blob: af68b15a8fc1f99ede5cf82a38aecfc6b3b6eda6 [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>
Andrew Morton250297e2013-04-29 15:06:12 -070028#include <linux/ramfs.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070029#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080030#include <linux/file.h>
31#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040032#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080033#include <linux/swap.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070034#include <linux/aio.h>
Matt Mackall853ac432009-01-06 14:40:20 -080035
36static struct vfsmount *shm_mnt;
37
38#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
43 */
44
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070045#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070046#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010047#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080048#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/string.h>
51#include <linux/slab.h>
52#include <linux/backing-dev.h>
53#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070056#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070057#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070058#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070059#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/security.h>
61#include <linux/swapops.h>
62#include <linux/mempolicy.h>
63#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000064#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070065#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070066#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080067#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040068#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070069
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <asm/pgtable.h>
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* Pretend that each entry is of this size in directory's i_size */
77#define BOGO_DIRENT_SIZE 20
78
Hugh Dickins69f07ec2011-08-03 16:21:26 -070079/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80#define SHORT_SYMLINK_LEN 128
81
Hugh Dickins1aac1402012-05-29 15:06:42 -070082/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070083 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
84 * inode->i_private (with i_mutex making sure that it has only one user at
85 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070086 */
87struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -070088 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -070089 pgoff_t start; /* start of range currently being fallocated */
90 pgoff_t next; /* the next page offset to be fallocated */
91 pgoff_t nr_falloced; /* how many new pages have been fallocated */
92 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
93};
94
Hugh Dickins285b2c42011-08-03 16:21:20 -070095/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -070096enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 SGP_READ, /* don't exceed i_size, don't allocate page */
98 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080099 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700100 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
101 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102};
103
Andrew Mortonb76db732008-02-08 04:21:49 -0800104#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800105static unsigned long shmem_default_max_blocks(void)
106{
107 return totalram_pages / 2;
108}
109
110static unsigned long shmem_default_max_inodes(void)
111{
112 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
113}
Andrew Mortonb76db732008-02-08 04:21:49 -0800114#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800115
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700116static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
117static int shmem_replace_page(struct page **pagep, gfp_t gfp,
118 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700119static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
120 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
121
122static inline int shmem_getpage(struct inode *inode, pgoff_t index,
123 struct page **pagep, enum sgp_type sgp, int *fault_type)
124{
125 return shmem_getpage_gfp(inode, index, pagep, sgp,
126 mapping_gfp_mask(inode->i_mapping), fault_type);
127}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
130{
131 return sb->s_fs_info;
132}
133
134/*
135 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
136 * for shared memory and for shared anonymous (/dev/zero) mappings
137 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
138 * consistent with the pre-accounting of private mappings ...
139 */
140static inline int shmem_acct_size(unsigned long flags, loff_t size)
141{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000142 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000143 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
146static inline void shmem_unacct_size(unsigned long flags, loff_t size)
147{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000148 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 vm_unacct_memory(VM_ACCT(size));
150}
151
152/*
153 * ... whereas tmpfs objects are accounted incrementally as
154 * pages are allocated, in order to allow huge sparse files.
155 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
156 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
157 */
158static inline int shmem_acct_block(unsigned long flags)
159{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000160 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000161 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
164static inline void shmem_unacct_blocks(unsigned long flags, long pages)
165{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000166 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
168}
169
Hugh Dickins759b9772007-03-05 00:30:28 -0800170static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700171static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800172static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800173static const struct inode_operations shmem_inode_operations;
174static const struct inode_operations shmem_dir_inode_operations;
175static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400176static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700178static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700180 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181};
182
183static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800184static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800186static int shmem_reserve_inode(struct super_block *sb)
187{
188 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
189 if (sbinfo->max_inodes) {
190 spin_lock(&sbinfo->stat_lock);
191 if (!sbinfo->free_inodes) {
192 spin_unlock(&sbinfo->stat_lock);
193 return -ENOSPC;
194 }
195 sbinfo->free_inodes--;
196 spin_unlock(&sbinfo->stat_lock);
197 }
198 return 0;
199}
200
201static void shmem_free_inode(struct super_block *sb)
202{
203 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
204 if (sbinfo->max_inodes) {
205 spin_lock(&sbinfo->stat_lock);
206 sbinfo->free_inodes++;
207 spin_unlock(&sbinfo->stat_lock);
208 }
209}
210
Randy Dunlap46711812008-03-19 17:00:41 -0700211/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700212 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 * @inode: inode to recalc
214 *
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
217 *
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
220 *
221 * It has to be called with the spinlock held.
222 */
223static void shmem_recalc_inode(struct inode *inode)
224{
225 struct shmem_inode_info *info = SHMEM_I(inode);
226 long freed;
227
228 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
229 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700230 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
231 if (sbinfo->max_blocks)
232 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700234 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
237}
238
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700239/*
240 * Replace item expected in radix tree by a new item, while holding tree lock.
