blob: 03f9ba8fb8e543d96d2fad317a2a19126cf99222 [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{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800914 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800916 /* Create a pseudo vma that just contains the policy */
917 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700918 /* Bias interleave by inode number to distribute better across nodes */
919 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800920 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700921 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700922
Mel Gorman18a2f372012-12-05 14:01:41 -0800923 page = swapin_readahead(swap, gfp, &pvma, 0);
924
925 /* Drop reference taken by mpol_shared_policy_lookup() */
926 mpol_cond_put(pvma.vm_policy);
927
928 return page;
929}
930
931static struct page *shmem_alloc_page(gfp_t gfp,
932 struct shmem_inode_info *info, pgoff_t index)
933{
934 struct vm_area_struct pvma;
935 struct page *page;
936
937 /* Create a pseudo vma that just contains the policy */
938 pvma.vm_start = 0;
939 /* Bias interleave by inode number to distribute better across nodes */
940 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
941 pvma.vm_ops = NULL;
942 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
943
944 page = alloc_page_vma(gfp, &pvma, 0);
945
946 /* Drop reference taken by mpol_shared_policy_lookup() */
947 mpol_cond_put(pvma.vm_policy);
948
949 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800951#else /* !CONFIG_NUMA */
952#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700953static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800954{
955}
956#endif /* CONFIG_TMPFS */
957
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700958static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
959 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700961 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
Hugh Dickins02098fe2008-02-04 22:28:42 -0800964static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700965 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000967 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800969#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700971#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
972static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
973{
974 return NULL;
975}
976#endif
977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700979 * When a page is moved from swapcache to shmem filecache (either by the
980 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
981 * shmem_unuse_inode()), it may have been read in earlier from swap, in
982 * ignorance of the mapping it belongs to. If that mapping has special
983 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
984 * we may need to copy to a suitable page before moving to filecache.
985 *
986 * In a future release, this may well be extended to respect cpuset and
987 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
988 * but for now it is a simple matter of zone.
989 */
990static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
991{
992 return page_zonenum(page) > gfp_zone(gfp);
993}
994
995static int shmem_replace_page(struct page **pagep, gfp_t gfp,
996 struct shmem_inode_info *info, pgoff_t index)
997{
998 struct page *oldpage, *newpage;
999 struct address_space *swap_mapping;
1000 pgoff_t swap_index;
1001 int error;
1002
1003 oldpage = *pagep;
1004 swap_index = page_private(oldpage);
1005 swap_mapping = page_mapping(oldpage);
1006
1007 /*
1008 * We have arrived here because our zones are constrained, so don't
1009 * limit chance of success by further cpuset and node constraints.
1010 */
1011 gfp &= ~GFP_CONSTRAINT_MASK;
1012 newpage = shmem_alloc_page(gfp, info, index);
1013 if (!newpage)
1014 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001015
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001016 page_cache_get(newpage);
1017 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001018 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001019
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001020 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001021 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001022 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001023 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001024 SetPageSwapCache(newpage);
1025
1026 /*
1027 * Our caller will very soon move newpage out of swapcache, but it's
1028 * a nice clean interface for us to replace oldpage by newpage there.
1029 */
1030 spin_lock_irq(&swap_mapping->tree_lock);
1031 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1032 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001033 if (!error) {
1034 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1035 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1036 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001037 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001038
Hugh Dickins0142ef62012-06-07 14:21:09 -07001039 if (unlikely(error)) {
1040 /*
1041 * Is this possible? I think not, now that our callers check
1042 * both PageSwapCache and page_private after getting page lock;
1043 * but be defensive. Reverse old to newpage for clear and free.
1044 */
1045 oldpage = newpage;
1046 } else {
1047 mem_cgroup_replace_page_cache(oldpage, newpage);
1048 lru_cache_add_anon(newpage);
1049 *pagep = newpage;
1050 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001051
1052 ClearPageSwapCache(oldpage);
1053 set_page_private(oldpage, 0);
1054
1055 unlock_page(oldpage);
1056 page_cache_release(oldpage);
1057 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001058 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001059}
1060
1061/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001062 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 *
1064 * If we allocate a new one we do not mark it dirty. That's up to the
1065 * vm. If we swap it in we mark it dirty since we also free the swap
1066 * entry since a page cannot live in both the swap and page cache
1067 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001068static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001069 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
1071 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001072 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001074 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 swp_entry_t swap;
1076 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001077 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001078 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001080 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001083 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001084 page = find_lock_page(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001085 if (radix_tree_exceptional_entry(page)) {
1086 swap = radix_to_swp_entry(page);
1087 page = NULL;
1088 }
1089
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001090 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001091 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1092 error = -EINVAL;
1093 goto failed;
1094 }
1095
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001096 /* fallocated page? */
1097 if (page && !PageUptodate(page)) {
1098 if (sgp != SGP_READ)
1099 goto clear;
1100 unlock_page(page);
1101 page_cache_release(page);
1102 page = NULL;
1103 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001104 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001105 *pagep = page;
1106 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001110 * Fast cache lookup did not find it:
1111 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001113 info = SHMEM_I(inode);
1114 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (swap.val) {
1117 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001118 page = lookup_swap_cache(swap);
1119 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001120 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001121 if (fault_type)
1122 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001123 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001124 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001125 error = -ENOMEM;
1126 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
1129
1130 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001131 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001132 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001133 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001134 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001135 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001136 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001137 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001139 goto failed;
1140 }
1141 wait_on_page_writeback(page);
1142
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001143 if (shmem_should_replace_page(page, gfp)) {
1144 error = shmem_replace_page(&page, gfp, info, index);
1145 if (error)
1146 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
1148
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001149 error = mem_cgroup_cache_charge(page, current->mm,
1150 gfp & GFP_RECLAIM_MASK);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001151 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001152 error = shmem_add_to_page_cache(page, mapping, index,
1153 gfp, swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001154 /*
1155 * We already confirmed swap under page lock, and make
1156 * no memory allocation here, so usually no possibility
1157 * of error; but free_swap_and_cache() only trylocks a
1158 * page, so it is just possible that the entry has been
1159 * truncated or holepunched since swap was confirmed.
1160 * shmem_undo_range() will have done some of the
1161 * unaccounting, now delete_from_swap_cache() will do
1162 * the rest (including mem_cgroup_uncharge_swapcache).
1163 * Reset swap.val? No, leave it so "failed" goes back to
1164 * "repeat": reading a hole and writing should succeed.
