 -07002487 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07002488 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489#else
2490 sb->s_flags |= MS_NOUSER;
2491#endif
2492
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002493 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002494 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2495 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002496 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002497
Hugh Dickins285b2c42011-08-03 16:21:20 -07002498 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 sb->s_blocksize = PAGE_CACHE_SIZE;
2500 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2501 sb->s_magic = TMPFS_MAGIC;
2502 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002503 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002504#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002505 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002506#endif
2507#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002508 sb->s_flags |= MS_POSIXACL;
2509#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002510
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002511 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 if (!inode)
2513 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002514 inode->i_uid = sbinfo->uid;
2515 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002516 sb->s_root = d_make_root(inode);
2517 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002518 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 return 0;
2520
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521failed:
2522 shmem_put_super(sb);
2523 return err;
2524}
2525
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002526static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
2528static struct inode *shmem_alloc_inode(struct super_block *sb)
2529{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002530 struct shmem_inode_info *info;
2531 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2532 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002534 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535}
2536
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002537static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002538{
2539 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002540 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2541}
2542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543static void shmem_destroy_inode(struct inode *inode)
2544{
Al Viro09208d12011-07-26 03:15:03 -04002545 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002547 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548}
2549
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002550static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002552 struct shmem_inode_info *info = foo;
2553 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554}
2555
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002556static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
2558 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2559 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002560 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 return 0;
2562}
2563
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002564static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002566 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567}
2568
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002569static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002571 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002573 .write_begin = shmem_write_begin,
2574 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002576 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002577 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578};
2579
Helge Deller15ad7cd2006-12-06 20:40:36 -08002580static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 .mmap = shmem_mmap,
2582#ifdef CONFIG_TMPFS
Hugh Dickinsf21f8062012-07-11 14:02:45 -07002583 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002584 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002585 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002586 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002587 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002588 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002589 .splice_read = shmem_file_splice_read,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002590 .splice_write = generic_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002591 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592#endif
2593};
2594
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002595static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002596 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002597#ifdef CONFIG_TMPFS_XATTR
2598 .setxattr = shmem_setxattr,
2599 .getxattr = shmem_getxattr,
2600 .listxattr = shmem_listxattr,
2601 .removexattr = shmem_removexattr,
2602#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603};
2604
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002605static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606#ifdef CONFIG_TMPFS
2607 .create = shmem_create,
2608 .lookup = simple_lookup,
2609 .link = shmem_link,
2610 .unlink = shmem_unlink,
2611 .symlink = shmem_symlink,
2612 .mkdir = shmem_mkdir,
2613 .rmdir = shmem_rmdir,
2614 .mknod = shmem_mknod,
2615 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002617#ifdef CONFIG_TMPFS_XATTR
2618 .setxattr = shmem_setxattr,
2619 .getxattr = shmem_getxattr,
2620 .listxattr = shmem_listxattr,
2621 .removexattr = shmem_removexattr,
2622#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002623#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002624 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002625#endif
2626};
2627
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002628static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002629#ifdef CONFIG_TMPFS_XATTR
2630 .setxattr = shmem_setxattr,
2631 .getxattr = shmem_getxattr,
2632 .listxattr = shmem_listxattr,
2633 .removexattr = shmem_removexattr,
2634#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002635#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002636 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002637#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638};
2639
Hugh Dickins759b9772007-03-05 00:30:28 -08002640static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 .alloc_inode = shmem_alloc_inode,
2642 .destroy_inode = shmem_destroy_inode,
2643#ifdef CONFIG_TMPFS
2644 .statfs = shmem_statfs,
2645 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002646 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647#endif
Al Viro1f895f72010-06-05 19:10:41 -04002648 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 .drop_inode = generic_delete_inode,
2650 .put_super = shmem_put_super,
2651};
2652
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002653static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002654 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655#ifdef CONFIG_NUMA
2656 .set_policy = shmem_set_policy,
2657 .get_policy = shmem_get_policy,
2658#endif
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07002659 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660};
2661
Al Viro3c26ff62010-07-25 11:46:36 +04002662static struct dentry *shmem_mount(struct file_system_type *fs_type,
2663 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664{
Al Viro3c26ff62010-07-25 11:46:36 +04002665 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666}
2667
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002668static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 .owner = THIS_MODULE,
2670 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002671 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 .kill_sb = kill_litter_super,
2673};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002675int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676{
2677 int error;
2678
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002679 error = bdi_init(&shmem_backing_dev_info);
2680 if (error)
2681 goto out4;
2682
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002683 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 if (error)
2685 goto out3;
2686
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002687 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (error) {
2689 printk(KERN_ERR "Could not register tmpfs\n");
2690 goto out2;
2691 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002692
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002693 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2694 shmem_fs_type.name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 if (IS_ERR(shm_mnt)) {
2696 error = PTR_ERR(shm_mnt);
2697 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2698 goto out1;
2699 }
2700 return 0;
2701
2702out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002703 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002705 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002707 bdi_destroy(&shmem_backing_dev_info);
2708out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 shm_mnt = ERR_PTR(error);
2710 return error;
2711}
Matt Mackall853ac432009-01-06 14:40:20 -08002712
2713#else /* !CONFIG_SHMEM */
2714
2715/*
2716 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2717 *
2718 * This is intended for small system where the benefits of the full
2719 * shmem code (swap-backed and resource-limited) are outweighed by
2720 * their complexity. On systems without swap this code should be
2721 * effectively equivalent, but much lighter weight.