241 */
242static int shmem_radix_tree_replace(struct address_space *mapping,
243 pgoff_t index, void *expected, void *replacement)
244{
245 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700246 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700247
248 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700249 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700250 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700251 if (!pslot)
252 return -ENOENT;
253 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700254 if (item != expected)
255 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700256 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700257 return 0;
258}
259
260/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700261 * Sometimes, before we decide whether to proceed or to fail, we must check
262 * that an entry was not already brought back from swap by a racing thread.
263 *
264 * Checking page is not enough: by the time a SwapCache page is locked, it
265 * might be reused, and again be SwapCache, using the same swap as before.
266 */
267static bool shmem_confirm_swap(struct address_space *mapping,
268 pgoff_t index, swp_entry_t swap)
269{
270 void *item;
271
272 rcu_read_lock();
273 item = radix_tree_lookup(&mapping->page_tree, index);
274 rcu_read_unlock();
275 return item == swp_to_radix_entry(swap);
276}
277
278/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700279 * Like add_to_page_cache_locked, but error if expected item has gone.
280 */
281static int shmem_add_to_page_cache(struct page *page,
282 struct address_space *mapping,
283 pgoff_t index, gfp_t gfp, void *expected)
284{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700285 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700286
Sasha Levin309381fea2014-01-23 15:52:54 -0800287 VM_BUG_ON_PAGE(!PageLocked(page), page);
288 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700289
Hugh Dickinsb065b432012-07-11 14:02:48 -0700290 page_cache_get(page);
291 page->mapping = mapping;
292 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700293
Hugh Dickinsb065b432012-07-11 14:02:48 -0700294 spin_lock_irq(&mapping->tree_lock);
295 if (!expected)
296 error = radix_tree_insert(&mapping->page_tree, index, page);
297 else
298 error = shmem_radix_tree_replace(mapping, index, expected,
299 page);
300 if (!error) {
301 mapping->nrpages++;
302 __inc_zone_page_state(page, NR_FILE_PAGES);
303 __inc_zone_page_state(page, NR_SHMEM);
304 spin_unlock_irq(&mapping->tree_lock);
305 } else {
306 page->mapping = NULL;
307 spin_unlock_irq(&mapping->tree_lock);
308 page_cache_release(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700309 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700310 return error;
311}
312
313/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700314 * Like delete_from_page_cache, but substitutes swap for page.
315 */
316static void shmem_delete_from_page_cache(struct page *page, void *radswap)
317{
318 struct address_space *mapping = page->mapping;
319 int error;
320
321 spin_lock_irq(&mapping->tree_lock);
322 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
323 page->mapping = NULL;
324 mapping->nrpages--;
325 __dec_zone_page_state(page, NR_FILE_PAGES);
326 __dec_zone_page_state(page, NR_SHMEM);
327 spin_unlock_irq(&mapping->tree_lock);
328 page_cache_release(page);
329 BUG_ON(error);
330}
331
332/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700333 * Remove swap entry from radix tree, free the swap and its page cache.
334 */
335static int shmem_free_swap(struct address_space *mapping,
336 pgoff_t index, void *radswap)
337{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700338 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700339
340 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700341 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700342 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700343 if (old != radswap)
344 return -ENOENT;
345 free_swap_and_cache(radix_to_swp_entry(radswap));
346 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700347}
348
349/*
Hugh Dickins24513262012-01-20 14:34:21 -0800350 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
351 */
352void shmem_unlock_mapping(struct address_space *mapping)
353{
354 struct pagevec pvec;
355 pgoff_t indices[PAGEVEC_SIZE];
356 pgoff_t index = 0;
357
358 pagevec_init(&pvec, 0);
359 /*
360 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
361 */
362 while (!mapping_unevictable(mapping)) {
363 /*
364 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
365 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
366 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700367 pvec.nr = find_get_entries(mapping, index,
368 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800369 if (!pvec.nr)
370 break;
371 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700372 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800373 check_move_unevictable_pages(pvec.pages, pvec.nr);
374 pagevec_release(&pvec);
375 cond_resched();
376 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700377}
378
379/*
380 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700381 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700382 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700383static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
384 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700386 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700388 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700389 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
390 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
391 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700392 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700393 pgoff_t indices[PAGEVEC_SIZE];
394 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700395 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700396 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700398 if (lend == -1)
399 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700400
401 pagevec_init(&pvec, 0);
402 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700403 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700404 pvec.nr = find_get_entries(mapping, index,
405 min(end - index, (pgoff_t)PAGEVEC_SIZE),
406 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700407 if (!pvec.nr)
408 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700409 mem_cgroup_uncharge_start();
410 for (i = 0; i < pagevec_count(&pvec); i++) {
411 struct page *page = pvec.pages[i];
412
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700413 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700414 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700415 break;
416
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700417 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700418 if (unfalloc)
419 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700420 nr_swaps_freed += !shmem_free_swap(mapping,
421 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700422 continue;
423 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700424
425 if (!trylock_page(page))
426 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700427 if (!unfalloc || !PageUptodate(page)) {
428 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800429 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700430 truncate_inode_page(mapping, page);
431 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700432 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700433 unlock_page(page);
434 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700435 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800436 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700437 mem_cgroup_uncharge_end();
438 cond_resched();
439 index++;
440 }
441
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700442 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700443 struct page *page = NULL;
444 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
445 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700446 unsigned int top = PAGE_CACHE_SIZE;
447 if (start > end) {
448 top = partial_end;
449 partial_end = 0;
450 }
451 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700452 set_page_dirty(page);
453 unlock_page(page);
454 page_cache_release(page);
455 }
456 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700457 if (partial_end) {
458 struct page *page = NULL;