1165 */
1166 if (error)
1167 delete_from_swap_cache(page);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001168 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001169 if (error)
1170 goto failed;
1171
1172 spin_lock(&info->lock);
1173 info->swapped--;
1174 shmem_recalc_inode(inode);
1175 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001176
Hugh Dickins27ab7002011-07-25 17:12:36 -07001177 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001178 set_page_dirty(page);
1179 swap_free(swap);
1180
Hugh Dickins54af60422011-08-03 16:21:24 -07001181 } else {
1182 if (shmem_acct_block(info->flags)) {
1183 error = -ENOSPC;
1184 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001186 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001187 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001188 sbinfo->max_blocks) >= 0) {
1189 error = -ENOSPC;
1190 goto unacct;
1191 }
Tim Chen7e496292010-08-09 17:19:05 -07001192 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Hugh Dickins54af60422011-08-03 16:21:24 -07001195 page = shmem_alloc_page(gfp, info, index);
1196 if (!page) {
1197 error = -ENOMEM;
1198 goto decused;
1199 }
1200
1201 SetPageSwapBacked(page);
1202 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001203 error = mem_cgroup_cache_charge(page, current->mm,
1204 gfp & GFP_RECLAIM_MASK);
Hugh Dickins54af60422011-08-03 16:21:24 -07001205 if (error)
1206 goto decused;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001207 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1208 if (!error) {
1209 error = shmem_add_to_page_cache(page, mapping, index,
1210 gfp, NULL);
1211 radix_tree_preload_end();
1212 }
1213 if (error) {
1214 mem_cgroup_uncharge_cache_page(page);
1215 goto decused;
1216 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001217 lru_cache_add_anon(page);
1218
1219 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001221 inode->i_blocks += BLOCKS_PER_PAGE;
1222 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001224 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001225
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001226 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001227 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1228 */
1229 if (sgp == SGP_FALLOC)
1230 sgp = SGP_WRITE;
1231clear:
1232 /*
1233 * Let SGP_WRITE caller clear ends if write does not fill page;
1234 * but SGP_FALLOC on a page fallocated earlier must initialize
1235 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001236 */
1237 if (sgp != SGP_WRITE) {
1238 clear_highpage(page);
1239 flush_dcache_page(page);
1240 SetPageUptodate(page);
1241 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001242 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001243 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001244 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001245
Hugh Dickins54af60422011-08-03 16:21:24 -07001246 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001247 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001248 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1249 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001250 if (alloced)
1251 goto trunc;
1252 else
1253 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001254 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001255 *pagep = page;
1256 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001257
Nick Piggind0217ac2007-07-19 01:47:03 -07001258 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001259 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001261trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001262 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001263 ClearPageDirty(page);
1264 delete_from_page_cache(page);
1265 spin_lock(&info->lock);
1266 info->alloced--;
1267 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001269decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001270 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001271 if (sbinfo->max_blocks)
1272 percpu_counter_add(&sbinfo->used_blocks, -1);
1273unacct:
1274 shmem_unacct_blocks(info->flags, 1);
1275failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001276 if (swap.val && error != -EINVAL &&
1277 !shmem_confirm_swap(mapping, index, swap))
1278 error = -EEXIST;
1279unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001280 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001281 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001282 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001283 }
1284 if (error == -ENOSPC && !once++) {
1285 info = SHMEM_I(inode);
1286 spin_lock(&info->lock);
1287 shmem_recalc_inode(inode);
1288 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001289 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001290 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001291 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001292 goto repeat;
1293 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294}
1295
1296static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1297{
1298 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1299 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001300 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1303 if (error)
1304 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001305
Ying Han456f9982011-05-26 16:25:38 -07001306 if (ret & VM_FAULT_MAJOR) {
1307 count_vm_event(PGMAJFAULT);
1308 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1309 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001310 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311}
1312
1313#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001314static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001316 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1317 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001320static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1321 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001323 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1324 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001326 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1327 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328}
1329#endif
1330
1331int shmem_lock(struct file *file, int lock, struct user_struct *user)
1332{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001333 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 struct shmem_inode_info *info = SHMEM_I(inode);
1335 int retval = -ENOMEM;
1336
1337 spin_lock(&info->lock);
1338 if (lock && !(info->flags & VM_LOCKED)) {
1339 if (!user_shm_lock(inode->i_size, user))
1340 goto out_nomem;
1341 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001342 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
1344 if (!lock && (info->flags & VM_LOCKED) && user) {
1345 user_shm_unlock(inode->i_size, user);
1346 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001347 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
1349 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351out_nomem:
1352 spin_unlock(&info->lock);
1353 return retval;
1354}
1355
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001356static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
1358 file_accessed(file);
1359 vma->vm_ops = &shmem_vm_ops;
1360 return 0;
1361}
1362
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001363static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001364 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365{
1366 struct inode *inode;
1367 struct shmem_inode_info *info;
1368 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1369
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001370 if (shmem_reserve_inode(sb))
1371 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
1373 inode = new_inode(sb);
1374 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001375 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001376 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1379 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001380 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 info = SHMEM_I(inode);
1382 memset(info, 0, (char *)inode - (char *)info);
1383 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001384 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001386 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001387 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 switch (mode & S_IFMT) {
1390 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001391 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 init_special_inode(inode, mode, dev);
1393 break;
1394 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001395 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 inode->i_op = &shmem_inode_operations;
1397 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001398 mpol_shared_policy_init(&info->policy,
1399 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 break;
1401 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001402 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 /* Some things misbehave if size == 0 on a directory */
1404 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1405 inode->i_op = &shmem_dir_inode_operations;
1406 inode->i_fop = &simple_dir_operations;
1407 break;
1408 case S_IFLNK:
1409 /*
1410 * Must not load anything in the rbtree,
1411 * mpol_free_shared_policy will not be called.