2722 */
2723
2724#include <linux/ramfs.h>
2725
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002726static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002727 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002728 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002729 .kill_sb = kill_litter_super,
2730};
2731
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002732int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002733{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002734 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002735
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002736 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002737 BUG_ON(IS_ERR(shm_mnt));
2738
2739 return 0;
2740}
2741
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002742int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002743{
2744 return 0;
2745}
2746
Hugh Dickins3f96b792009-09-21 17:03:37 -07002747int shmem_lock(struct file *file, int lock, struct user_struct *user)
2748{
2749 return 0;
2750}
2751
Hugh Dickins24513262012-01-20 14:34:21 -08002752void shmem_unlock_mapping(struct address_space *mapping)
2753{
2754}
2755
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002756void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002757{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002758 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002759}
2760EXPORT_SYMBOL_GPL(shmem_truncate_range);
2761
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002762#define shmem_vm_ops generic_file_vm_ops
2763#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002764#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002765#define shmem_acct_size(flags, size) 0
2766#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002767
2768#endif /* CONFIG_SHMEM */
2769
2770/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771
Randy Dunlap46711812008-03-19 17:00:41 -07002772/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 * @name: name for dentry (to be seen in /proc/<pid>/maps
2775 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002776 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002778struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779{
2780 int error;
2781 struct file *file;
2782 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002783 struct path path;
2784 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 struct qstr this;
2786
2787 if (IS_ERR(shm_mnt))
2788 return (void *)shm_mnt;
2789
Hugh Dickins285b2c42011-08-03 16:21:20 -07002790 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 return ERR_PTR(-EINVAL);
2792
2793 if (shmem_acct_size(flags, size))
2794 return ERR_PTR(-ENOMEM);
2795
2796 error = -ENOMEM;
2797 this.name = name;
2798 this.len = strlen(name);
2799 this.hash = 0; /* will go */
2800 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002801 path.dentry = d_alloc(root, &this);
2802 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002804 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002807 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002809 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
Al Viro2c48b9c2009-08-09 00:52:35 +04002811 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002813 clear_nlink(inode); /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002814#ifndef CONFIG_MMU
2815 error = ramfs_nommu_expand_for_mapping(inode, size);
2816 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002817 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002818#endif
Al Viro4b42af82009-08-05 18:25:56 +04002819
2820 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002821 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002822 &shmem_file_operations);
2823 if (!file)
2824 goto put_dentry;
2825
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 return file;
2827
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002829 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830put_memory:
2831 shmem_unacct_size(flags, size);
2832 return ERR_PTR(error);
2833}
Keith Packard395e0dd2008-06-20 00:08:06 -07002834EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835
Randy Dunlap46711812008-03-19 17:00:41 -07002836/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2839 */
2840int shmem_zero_setup(struct vm_area_struct *vma)
2841{
2842 struct file *file;
2843 loff_t size = vma->vm_end - vma->vm_start;
2844
2845 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2846 if (IS_ERR(file))
2847 return PTR_ERR(file);
2848
2849 if (vma->vm_file)
2850 fput(vma->vm_file);
2851 vma->vm_file = file;
2852 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 return 0;
2854}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002855
2856/**
2857 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2858 * @mapping: the page's address_space
2859 * @index: the page index
2860 * @gfp: the page allocator flags to use if allocating
2861 *
2862 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2863 * with any new page allocations done using the specified allocation flags.
2864 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2865 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2866 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2867 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07002868 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2869 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002870 */
2871struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2872 pgoff_t index, gfp_t gfp)
2873{
Hugh Dickins68da9f02011-07-25 17:12:34 -07002874#ifdef CONFIG_SHMEM
2875 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002876 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07002877 int error;
2878
2879 BUG_ON(mapping->a_ops != &shmem_aops);
2880 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2881 if (error)
2882 page = ERR_PTR(error);
2883 else
2884 unlock_page(page);
2885 return page;
2886#else
2887 /*
2888 * The tiny !SHMEM case uses ramfs without swap
2889 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002890 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002891#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002892}
2893EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);