459 shmem_getpage(inode, end, &page, SGP_READ, NULL);
460 if (page) {
461 zero_user_segment(page, 0, partial_end);
462 set_page_dirty(page);
463 unlock_page(page);
464 page_cache_release(page);
465 }
466 }
467 if (start >= end)
468 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700469
470 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700471 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700472 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700473
474 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700475 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700476 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700477 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700478 /* If all gone or hole-punch or unfalloc, we're done */
479 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700480 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700481 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700482 index = start;
483 continue;
484 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700485 mem_cgroup_uncharge_start();
486 for (i = 0; i < pagevec_count(&pvec); i++) {
487 struct page *page = pvec.pages[i];
488
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700489 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700490 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700491 break;
492
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700493 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700494 if (unfalloc)
495 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700496 if (shmem_free_swap(mapping, index, page)) {
497 /* Swap was replaced by page: retry */
498 index--;
499 break;
500 }
501 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700502 continue;
503 }
504
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700505 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700506 if (!unfalloc || !PageUptodate(page)) {
507 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800508 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700509 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700510 } else {
511 /* Page was replaced by swap: retry */
512 unlock_page(page);
513 index--;
514 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700515 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700516 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700517 unlock_page(page);
518 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700519 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800520 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700521 mem_cgroup_uncharge_end();
522 index++;
523 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700526 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 shmem_recalc_inode(inode);
528 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700529}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700531void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
532{
533 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700534 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700536EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
Hugh Dickins94c1e622011-06-27 16:18:03 -0700538static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 int error;
542
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200543 error = inode_change_ok(inode, attr);
544 if (error)
545 return error;
546
Hugh Dickins94c1e622011-06-27 16:18:03 -0700547 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
548 loff_t oldsize = inode->i_size;
549 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000550
Hugh Dickins94c1e622011-06-27 16:18:03 -0700551 if (newsize != oldsize) {
552 i_size_write(inode, newsize);
553 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
554 }
555 if (newsize < oldsize) {
556 loff_t holebegin = round_up(newsize, PAGE_SIZE);
557 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
558 shmem_truncate_range(inode, newsize, (loff_t)-1);
559 /* unmap again to remove racily COWed private pages */
560 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200564 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200565 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -0800566 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 return error;
568}
569
Al Viro1f895f72010-06-05 19:10:41 -0400570static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 struct shmem_inode_info *info = SHMEM_I(inode);
573
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000574 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 shmem_unacct_size(info->flags, inode->i_size);
576 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000577 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800579 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800581 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700583 } else
584 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700585
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400586 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800587 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800588 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200589 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}
591
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700592/*
593 * If swap found in inode, free it and move page from swapcache to filecache.
594 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700595static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700596 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700598 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700599 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700600 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700601 gfp_t gfp;
602 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700604 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700605 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700606 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700607 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800608
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800609 /*
610 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400611 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800612 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700613 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800614 */
615 if (shmem_swaplist.next != &info->swaplist)
616 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800617
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700618 gfp = mapping_gfp_mask(mapping);
619 if (shmem_should_replace_page(*pagep, gfp)) {
620 mutex_unlock(&shmem_swaplist_mutex);
621 error = shmem_replace_page(pagep, gfp, info, index);
622 mutex_lock(&shmem_swaplist_mutex);
623 /*
624 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700625 * allocation, but the inode might have been freed while we
626 * dropped it: although a racing shmem_evict_inode() cannot
627 * complete without emptying the radix_tree, our page lock
628 * on this swapcache page is not enough to prevent that -
629 * free_swap_and_cache() of our swap entry will only
630 * trylock_page(), removing swap from radix_tree whatever.
631 *
632 * We must not proceed to shmem_add_to_page_cache() if the
633 * inode has been freed, but of course we cannot rely on
634 * inode or mapping or info to check that. However, we can
635 * safely check if our swap entry is still in use (and here
636 * it can't have got reused for another page): if it's still
637 * in use, then the inode cannot have been freed yet, and we
638 * can safely proceed (if it's no longer in use, that tells
639 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700640 */
641 if (!page_swapcount(*pagep))
642 error = -ENOENT;
643 }
644
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800645 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700646 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
647 * but also to hold up shmem_evict_inode(): so inode cannot be freed
648 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800649 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700650 if (!error)
651 error = shmem_add_to_page_cache(*pagep, mapping, index,
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700652 GFP_NOWAIT, radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700653 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700654 /*
655 * Truncation and eviction use free_swap_and_cache(), which
656 * only does trylock page: if we raced, best clean up here.