1412 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001413 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 break;
1415 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001416 } else
1417 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 return inode;
1419}
1420
1421#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001422static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001423static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001425#ifdef CONFIG_TMPFS_XATTR
1426static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1427#else
1428#define shmem_initxattrs NULL
1429#endif
1430
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001432shmem_write_begin(struct file *file, struct address_space *mapping,
1433 loff_t pos, unsigned len, unsigned flags,
1434 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Nick Piggin800d15a2007-10-16 01:25:03 -07001436 struct inode *inode = mapping->host;
1437 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001438 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1439}
1440
1441static int
1442shmem_write_end(struct file *file, struct address_space *mapping,
1443 loff_t pos, unsigned len, unsigned copied,
1444 struct page *page, void *fsdata)
1445{
1446 struct inode *inode = mapping->host;
1447
Hugh Dickinsd3602442008-02-04 22:28:44 -08001448 if (pos + copied > inode->i_size)
1449 i_size_write(inode, pos + copied);
1450
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001451 if (!PageUptodate(page)) {
1452 if (copied < PAGE_CACHE_SIZE) {
1453 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1454 zero_user_segments(page, 0, from,
1455 from + copied, PAGE_CACHE_SIZE);
1456 }
1457 SetPageUptodate(page);
1458 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001459 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001460 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001461 page_cache_release(page);
1462
Nick Piggin800d15a2007-10-16 01:25:03 -07001463 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464}
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1467{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001468 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001470 pgoff_t index;
1471 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001472 enum sgp_type sgp = SGP_READ;
1473
1474 /*
1475 * Might this read be for a stacking filesystem? Then when reading
1476 * holes of a sparse file, we actually need to allocate those pages,
1477 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1478 */
1479 if (segment_eq(get_fs(), KERNEL_DS))
1480 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
1482 index = *ppos >> PAGE_CACHE_SHIFT;
1483 offset = *ppos & ~PAGE_CACHE_MASK;
1484
1485 for (;;) {
1486 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001487 pgoff_t end_index;
1488 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 loff_t i_size = i_size_read(inode);
1490
1491 end_index = i_size >> PAGE_CACHE_SHIFT;
1492 if (index > end_index)
1493 break;
1494 if (index == end_index) {
1495 nr = i_size & ~PAGE_CACHE_MASK;
1496 if (nr <= offset)
1497 break;
1498 }
1499
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001500 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 if (desc->error) {
1502 if (desc->error == -EINVAL)
1503 desc->error = 0;
1504 break;
1505 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001506 if (page)
1507 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
1509 /*
1510 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001511 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 */
1513 nr = PAGE_CACHE_SIZE;
1514 i_size = i_size_read(inode);
1515 end_index = i_size >> PAGE_CACHE_SHIFT;
1516 if (index == end_index) {
1517 nr = i_size & ~PAGE_CACHE_MASK;
1518 if (nr <= offset) {
1519 if (page)
1520 page_cache_release(page);
1521 break;
1522 }
1523 }
1524 nr -= offset;
1525
1526 if (page) {
1527 /*
1528 * If users can be writing to this page using arbitrary
1529 * virtual addresses, take care about potential aliasing
1530 * before reading the page on the kernel side.
1531 */
1532 if (mapping_writably_mapped(mapping))
1533 flush_dcache_page(page);
1534 /*
1535 * Mark the page accessed if we read the beginning.
1536 */
1537 if (!offset)
1538 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001539 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001541 page_cache_get(page);
1542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
1544 /*
1545 * Ok, we have the page, and it's up-to-date, so
1546 * now we can copy it to user space...
1547 *
1548 * The actor routine returns how many bytes were actually used..
1549 * NOTE! This may not be the same as how much of a user buffer
1550 * we filled up (we may be padding etc), so we can only update
1551 * "pos" here (the actor routine has to update the user buffer
1552 * pointers and the remaining count).
1553 */
1554 ret = actor(desc, page, offset, nr);
1555 offset += ret;
1556 index += offset >> PAGE_CACHE_SHIFT;
1557 offset &= ~PAGE_CACHE_MASK;
1558
1559 page_cache_release(page);
1560 if (ret != nr || !desc->count)
1561 break;
1562
1563 cond_resched();
1564 }
1565
1566 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1567 file_accessed(filp);
1568}
1569
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001570static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1571 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001573 struct file *filp = iocb->ki_filp;
1574 ssize_t retval;
1575 unsigned long seg;
1576 size_t count;
1577 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001579 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1580 if (retval)
1581 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001583 for (seg = 0; seg < nr_segs; seg++) {
1584 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001586 desc.written = 0;
1587 desc.arg.buf = iov[seg].iov_base;
1588 desc.count = iov[seg].iov_len;
1589 if (desc.count == 0)
1590 continue;
1591 desc.error = 0;
1592 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1593 retval += desc.written;
1594 if (desc.error) {
1595 retval = retval ?: desc.error;
1596 break;
1597 }
1598 if (desc.count > 0)
1599 break;
1600 }
1601 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
1603
Hugh Dickins708e3502011-07-25 17:12:32 -07001604static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1605 struct pipe_inode_info *pipe, size_t len,
1606 unsigned int flags)
1607{
1608 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001609 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001610 unsigned int loff, nr_pages, req_pages;
1611 struct page *pages[PIPE_DEF_BUFFERS];
1612 struct partial_page partial[PIPE_DEF_BUFFERS];
1613 struct page *page;
1614 pgoff_t index, end_index;
1615 loff_t isize, left;
1616 int error, page_nr;
1617 struct splice_pipe_desc spd = {
1618 .pages = pages,
1619 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001620 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001621 .flags = flags,
1622 .ops = &page_cache_pipe_buf_ops,
1623 .spd_release = spd_release_page,
1624 };
1625
Hugh Dickins71f0e072011-07-25 17:12:33 -07001626 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001627 if (unlikely(*ppos >= isize))
1628 return 0;
1629
1630 left = isize - *ppos;
1631 if (unlikely(left < len))
1632 len = left;
1633
1634 if (splice_grow_spd(pipe, &spd))
1635 return -ENOMEM;
1636
1637 index = *ppos >> PAGE_CACHE_SHIFT;
1638 loff = *ppos & ~PAGE_CACHE_MASK;
1639 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1640 nr_pages = min(req_pages, pipe->buffers);
1641
Hugh Dickins708e3502011-07-25 17:12:32 -07001642 spd.nr_pages = find_get_pages_contig(mapping, index,
1643 nr_pages, spd.pages);
1644 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001645 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001646
Hugh Dickins708e3502011-07-25 17:12:32 -07001647 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001648 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1649 if (error)
1650 break;