657 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700658 delete_from_swap_cache(*pagep);
659 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700660 if (!error) {
661 spin_lock(&info->lock);
662 info->swapped--;
663 spin_unlock(&info->lock);
664 swap_free(swap);
665 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800666 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800668 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669}
670
671/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700672 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700674int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700676 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 struct shmem_inode_info *info;
678 int found = 0;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700679 int error = 0;
680
681 /*
682 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700683 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700684 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700685 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700686 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700687
688 /*
689 * Charge page using GFP_KERNEL while we can wait, before taking
690 * the shmem_swaplist_mutex which might hold up shmem_writepage().
691 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700692 */
Michal Hockod715ae02014-04-07 15:37:46 -0700693 error = mem_cgroup_charge_file(page, current->mm, GFP_KERNEL);
Hugh Dickins778dd892011-05-11 15:13:37 -0700694 if (error)
695 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700696 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800698 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700699 list_for_each_safe(this, next, &shmem_swaplist) {
700 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700701 if (info->swapped)
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700702 found = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700703 else
704 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800705 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800706 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700707 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800709 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700710
Hugh Dickins778dd892011-05-11 15:13:37 -0700711 if (found < 0)
712 error = found;
713out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800714 unlock_page(page);
715 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700716 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
719/*
720 * Move the page from the page cache to the swap cache.
721 */
722static int shmem_writepage(struct page *page, struct writeback_control *wbc)
723{
724 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700727 swp_entry_t swap;
728 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 mapping = page->mapping;
732 index = page->index;
733 inode = mapping->host;
734 info = SHMEM_I(inode);
735 if (info->flags & VM_LOCKED)
736 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800737 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 goto redirty;
739
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800740 /*
741 * shmem_backing_dev_info's capabilities prevent regular writeback or
742 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700743 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800744 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700745 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800746 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700747 if (!wbc->for_reclaim) {
748 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
749 goto redirty;
750 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700751
752 /*
753 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
754 * value into swapfile.c, the only way we can correctly account for a
755 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700756 *
757 * That's okay for a page already fallocated earlier, but if we have
758 * not yet completed the fallocation, then (a) we want to keep track
759 * of this page in case we have to undo it, and (b) it may not be a
760 * good idea to continue anyway, once we're pushing into swap. So
761 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700762 */
763 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700764 if (inode->i_private) {
765 struct shmem_falloc *shmem_falloc;
766 spin_lock(&inode->i_lock);
767 shmem_falloc = inode->i_private;
768 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -0700769 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -0700770 index >= shmem_falloc->start &&
771 index < shmem_falloc->next)
772 shmem_falloc->nr_unswapped++;
773 else
774 shmem_falloc = NULL;
775 spin_unlock(&inode->i_lock);
776 if (shmem_falloc)
777 goto redirty;
778 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700779 clear_highpage(page);
780 flush_dcache_page(page);
781 SetPageUptodate(page);
782 }
783
Hugh Dickins48f170f2011-07-25 17:12:37 -0700784 swap = get_swap_page();
785 if (!swap.val)
786 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800787
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700788 /*
789 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700790 * if it's not already there. Do it now before the page is
791 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700792 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700793 * we've incremented swapped, because shmem_unuse_inode() will
794 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700795 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700796 mutex_lock(&shmem_swaplist_mutex);
797 if (list_empty(&info->swaplist))
798 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700799
Hugh Dickins48f170f2011-07-25 17:12:37 -0700800 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800801 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700802 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
803
804 spin_lock(&info->lock);
805 info->swapped++;
806 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700807 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700808
809 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800810 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700811 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return 0;
813 }
814
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700815 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700816 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817redirty:
818 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800819 if (wbc->for_reclaim)
820 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
821 unlock_page(page);
822 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823}
824
825#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800826#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700827static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800828{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700829 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800830
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700831 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700832 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800833
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800834 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800835
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700836 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800837}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700838
839static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
840{
841 struct mempolicy *mpol = NULL;
842 if (sbinfo->mpol) {
843 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
844 mpol = sbinfo->mpol;
845 mpol_get(mpol);
846 spin_unlock(&sbinfo->stat_lock);
847 }
848 return mpol;
849}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800850#endif /* CONFIG_TMPFS */
851
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700852static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
853 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800856 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800858 /* Create a pseudo vma that just contains the policy */
859 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700860 /* Bias interleave by inode number to distribute better across nodes */
861 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800862 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700863 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700864
Mel Gorman18a2f372012-12-05 14:01:41 -0800865 page = swapin_readahead(swap, gfp, &pvma, 0);
866
867 /* Drop reference taken by mpol_shared_policy_lookup() */
868 mpol_cond_put(pvma.vm_policy);
869
870 return page;
871}
872
873static struct page *shmem_alloc_page(gfp_t gfp,
874 struct shmem_inode_info *info, pgoff_t index)
875{
876 struct vm_area_struct pvma;
877 struct page *page;
878
879 /* Create a pseudo vma that just contains the policy */
880 pvma.vm_start = 0;
881 /* Bias interleave by inode number to distribute better across nodes */
882 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
883 pvma.vm_ops = NULL;
884 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
885
886 page = alloc_page_vma(gfp, &pvma, 0);
887
888 /* Drop reference taken by mpol_shared_policy_lookup() */
889 mpol_cond_put(pvma.vm_policy);
890
891 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800893#else /* !CONFIG_NUMA */
894#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700895static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800896{
897}
898#endif /* CONFIG_TMPFS */
899
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700900static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
901 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700903 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
Hugh Dickins02098fe2008-02-04 22:28:42 -0800906static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700907 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000909 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800911#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700913#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
914static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
915{
916 return NULL;
917}
918#endif
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700921 * When a page is moved from swapcache to shmem filecache (either by the
922 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
923 * shmem_unuse_inode()), it may have been read in earlier from swap, in
924 * ignorance of the mapping it belongs to. If that mapping has special
925 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
926 * we may need to copy to a suitable page before moving to filecache.
927 *
928 * In a future release, this may well be extended to respect cpuset and
929 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
930 * but for now it is a simple matter of zone.
931 */
932static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
933{
934 return page_zonenum(page) > gfp_zone(gfp);
935}
936
937static int shmem_replace_page(struct page **pagep, gfp_t gfp,
938 struct shmem_inode_info *info, pgoff_t index)
939{
940 struct page *oldpage, *newpage;
941 struct address_space *swap_mapping;
942 pgoff_t swap_index;
943 int error;
944
945 oldpage = *pagep;
946 swap_index = page_private(oldpage);
947 swap_mapping = page_mapping(oldpage);
948
949 /*
950 * We have arrived here because our zones are constrained, so don't
951 * limit chance of success by further cpuset and node constraints.
952 */
953 gfp &= ~GFP_CONSTRAINT_MASK;
954 newpage = shmem_alloc_page(gfp, info, index);
955 if (!newpage)
956 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700957
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700958 page_cache_get(newpage);
959 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -0700960 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700961
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700962 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700963 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700964 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700965 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700966 SetPageSwapCache(newpage);
967
968 /*
969 * Our caller will very soon move newpage out of swapcache, but it's
970 * a nice clean interface for us to replace oldpage by newpage there.
971 */
972 spin_lock_irq(&swap_mapping->tree_lock);
973 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
974 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -0700975 if (!error) {
976 __inc_zone_page_state(newpage, NR_FILE_PAGES);
977 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
978 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700979 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700980
Hugh Dickins0142ef62012-06-07 14:21:09 -0700981 if (unlikely(error)) {
982 /*
983 * Is this possible? I think not, now that our callers check
984 * both PageSwapCache and page_private after getting page lock;
985 * but be defensive. Reverse old to newpage for clear and free.
986 */
987 oldpage = newpage;
988 } else {
989 mem_cgroup_replace_page_cache(oldpage, newpage);
990 lru_cache_add_anon(newpage);
991 *pagep = newpage;
992 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700993
994 ClearPageSwapCache(oldpage);
995 set_page_private(oldpage, 0);
996
997 unlock_page(oldpage);
998 page_cache_release(oldpage);
999 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001000 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001001}
1002
1003/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001004 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 *
1006 * If we allocate a new one we do not mark it dirty. That's up to the
1007 * vm. If we swap it in we mark it dirty since we also free the swap
1008 * entry since a page cannot live in both the swap and page cache
1009 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001010static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001011 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012{
1013 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001014 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001016 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 swp_entry_t swap;
1018 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001019 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001020 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001022 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001025 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001026 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001027 if (radix_tree_exceptional_entry(page)) {
1028 swap = radix_to_swp_entry(page);
1029 page = NULL;
1030 }
1031
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001032 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001033 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1034 error = -EINVAL;
1035 goto failed;
1036 }
1037
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001038 if (page && sgp == SGP_WRITE)
1039 mark_page_accessed(page);
1040
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001041 /* fallocated page? */
1042 if (page && !PageUptodate(page)) {
1043 if (sgp != SGP_READ)
1044 goto clear;
1045 unlock_page(page);
1046 page_cache_release(page);
1047 page = NULL;
1048 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001049 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001050 *pagep = page;
1051 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
1054 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001055 * Fast cache lookup did not find it:
1056 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001058 info = SHMEM_I(inode);
1059 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (swap.val) {
1062 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001063 page = lookup_swap_cache(swap);
1064 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001065 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001066 if (fault_type)
1067 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001068 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001069 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001070 error = -ENOMEM;
1071 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074
1075 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001076 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001077 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001078 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001079 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001080 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001081 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001082 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001084 goto failed;
1085 }
1086 wait_on_page_writeback(page);
1087
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001088 if (shmem_should_replace_page(page, gfp)) {
1089 error = shmem_replace_page(&page, gfp, info, index);
1090 if (error)
1091 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 }
1093
Michal Hockod715ae02014-04-07 15:37:46 -07001094 error = mem_cgroup_charge_file(page, current->mm,
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001095 gfp & GFP_RECLAIM_MASK);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001096 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001097 error = shmem_add_to_page_cache(page, mapping, index,
1098 gfp, swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001099 /*
1100 * We already confirmed swap under page lock, and make
1101 * no memory allocation here, so usually no possibility
1102 * of error; but free_swap_and_cache() only trylocks a
1103 * page, so it is just possible that the entry has been
1104 * truncated or holepunched since swap was confirmed.
1105 * shmem_undo_range() will have done some of the
1106 * unaccounting, now delete_from_swap_cache() will do
1107 * the rest (including mem_cgroup_uncharge_swapcache).