1651 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001652 spd.pages[spd.nr_pages++] = page;
1653 index++;
1654 }
1655
Hugh Dickins708e3502011-07-25 17:12:32 -07001656 index = *ppos >> PAGE_CACHE_SHIFT;
1657 nr_pages = spd.nr_pages;
1658 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001659
Hugh Dickins708e3502011-07-25 17:12:32 -07001660 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1661 unsigned int this_len;
1662
1663 if (!len)
1664 break;
1665
Hugh Dickins708e3502011-07-25 17:12:32 -07001666 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1667 page = spd.pages[page_nr];
1668
Hugh Dickins71f0e072011-07-25 17:12:33 -07001669 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001670 error = shmem_getpage(inode, index, &page,
1671 SGP_CACHE, NULL);
1672 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001673 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001674 unlock_page(page);
1675 page_cache_release(spd.pages[page_nr]);
1676 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001677 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001678
1679 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001680 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1681 if (unlikely(!isize || index > end_index))
1682 break;
1683
Hugh Dickins708e3502011-07-25 17:12:32 -07001684 if (end_index == index) {
1685 unsigned int plen;
1686
Hugh Dickins708e3502011-07-25 17:12:32 -07001687 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1688 if (plen <= loff)
1689 break;
1690
Hugh Dickins708e3502011-07-25 17:12:32 -07001691 this_len = min(this_len, plen - loff);
1692 len = this_len;
1693 }
1694
1695 spd.partial[page_nr].offset = loff;
1696 spd.partial[page_nr].len = this_len;
1697 len -= this_len;
1698 loff = 0;
1699 spd.nr_pages++;
1700 index++;
1701 }
1702
Hugh Dickins708e3502011-07-25 17:12:32 -07001703 while (page_nr < nr_pages)
1704 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001705
1706 if (spd.nr_pages)
1707 error = splice_to_pipe(pipe, &spd);
1708
Eric Dumazet047fe362012-06-12 15:24:40 +02001709 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001710
1711 if (error > 0) {
1712 *ppos += error;
1713 file_accessed(in);
1714 }
1715 return error;
1716}
1717
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001718/*
1719 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1720 */
1721static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1722 pgoff_t index, pgoff_t end, int origin)
1723{
1724 struct page *page;
1725 struct pagevec pvec;
1726 pgoff_t indices[PAGEVEC_SIZE];
1727 bool done = false;
1728 int i;
1729
1730 pagevec_init(&pvec, 0);
1731 pvec.nr = 1; /* start small: we may be there already */
1732 while (!done) {
1733 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1734 pvec.nr, pvec.pages, indices);
1735 if (!pvec.nr) {
1736 if (origin == SEEK_DATA)
1737 index = end;
1738 break;
1739 }
1740 for (i = 0; i < pvec.nr; i++, index++) {
1741 if (index < indices[i]) {
1742 if (origin == SEEK_HOLE) {
1743 done = true;
1744 break;
1745 }
1746 index = indices[i];
1747 }
1748 page = pvec.pages[i];
1749 if (page && !radix_tree_exceptional_entry(page)) {
1750 if (!PageUptodate(page))
1751 page = NULL;
1752 }
1753 if (index >= end ||
1754 (page && origin == SEEK_DATA) ||
1755 (!page && origin == SEEK_HOLE)) {
1756 done = true;
1757 break;
1758 }
1759 }
1760 shmem_deswap_pagevec(&pvec);
1761 pagevec_release(&pvec);
1762 pvec.nr = PAGEVEC_SIZE;
1763 cond_resched();
1764 }
1765 return index;
1766}
1767
1768static loff_t shmem_file_llseek(struct file *file, loff_t offset, int origin)
1769{
1770 struct address_space *mapping = file->f_mapping;
1771 struct inode *inode = mapping->host;
1772 pgoff_t start, end;
1773 loff_t new_offset;
1774
1775 if (origin != SEEK_DATA && origin != SEEK_HOLE)
1776 return generic_file_llseek_size(file, offset, origin,
1777 MAX_LFS_FILESIZE, i_size_read(inode));
1778 mutex_lock(&inode->i_mutex);
1779 /* We're holding i_mutex so we can access i_size directly */
1780
1781 if (offset < 0)
1782 offset = -EINVAL;
1783 else if (offset >= inode->i_size)
1784 offset = -ENXIO;
1785 else {
1786 start = offset >> PAGE_CACHE_SHIFT;
1787 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1788 new_offset = shmem_seek_hole_data(mapping, start, end, origin);
1789 new_offset <<= PAGE_CACHE_SHIFT;
1790 if (new_offset > offset) {
1791 if (new_offset < inode->i_size)
1792 offset = new_offset;
1793 else if (origin == SEEK_DATA)
1794 offset = -ENXIO;
1795 else
1796 offset = inode->i_size;
1797 }
1798 }
1799
1800 if (offset >= 0 && offset != file->f_pos) {
1801 file->f_pos = offset;
1802 file->f_version = 0;
1803 }
1804 mutex_unlock(&inode->i_mutex);
1805 return offset;
1806}
1807
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001808static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1809 loff_t len)
1810{
1811 struct inode *inode = file->f_path.dentry->d_inode;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001812 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001813 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001814 pgoff_t start, index, end;
1815 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001816
1817 mutex_lock(&inode->i_mutex);
1818
1819 if (mode & FALLOC_FL_PUNCH_HOLE) {
1820 struct address_space *mapping = file->f_mapping;
1821 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1822 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1823
1824 if ((u64)unmap_end > (u64)unmap_start)
1825 unmap_mapping_range(mapping, unmap_start,
1826 1 + unmap_end - unmap_start, 0);
1827 shmem_truncate_range(inode, offset, offset + len - 1);
1828 /* No need to unmap again: hole-punching leaves COWed pages */
1829 error = 0;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001830 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001831 }
1832
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001833 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1834 error = inode_newsize_ok(inode, offset + len);
1835 if (error)
1836 goto out;
1837
1838 start = offset >> PAGE_CACHE_SHIFT;
1839 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1840 /* Try to avoid a swapstorm if len is impossible to satisfy */
1841 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1842 error = -ENOSPC;
1843 goto out;
1844 }
1845
Hugh Dickins1aac1402012-05-29 15:06:42 -07001846 shmem_falloc.start = start;
1847 shmem_falloc.next = start;
1848 shmem_falloc.nr_falloced = 0;
1849 shmem_falloc.nr_unswapped = 0;
1850 spin_lock(&inode->i_lock);
1851 inode->i_private = &shmem_falloc;
1852 spin_unlock(&inode->i_lock);
1853
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001854 for (index = start; index < end; index++) {
1855 struct page *page;
1856
1857 /*
1858 * Good, the fallocate(2) manpage permits EINTR: we may have
1859 * been interrupted because we are using up too much memory.
1860 */
1861 if (signal_pending(current))
1862 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001863 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1864 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001865 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001866 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001867 NULL);
1868 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001869 /* Remove the !PageUptodate pages we added */
1870 shmem_undo_range(inode,
1871 (loff_t)start << PAGE_CACHE_SHIFT,
1872 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001873 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001874 }
1875
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001876 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07001877 * Inform shmem_writepage() how far we have reached.
1878 * No need for lock or barrier: we have the page lock.
1879 */
1880 shmem_falloc.next++;
1881 if (!PageUptodate(page))
1882 shmem_falloc.nr_falloced++;
1883
1884 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001885 * If !PageUptodate, leave it that way so that freeable pages
1886 * can be recognized if we need to rollback on error later.