1108 * Reset swap.val? No, leave it so "failed" goes back to
1109 * "repeat": reading a hole and writing should succeed.
1110 */
1111 if (error)
1112 delete_from_swap_cache(page);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001113 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001114 if (error)
1115 goto failed;
1116
1117 spin_lock(&info->lock);
1118 info->swapped--;
1119 shmem_recalc_inode(inode);
1120 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001121
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001122 if (sgp == SGP_WRITE)
1123 mark_page_accessed(page);
1124
Hugh Dickins27ab7002011-07-25 17:12:36 -07001125 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001126 set_page_dirty(page);
1127 swap_free(swap);
1128
Hugh Dickins54af60422011-08-03 16:21:24 -07001129 } else {
1130 if (shmem_acct_block(info->flags)) {
1131 error = -ENOSPC;
1132 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001134 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001135 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001136 sbinfo->max_blocks) >= 0) {
1137 error = -ENOSPC;
1138 goto unacct;
1139 }
Tim Chen7e496292010-08-09 17:19:05 -07001140 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Hugh Dickins54af60422011-08-03 16:21:24 -07001143 page = shmem_alloc_page(gfp, info, index);
1144 if (!page) {
1145 error = -ENOMEM;
1146 goto decused;
1147 }
1148
Mel Gorman07a42782014-06-04 16:10:24 -07001149 __SetPageSwapBacked(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001150 __set_page_locked(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001151 if (sgp == SGP_WRITE)
1152 init_page_accessed(page);
1153
Michal Hockod715ae02014-04-07 15:37:46 -07001154 error = mem_cgroup_charge_file(page, current->mm,
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001155 gfp & GFP_RECLAIM_MASK);
Hugh Dickins54af60422011-08-03 16:21:24 -07001156 if (error)
1157 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001158 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001159 if (!error) {
1160 error = shmem_add_to_page_cache(page, mapping, index,
1161 gfp, NULL);
1162 radix_tree_preload_end();
1163 }
1164 if (error) {
1165 mem_cgroup_uncharge_cache_page(page);
1166 goto decused;
1167 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001168 lru_cache_add_anon(page);
1169
1170 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001172 inode->i_blocks += BLOCKS_PER_PAGE;
1173 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001175 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001176
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001177 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001178 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1179 */
1180 if (sgp == SGP_FALLOC)
1181 sgp = SGP_WRITE;
1182clear:
1183 /*
1184 * Let SGP_WRITE caller clear ends if write does not fill page;
1185 * but SGP_FALLOC on a page fallocated earlier must initialize
1186 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001187 */
1188 if (sgp != SGP_WRITE) {
1189 clear_highpage(page);
1190 flush_dcache_page(page);
1191 SetPageUptodate(page);
1192 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001193 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001194 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001195 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001196
Hugh Dickins54af60422011-08-03 16:21:24 -07001197 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001198 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001199 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1200 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001201 if (alloced)
1202 goto trunc;
1203 else
1204 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001205 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001206 *pagep = page;
1207 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001208
Nick Piggind0217ac2007-07-19 01:47:03 -07001209 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001210 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001212trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001213 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001214 ClearPageDirty(page);
1215 delete_from_page_cache(page);
1216 spin_lock(&info->lock);
1217 info->alloced--;
1218 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001220decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001221 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001222 if (sbinfo->max_blocks)
1223 percpu_counter_add(&sbinfo->used_blocks, -1);
1224unacct:
1225 shmem_unacct_blocks(info->flags, 1);
1226failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001227 if (swap.val && error != -EINVAL &&
1228 !shmem_confirm_swap(mapping, index, swap))
1229 error = -EEXIST;
1230unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001231 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001232 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001233 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001234 }
1235 if (error == -ENOSPC && !once++) {
1236 info = SHMEM_I(inode);
1237 spin_lock(&info->lock);
1238 shmem_recalc_inode(inode);
1239 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001240 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001241 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001242 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001243 goto repeat;
1244 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245}
1246
1247static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1248{
Al Viro496ad9a2013-01-23 17:07:38 -05001249 struct inode *inode = file_inode(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001251 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001253 /*
1254 * Trinity finds that probing a hole which tmpfs is punching can
1255 * prevent the hole-punch from ever completing: which in turn
1256 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001257 * faulting pages into the hole while it's being punched. Although
1258 * shmem_undo_range() does remove the additions, it may be unable to
1259 * keep up, as each new page needs its own unmap_mapping_range() call,
1260 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1261 *
1262 * It does not matter if we sometimes reach this check just before the
1263 * hole-punch begins, so that one fault then races with the punch:
1264 * we just need to make racing faults a rare case.
1265 *
1266 * The implementation below would be much simpler if we just used a
1267 * standard mutex or completion: but we cannot take i_mutex in fault,
1268 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001269 */
1270 if (unlikely(inode->i_private)) {
1271 struct shmem_falloc *shmem_falloc;
1272
1273 spin_lock(&inode->i_lock);
1274 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001275 if (shmem_falloc &&
1276 shmem_falloc->waitq &&
1277 vmf->pgoff >= shmem_falloc->start &&
1278 vmf->pgoff < shmem_falloc->next) {
1279 wait_queue_head_t *shmem_falloc_waitq;
1280 DEFINE_WAIT(shmem_fault_wait);
1281
1282 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001283 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1284 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001285 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001286 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001287 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001288 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001289
1290 shmem_falloc_waitq = shmem_falloc->waitq;
1291 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1292 TASK_UNINTERRUPTIBLE);
1293 spin_unlock(&inode->i_lock);
1294 schedule();
1295
1296 /*
1297 * shmem_falloc_waitq points into the shmem_fallocate()
1298 * stack of the hole-punching task: shmem_falloc_waitq
1299 * is usually invalid by the time we reach here, but
1300 * finish_wait() does not dereference it in that case;
1301 * though i_lock needed lest racing with wake_up_all().
1302 */
1303 spin_lock(&inode->i_lock);
1304 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1305 spin_unlock(&inode->i_lock);
1306 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001307 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001308 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001309 }
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1312 if (error)
1313 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001314
Ying Han456f9982011-05-26 16:25:38 -07001315 if (ret & VM_FAULT_MAJOR) {
1316 count_vm_event(PGMAJFAULT);
1317 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1318 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001319 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320}
1321
1322#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001323static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324{
Al Viro496ad9a2013-01-23 17:07:38 -05001325 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001326 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327}
1328
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001329static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1330 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
Al Viro496ad9a2013-01-23 17:07:38 -05001332 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001333 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001335 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1336 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337}
1338#endif
1339
1340int shmem_lock(struct file *file, int lock, struct user_struct *user)
1341{
Al Viro496ad9a2013-01-23 17:07:38 -05001342 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 struct shmem_inode_info *info = SHMEM_I(inode);
1344 int retval = -ENOMEM;
1345
1346 spin_lock(&info->lock);
1347 if (lock && !(info->flags & VM_LOCKED)) {
1348 if (!user_shm_lock(inode->i_size, user))
1349 goto out_nomem;
1350 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001351 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353 if (!lock && (info->flags & VM_LOCKED) && user) {
1354 user_shm_unlock(inode->i_size, user);
1355 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001356 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360out_nomem:
1361 spin_unlock(&info->lock);
1362 return retval;
1363}
1364
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001365static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
1367 file_accessed(file);
1368 vma->vm_ops = &shmem_vm_ops;
1369 return 0;
1370}
1371
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001372static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001373 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
1375 struct inode *inode;
1376 struct shmem_inode_info *info;
1377 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1378
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001379 if (shmem_reserve_inode(sb))
1380 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382 inode = new_inode(sb);
1383 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001384 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001385 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1388 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001389 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 info = SHMEM_I(inode);
1391 memset(info, 0, (char *)inode - (char *)info);
1392 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001393 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001395 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001396 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
1398 switch (mode & S_IFMT) {
1399 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001400 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 init_special_inode(inode, mode, dev);
1402 break;
1403 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001404 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 inode->i_op = &shmem_inode_operations;
1406 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001407 mpol_shared_policy_init(&info->policy,
1408 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 break;
1410 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001411 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 /* Some things misbehave if size == 0 on a directory */
1413 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1414 inode->i_op = &shmem_dir_inode_operations;
1415 inode->i_fop = &simple_dir_operations;
1416 break;
1417 case S_IFLNK:
1418 /*
1419 * Must not load anything in the rbtree,
1420 * mpol_free_shared_policy will not be called.
1421 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001422 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 break;
1424 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001425 } else
1426 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return inode;
1428}
1429
Johannes Weiner0cd61442014-04-03 14:47:46 -07001430bool shmem_mapping(struct address_space *mapping)
1431{
1432 return mapping->backing_dev_info == &shmem_backing_dev_info;
1433}
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001436static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001437static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001439#ifdef CONFIG_TMPFS_XATTR
1440static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1441#else
1442#define shmem_initxattrs NULL
1443#endif
1444
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001446shmem_write_begin(struct file *file, struct address_space *mapping,
1447 loff_t pos, unsigned len, unsigned flags,
1448 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