1887 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001888 * than free the pages we are allocating (and SGP_CACHE pages
1889 * might still be clean: we now need to mark those dirty too).
1890 */
1891 set_page_dirty(page);
1892 unlock_page(page);
1893 page_cache_release(page);
1894 cond_resched();
1895 }
1896
1897 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1898 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001899 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001900undone:
1901 spin_lock(&inode->i_lock);
1902 inode->i_private = NULL;
1903 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001904out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001905 mutex_unlock(&inode->i_mutex);
1906 return error;
1907}
1908
David Howells726c3342006-06-23 02:02:58 -07001909static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910{
David Howells726c3342006-06-23 02:02:58 -07001911 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
1913 buf->f_type = TMPFS_MAGIC;
1914 buf->f_bsize = PAGE_CACHE_SIZE;
1915 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001916 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001918 buf->f_bavail =
1919 buf->f_bfree = sbinfo->max_blocks -
1920 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001921 }
1922 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 buf->f_files = sbinfo->max_inodes;
1924 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 }
1926 /* else leave those fields 0 like simple_statfs */
1927 return 0;
1928}
1929
1930/*
1931 * File creation. Allocate an inode, and we're done..
1932 */
1933static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001934shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001936 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 int error = -ENOSPC;
1938
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001939 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001941 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001942 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001943 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001944 if (error) {
1945 if (error != -EOPNOTSUPP) {
1946 iput(inode);
1947 return error;
1948 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001949 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001950#ifdef CONFIG_TMPFS_POSIX_ACL
1951 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001952 if (error) {
1953 iput(inode);
1954 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001955 }
Al Viro718deb62009-12-16 19:35:36 -05001956#else
1957 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001958#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 dir->i_size += BOGO_DIRENT_SIZE;
1960 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1961 d_instantiate(dentry, inode);
1962 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 }
1964 return error;
1965}
1966
Al Viro18bb1db2011-07-26 01:41:39 -04001967static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968{
1969 int error;
1970
1971 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1972 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001973 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 return 0;
1975}
1976
Al Viro4acdaf22011-07-26 01:42:34 -04001977static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04001978 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979{
1980 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1981}
1982
1983/*
1984 * Link a file..
1985 */
1986static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1987{
1988 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001989 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991 /*
1992 * No ordinary (disk based) filesystem counts links as inodes;
1993 * but each new link needs a new dentry, pinning lowmem, and
1994 * tmpfs dentries cannot be pruned until they are unlinked.
1995 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001996 ret = shmem_reserve_inode(inode->i_sb);
1997 if (ret)
1998 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 dir->i_size += BOGO_DIRENT_SIZE;
2001 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002002 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002003 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 dget(dentry); /* Extra pinning count for the created dentry */
2005 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002006out:
2007 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008}
2009
2010static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2011{
2012 struct inode *inode = dentry->d_inode;
2013
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002014 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2015 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
2017 dir->i_size -= BOGO_DIRENT_SIZE;
2018 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002019 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 dput(dentry); /* Undo the count from "create" - this does all the work */
2021 return 0;
2022}
2023
2024static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2025{
2026 if (!simple_empty(dentry))
2027 return -ENOTEMPTY;
2028
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002029 drop_nlink(dentry->d_inode);
2030 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 return shmem_unlink(dir, dentry);
2032}
2033
2034/*
2035 * The VFS layer already does all the dentry stuff for rename,
2036 * we just have to decrement the usage count for the target if
2037 * it exists so that the VFS layer correctly free's it when it
2038 * gets overwritten.
2039 */
2040static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2041{
2042 struct inode *inode = old_dentry->d_inode;
2043 int they_are_dirs = S_ISDIR(inode->i_mode);
2044
2045 if (!simple_empty(new_dentry))
2046 return -ENOTEMPTY;
2047
2048 if (new_dentry->d_inode) {
2049 (void) shmem_unlink(new_dir, new_dentry);
2050 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002051 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002053 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002054 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 }
2056
2057 old_dir->i_size -= BOGO_DIRENT_SIZE;
2058 new_dir->i_size += BOGO_DIRENT_SIZE;
2059 old_dir->i_ctime = old_dir->i_mtime =
2060 new_dir->i_ctime = new_dir->i_mtime =
2061 inode->i_ctime = CURRENT_TIME;
2062 return 0;
2063}
2064
2065static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2066{
2067 int error;
2068 int len;
2069 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002070 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 char *kaddr;
2072 struct shmem_inode_info *info;
2073
2074 len = strlen(symname) + 1;
2075 if (len > PAGE_CACHE_SIZE)
2076 return -ENAMETOOLONG;
2077
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002078 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (!inode)
2080 return -ENOSPC;
2081
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002082 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002083 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002084 if (error) {
2085 if (error != -EOPNOTSUPP) {
2086 iput(inode);
2087 return error;
2088 }
2089 error = 0;
2090 }
2091
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 info = SHMEM_I(inode);
2093 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002094 if (len <= SHORT_SYMLINK_LEN) {
2095 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2096 if (!info->symlink) {
2097 iput(inode);
2098 return -ENOMEM;
2099 }
2100 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 } else {
2102 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2103 if (error) {
2104 iput(inode);
2105 return error;
2106 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002107 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002109 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002111 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002112 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002114 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 page_cache_release(page);
2116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 dir->i_size += BOGO_DIRENT_SIZE;
2118 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2119 d_instantiate(dentry, inode);
2120 dget(dentry);
2121 return 0;
2122}
2123
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002124static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002126 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002127 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128}
2129
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002130static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131{
2132 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002133 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2134 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002135 if (page)
2136 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002137 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138}
2139
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002140static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141{
2142 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002143 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 kunmap(page);
2145 mark_page_accessed(page);
2146 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 }
2148}
2149
Eric Parisb09e0fa2011-05-24 17:12:39 -07002150#ifdef CONFIG_TMPFS_XATTR
2151/*
2152 * Superblocks without xattr inode operations may get some security.* xattr
2153 * support from the LSM "for free". As soon as we have any other xattrs
2154 * like ACLs, we also need to implement the security.* handlers at
2155 * filesystem level, though.
2156 */
2157
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002158/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002159 * Callback for security_inode_init_security() for acquiring xattrs.
2160 */
2161static int shmem_initxattrs(struct inode *inode,
2162 const struct xattr *xattr_array,
2163 void *fs_info)
2164{
2165 struct shmem_inode_info *info = SHMEM_I(inode);
2166 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002167 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002168 size_t len;
2169
2170 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002171 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002172 if (!new_xattr)
2173 return -ENOMEM;
2174
2175 len = strlen(xattr->name) + 1;
2176 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2177 GFP_KERNEL);
2178 if (!new_xattr->name) {
2179 kfree(new_xattr);
2180 return -ENOMEM;
2181 }
2182
2183 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2184 XATTR_SECURITY_PREFIX_LEN);
2185 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2186 xattr->name, len);
2187
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002188 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002189 }
2190
2191 return 0;
2192}
2193
Eric Parisb09e0fa2011-05-24 17:12:39 -07002194static const struct xattr_handler *shmem_xattr_handlers[] = {
2195#ifdef CONFIG_TMPFS_POSIX_ACL
2196 &generic_acl_access_handler,
2197 &generic_acl_default_handler,
2198#endif
2199 NULL
2200};
2201
2202static int shmem_xattr_validate(const char *name)
2203{
2204 struct { const char *prefix; size_t len; } arr[] = {
2205 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2206 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2207 };
2208 int i;
2209
2210 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2211 size_t preflen = arr[i].len;
2212 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2213 if (!name[preflen])
2214 return -EINVAL;
2215 return 0;
2216 }
2217 }
2218 return -EOPNOTSUPP;
2219}
2220
2221static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2222 void *buffer, size_t size)
2223{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002224 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002225 int err;
2226
2227 /*
2228 * If this is a request for a synthetic attribute in the system.*
2229 * namespace use the generic infrastructure to resolve a handler
2230 * for it via sb->s_xattr.
2231 */
2232 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2233 return generic_getxattr(dentry, name, buffer, size);
2234
2235 err = shmem_xattr_validate(name);
2236 if (err)
2237 return err;
2238
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002239 return simple_xattr_get(&info->xattrs, name, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002240}
2241
2242static int shmem_setxattr(struct dentry *dentry, const char *name,
2243 const void *value, size_t size, int flags)
2244{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002245 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002246 int err;
2247
2248 /*
2249 * If this is a request for a synthetic attribute in the system.*
2250 * namespace use the generic infrastructure to resolve a handler
2251 * for it via sb->s_xattr.
2252 */
2253 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2254 return generic_setxattr(dentry, name, value, size, flags);
2255
2256 err = shmem_xattr_validate(name);
2257 if (err)
2258 return err;
2259
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002260 return simple_xattr_set(&info->xattrs, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002261}
2262
2263static int shmem_removexattr(struct dentry *dentry, const char *name)
2264{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002265 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002266 int err;
2267
2268 /*
2269 * If this is a request for a synthetic attribute in the system.*
2270 * namespace use the generic infrastructure to resolve a handler
2271 * for it via sb->s_xattr.
2272 */
2273 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2274 return generic_removexattr(dentry, name);
2275
2276 err = shmem_xattr_validate(name);
2277 if (err)
2278 return err;
2279
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002280 return simple_xattr_remove(&info->xattrs, name);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002281}
2282
2283static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2284{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002285 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2286 return simple_xattr_list(&info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002287}
2288#endif /* CONFIG_TMPFS_XATTR */
2289
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002290static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002292 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002293#ifdef CONFIG_TMPFS_XATTR
2294 .setxattr = shmem_setxattr,
2295 .getxattr = shmem_getxattr,
2296 .listxattr = shmem_listxattr,
2297 .removexattr = shmem_removexattr,
2298#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299};
2300
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002301static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 .readlink = generic_readlink,
2303 .follow_link = shmem_follow_link,
2304 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002305#ifdef CONFIG_TMPFS_XATTR
2306 .setxattr = shmem_setxattr,
2307 .getxattr = shmem_getxattr,
2308 .listxattr = shmem_listxattr,
2309 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002310#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002311};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002312
David M. Grimes91828a42006-10-17 00:09:45 -07002313static struct dentry *shmem_get_parent(struct dentry *child)
2314{
2315 return ERR_PTR(-ESTALE);
2316}
2317
2318static int shmem_match(struct inode *ino, void *vfh)
2319{
2320 __u32 *fh = vfh;
2321 __u64 inum = fh[2];
2322 inum = (inum << 32) | fh[1];
2323 return ino->i_ino == inum && fh[0] == ino->i_generation;
2324}
2325
Christoph Hellwig480b1162007-10-21 16:42:13 -07002326static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2327 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002328{
David M. Grimes91828a42006-10-17 00:09:45 -07002329 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002330 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002331 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002332
Christoph Hellwig480b1162007-10-21 16:42:13 -07002333 if (fh_len < 3)
2334 return NULL;
2335
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002336 inum = fid->raw[2];
2337 inum = (inum << 32) | fid->raw[1];
2338
Christoph Hellwig480b1162007-10-21 16:42:13 -07002339 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2340 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002341 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002342 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002343 iput(inode);
2344 }
2345
Christoph Hellwig480b1162007-10-21 16:42:13 -07002346 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002347}
2348
Al Virob0b03822012-04-02 14:34:06 -04002349static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2350 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002351{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302352 if (*len < 3) {
2353 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07002354 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302355 }
David M. Grimes91828a42006-10-17 00:09:45 -07002356
Al Viro1d3382cb2010-10-23 15:19:20 -04002357 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002358 /* Unfortunately insert_inode_hash is not idempotent,
2359 * so as we hash inodes here rather than at creation
2360 * time, we need a lock to ensure we only try
2361 * to do it once
2362 */
2363 static DEFINE_SPINLOCK(lock);
2364 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002365 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002366 __insert_inode_hash(inode,
2367 inode->i_ino + inode->i_generation);
2368 spin_unlock(&lock);
2369 }
2370
2371 fh[0] = inode->i_generation;
2372 fh[1] = inode->i_ino;
2373 fh[2] = ((__u64)inode->i_ino) >> 32;
2374
2375 *len = 3;
2376 return 1;
2377}
2378
Christoph Hellwig39655162007-10-21 16:42:17 -07002379static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002380 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002381 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002382 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002383};
2384
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002385static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2386 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387{
2388 char *this_char, *value, *rest;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002389 uid_t uid;
2390 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002392 while (options != NULL) {
2393 this_char = options;
2394 for (;;) {
2395 /*
2396 * NUL-terminate this option: unfortunately,
2397 * mount options form a comma-separated list,
2398 * but mpol's nodelist may also contain commas.
2399 */
2400 options = strchr(options, ',');
2401 if (options == NULL)
2402 break;
2403 options++;
2404 if (!isdigit(*options)) {
2405 options[-1] = '\0';
2406 break;
2407 }
2408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 if (!*this_char)
2410 continue;
2411 if ((value = strchr(this_char,'=')) != NULL) {
2412 *value++ = 0;
2413 } else {
2414 printk(KERN_ERR
2415 "tmpfs: No value for mount option '%s'\n",
2416 this_char);
2417 return 1;
2418 }
2419
2420 if (!strcmp(this_char,"size")) {
2421 unsigned long long size;
2422 size = memparse(value,&rest);
2423 if (*rest == '%') {
2424 size <<= PAGE_SHIFT;
2425 size *= totalram_pages;
2426 do_div(size, 100);
2427 rest++;
2428 }
2429 if (*rest)
2430 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002431 sbinfo->max_blocks =
2432 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002434 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 if (*rest)
2436 goto bad_val;
2437 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002438 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 if (*rest)
2440 goto bad_val;
2441 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002442 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002444 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 if (*rest)
2446 goto bad_val;
2447 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002448 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002450 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 if (*rest)
2452 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002453 sbinfo->uid = make_kuid(current_user_ns(), uid);
2454 if (!uid_valid(sbinfo->uid))
2455 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002457 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002459 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 if (*rest)
2461 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002462 sbinfo->gid = make_kgid(current_user_ns(), gid);
2463 if (!gid_valid(sbinfo->gid))
2464 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002465 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002466 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002467 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 } else {
2469 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2470 this_char);
2471 return 1;
2472 }
2473 }
2474 return 0;
2475
2476bad_val:
2477 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2478 value, this_char);
2479 return 1;
2480
2481}
2482
2483static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2484{
2485 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002486 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002487 unsigned long inodes;
2488 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002490 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002491 return error;
2492
2493 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002494 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002495 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002496 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002497 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002498 goto out;
2499 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002500 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002501 * but we must separately disallow unlimited->limited, because
2502 * in that case we have no record of how much is already in use.
2503 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002504 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002505 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002506 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002507 goto out;
2508
2509 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002510 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002511 sbinfo->max_inodes = config.max_inodes;
2512 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002513
2514 mpol_put(sbinfo->mpol);
2515 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002516out:
2517 spin_unlock(&sbinfo->stat_lock);
2518 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002520
Al Viro34c80b12011-12-08 21:32:45 -05002521static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002522{
Al Viro34c80b12011-12-08 21:32:45 -05002523 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002524
2525 if (sbinfo->max_blocks != shmem_default_max_blocks())
2526 seq_printf(seq, ",size=%luk",
2527 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2528 if (sbinfo->max_inodes != shmem_default_max_inodes())
2529 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2530 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002531 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002532 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2533 seq_printf(seq, ",uid=%u",
2534 from_kuid_munged(&init_user_ns, sbinfo->uid));
2535 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2536 seq_printf(seq, ",gid=%u",
2537 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002538 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002539 return 0;
2540}
2541#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
2543static void shmem_put_super(struct super_block *sb)
2544{
Hugh Dickins602586a2010-08-17 15:23:56 -07002545 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2546
2547 percpu_counter_destroy(&sbinfo->used_blocks);
2548 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 sb->s_fs_info = NULL;
2550}
2551
Kay Sievers2b2af542009-04-30 15:23:42 +02002552int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553{
2554 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002555 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002556 int err = -ENOMEM;
2557
2558 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002559 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002560 L1_CACHE_BYTES), GFP_KERNEL);
2561 if (!sbinfo)
2562 return -ENOMEM;
2563
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002564 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002565 sbinfo->uid = current_fsuid();
2566 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002567 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002569#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 /*
2571 * Per default we only allow half of the physical ram per
2572 * tmpfs instance, limiting inodes to one per page of lowmem;
2573 * but the internal instance is left unlimited.
2574 */
2575 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002576 sbinfo->max_blocks = shmem_default_max_blocks();
2577 sbinfo->max_inodes = shmem_default_max_inodes();
2578 if (shmem_parse_options(data, sbinfo, false)) {
2579 err = -EINVAL;
2580 goto failed;
2581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 }
David M. Grimes91828a42006-10-17 00:09:45 -07002583 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07002584 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585#else
2586 sb->s_flags |= MS_NOUSER;
2587#endif
2588
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002589 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002590 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2591 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002592 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002593
Hugh Dickins285b2c42011-08-03 16:21:20 -07002594 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 sb->s_blocksize = PAGE_CACHE_SIZE;
2596 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2597 sb->s_magic = TMPFS_MAGIC;
2598 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002599 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002600#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002601 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002602#endif
2603#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002604 sb->s_flags |= MS_POSIXACL;
2605#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002606
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002607 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 if (!inode)
2609 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002610 inode->i_uid = sbinfo->uid;
2611 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002612 sb->s_root = d_make_root(inode);
2613 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002614 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 return 0;
2616
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617failed:
2618 shmem_put_super(sb);
2619 return err;
2620}
2621
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002622static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
2624static struct inode *shmem_alloc_inode(struct super_block *sb)
2625{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002626 struct shmem_inode_info *info;
2627 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2628 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002630 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631}
2632
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002633static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002634{
2635 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002636 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2637}
2638
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639static void shmem_destroy_inode(struct inode *inode)
2640{
Al Viro09208d12011-07-26 03:15:03 -04002641 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002643 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644}
2645
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002646static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002648 struct shmem_inode_info *info = foo;
2649 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650}
2651
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002652static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653{
2654 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2655 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002656 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 return 0;
2658}
2659
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002660static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002662 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663}
2664
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002665static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002667 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002669 .write_begin = shmem_write_begin,
2670 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002672 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002673 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674};
2675
Helge Deller15ad7cd2006-12-06 20:40:36 -08002676static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 .mmap = shmem_mmap,
2678#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002679 .llseek = shmem_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002680 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002681 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002682 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002683 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002684 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002685 .splice_read = shmem_file_splice_read,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002686 .splice_write = generic_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002687 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688#endif
2689};
2690
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002691static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002692 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002693#ifdef CONFIG_TMPFS_XATTR
2694 .setxattr = shmem_setxattr,
2695 .getxattr = shmem_getxattr,
2696 .listxattr = shmem_listxattr,
2697 .removexattr = shmem_removexattr,
2698#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699};
2700
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002701static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702#ifdef CONFIG_TMPFS
2703 .create = shmem_create,
2704 .lookup = simple_lookup,
2705 .link = shmem_link,
2706 .unlink = shmem_unlink,
2707 .symlink = shmem_symlink,
2708 .mkdir = shmem_mkdir,
2709 .rmdir = shmem_rmdir,
2710 .mknod = shmem_mknod,
2711 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002713#ifdef CONFIG_TMPFS_XATTR
2714 .setxattr = shmem_setxattr,
2715 .getxattr = shmem_getxattr,
2716 .listxattr = shmem_listxattr,
2717 .removexattr = shmem_removexattr,
2718#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002719#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002720 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002721#endif
2722};
2723
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002724static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002725#ifdef CONFIG_TMPFS_XATTR
2726 .setxattr = shmem_setxattr,
2727 .getxattr = shmem_getxattr,
2728 .listxattr = shmem_listxattr,
2729 .removexattr = shmem_removexattr,
2730#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002731#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002732 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002733#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734};
2735
Hugh Dickins759b9772007-03-05 00:30:28 -08002736static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 .alloc_inode = shmem_alloc_inode,
2738 .destroy_inode = shmem_destroy_inode,
2739#ifdef CONFIG_TMPFS
2740 .statfs = shmem_statfs,
2741 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002742 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743#endif
Al Viro1f895f72010-06-05 19:10:41 -04002744 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 .drop_inode = generic_delete_inode,
2746 .put_super = shmem_put_super,
2747};
2748
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002749static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002750 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751#ifdef CONFIG_NUMA
2752 .set_policy = shmem_set_policy,
2753 .get_policy = shmem_get_policy,
2754#endif
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07002755 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756};
2757
Al Viro3c26ff62010-07-25 11:46:36 +04002758static struct dentry *shmem_mount(struct file_system_type *fs_type,
2759 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760{
Al Viro3c26ff62010-07-25 11:46:36 +04002761 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762}
2763
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002764static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 .owner = THIS_MODULE,
2766 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002767 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 .kill_sb = kill_litter_super,
2769};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002771int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772{
2773 int error;
2774
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002775 error = bdi_init(&shmem_backing_dev_info);
2776 if (error)
2777 goto out4;
2778
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002779 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 if (error)
2781 goto out3;
2782
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002783 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 if (error) {
2785 printk(KERN_ERR "Could not register tmpfs\n");
2786 goto out2;
2787 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002788
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002789 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2790 shmem_fs_type.name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 if (IS_ERR(shm_mnt)) {
2792 error = PTR_ERR(shm_mnt);
2793 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2794 goto out1;
2795 }
2796 return 0;
2797
2798out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002799 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002801 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002803 bdi_destroy(&shmem_backing_dev_info);
2804out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 shm_mnt = ERR_PTR(error);
2806 return error;
2807}
Matt Mackall853ac432009-01-06 14:40:20 -08002808
2809#else /* !CONFIG_SHMEM */
2810
2811/*
2812 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2813 *
2814 * This is intended for small system where the benefits of the full
2815 * shmem code (swap-backed and resource-limited) are outweighed by
2816 * their complexity. On systems without swap this code should be
2817 * effectively equivalent, but much lighter weight.
2818 */
2819
2820#include <linux/ramfs.h>
2821
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002822static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002823 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002824 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002825 .kill_sb = kill_litter_super,
2826};
2827
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002828int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002829{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002830 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002831
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002832 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002833 BUG_ON(IS_ERR(shm_mnt));
2834
2835 return 0;
2836}
2837
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002838int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002839{
2840 return 0;
2841}
2842
Hugh Dickins3f96b792009-09-21 17:03:37 -07002843int shmem_lock(struct file *file, int lock, struct user_struct *user)
2844{
2845 return 0;
2846}
2847
Hugh Dickins24513262012-01-20 14:34:21 -08002848void shmem_unlock_mapping(struct address_space *mapping)
2849{
2850}
2851
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002852void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002853{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002854 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002855}
2856EXPORT_SYMBOL_GPL(shmem_truncate_range);
2857
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002858#define shmem_vm_ops generic_file_vm_ops
2859#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002860#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002861#define shmem_acct_size(flags, size) 0
2862#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002863
2864#endif /* CONFIG_SHMEM */
2865
2866/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
Randy Dunlap46711812008-03-19 17:00:41 -07002868/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 * @name: name for dentry (to be seen in /proc/<pid>/maps
2871 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002872 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002874struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875{
2876 int error;
2877 struct file *file;
2878 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002879 struct path path;
2880 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 struct qstr this;
2882
2883 if (IS_ERR(shm_mnt))
2884 return (void *)shm_mnt;
2885
Hugh Dickins285b2c42011-08-03 16:21:20 -07002886 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 return ERR_PTR(-EINVAL);
2888
2889 if (shmem_acct_size(flags, size))
2890 return ERR_PTR(-ENOMEM);
2891
2892 error = -ENOMEM;
2893 this.name = name;
2894 this.len = strlen(name);
2895 this.hash = 0; /* will go */
2896 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002897 path.dentry = d_alloc(root, &this);
2898 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002900 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002903 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002905 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
Al Viro2c48b9c2009-08-09 00:52:35 +04002907 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002909 clear_nlink(inode); /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002910#ifndef CONFIG_MMU
2911 error = ramfs_nommu_expand_for_mapping(inode, size);
2912 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002913 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002914#endif
Al Viro4b42af82009-08-05 18:25:56 +04002915
2916 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002917 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002918 &shmem_file_operations);
2919 if (!file)
2920 goto put_dentry;
2921
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 return file;
2923
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002925 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926put_memory:
2927 shmem_unacct_size(flags, size);
2928 return ERR_PTR(error);
2929}
Keith Packard395e0dd2008-06-20 00:08:06 -07002930EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
Randy Dunlap46711812008-03-19 17:00:41 -07002932/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2935 */
2936int shmem_zero_setup(struct vm_area_struct *vma)
2937{
2938 struct file *file;
2939 loff_t size = vma->vm_end - vma->vm_start;
2940
2941 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2942 if (IS_ERR(file))
2943 return PTR_ERR(file);
2944
2945 if (vma->vm_file)
2946 fput(vma->vm_file);
2947 vma->vm_file = file;
2948 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 return 0;
2950}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002951
2952/**
2953 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2954 * @mapping: the page's address_space
2955 * @index: the page index
2956 * @gfp: the page allocator flags to use if allocating
2957 *
2958 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2959 * with any new page allocations done using the specified allocation flags.
2960 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2961 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2962 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2963 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07002964 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2965 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002966 */
2967struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2968 pgoff_t index, gfp_t gfp)
2969{
Hugh Dickins68da9f02011-07-25 17:12:34 -07002970#ifdef CONFIG_SHMEM
2971 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002972 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07002973 int error;
2974
2975 BUG_ON(mapping->a_ops != &shmem_aops);
2976 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2977 if (error)
2978 page = ERR_PTR(error);
2979 else
2980 unlock_page(page);
2981 return page;
2982#else
2983 /*
2984 * The tiny !SHMEM case uses ramfs without swap
2985 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002986 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002987#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002988}
2989EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);