Nick Piggin800d15a2007-10-16 01:25:03 -07001450 struct inode *inode = mapping->host;
1451 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001452 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
Nick Piggin800d15a2007-10-16 01:25:03 -07001453}
1454
1455static int
1456shmem_write_end(struct file *file, struct address_space *mapping,
1457 loff_t pos, unsigned len, unsigned copied,
1458 struct page *page, void *fsdata)
1459{
1460 struct inode *inode = mapping->host;
1461
Hugh Dickinsd3602442008-02-04 22:28:44 -08001462 if (pos + copied > inode->i_size)
1463 i_size_write(inode, pos + copied);
1464
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001465 if (!PageUptodate(page)) {
1466 if (copied < PAGE_CACHE_SIZE) {
1467 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1468 zero_user_segments(page, 0, from,
1469 from + copied, PAGE_CACHE_SIZE);
1470 }
1471 SetPageUptodate(page);
1472 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001473 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001474 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001475 page_cache_release(page);
1476
Nick Piggin800d15a2007-10-16 01:25:03 -07001477 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
Al Viro2ba5bbe2014-04-02 20:00:02 -04001480static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481{
Al Viro6e58e792014-02-03 17:07:03 -05001482 struct file *file = iocb->ki_filp;
1483 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001485 pgoff_t index;
1486 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001487 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02001488 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05001489 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05001490 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001491
1492 /*
1493 * Might this read be for a stacking filesystem? Then when reading
1494 * holes of a sparse file, we actually need to allocate those pages,
1495 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1496 */
1497 if (segment_eq(get_fs(), KERNEL_DS))
1498 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500 index = *ppos >> PAGE_CACHE_SHIFT;
1501 offset = *ppos & ~PAGE_CACHE_MASK;
1502
1503 for (;;) {
1504 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001505 pgoff_t end_index;
1506 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 loff_t i_size = i_size_read(inode);
1508
1509 end_index = i_size >> PAGE_CACHE_SHIFT;
1510 if (index > end_index)
1511 break;
1512 if (index == end_index) {
1513 nr = i_size & ~PAGE_CACHE_MASK;
1514 if (nr <= offset)
1515 break;
1516 }
1517
Al Viro6e58e792014-02-03 17:07:03 -05001518 error = shmem_getpage(inode, index, &page, sgp, NULL);
1519 if (error) {
1520 if (error == -EINVAL)
1521 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 break;
1523 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001524 if (page)
1525 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
1527 /*
1528 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001529 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 */
1531 nr = PAGE_CACHE_SIZE;
1532 i_size = i_size_read(inode);
1533 end_index = i_size >> PAGE_CACHE_SHIFT;
1534 if (index == end_index) {
1535 nr = i_size & ~PAGE_CACHE_MASK;
1536 if (nr <= offset) {
1537 if (page)
1538 page_cache_release(page);
1539 break;
1540 }
1541 }
1542 nr -= offset;
1543
1544 if (page) {
1545 /*
1546 * If users can be writing to this page using arbitrary
1547 * virtual addresses, take care about potential aliasing
1548 * before reading the page on the kernel side.
1549 */
1550 if (mapping_writably_mapped(mapping))
1551 flush_dcache_page(page);
1552 /*
1553 * Mark the page accessed if we read the beginning.
1554 */
1555 if (!offset)
1556 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001557 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001559 page_cache_get(page);
1560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 /*
1563 * Ok, we have the page, and it's up-to-date, so
1564 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04001566 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05001567 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 offset += ret;
1569 index += offset >> PAGE_CACHE_SHIFT;
1570 offset &= ~PAGE_CACHE_MASK;
1571
1572 page_cache_release(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04001573 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 break;
Al Viro6e58e792014-02-03 17:07:03 -05001575 if (ret < nr) {
1576 error = -EFAULT;
1577 break;
1578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 cond_resched();
1580 }
1581
1582 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05001583 file_accessed(file);
1584 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585}
1586
Hugh Dickins708e3502011-07-25 17:12:32 -07001587static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1588 struct pipe_inode_info *pipe, size_t len,
1589 unsigned int flags)
1590{
1591 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001592 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001593 unsigned int loff, nr_pages, req_pages;
1594 struct page *pages[PIPE_DEF_BUFFERS];
1595 struct partial_page partial[PIPE_DEF_BUFFERS];
1596 struct page *page;
1597 pgoff_t index, end_index;
1598 loff_t isize, left;
1599 int error, page_nr;
1600 struct splice_pipe_desc spd = {
1601 .pages = pages,
1602 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001603 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001604 .flags = flags,
1605 .ops = &page_cache_pipe_buf_ops,
1606 .spd_release = spd_release_page,
1607 };
1608
Hugh Dickins71f0e072011-07-25 17:12:33 -07001609 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001610 if (unlikely(*ppos >= isize))
1611 return 0;
1612
1613 left = isize - *ppos;
1614 if (unlikely(left < len))
1615 len = left;
1616
1617 if (splice_grow_spd(pipe, &spd))
1618 return -ENOMEM;
1619
1620 index = *ppos >> PAGE_CACHE_SHIFT;
1621 loff = *ppos & ~PAGE_CACHE_MASK;
1622 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04001623 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07001624
Hugh Dickins708e3502011-07-25 17:12:32 -07001625 spd.nr_pages = find_get_pages_contig(mapping, index,
1626 nr_pages, spd.pages);
1627 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001628 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001629
Hugh Dickins708e3502011-07-25 17:12:32 -07001630 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001631 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1632 if (error)
1633 break;
1634 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001635 spd.pages[spd.nr_pages++] = page;
1636 index++;
1637 }
1638
Hugh Dickins708e3502011-07-25 17:12:32 -07001639 index = *ppos >> PAGE_CACHE_SHIFT;
1640 nr_pages = spd.nr_pages;
1641 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001642
Hugh Dickins708e3502011-07-25 17:12:32 -07001643 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1644 unsigned int this_len;
1645
1646 if (!len)
1647 break;
1648
Hugh Dickins708e3502011-07-25 17:12:32 -07001649 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1650 page = spd.pages[page_nr];
1651
Hugh Dickins71f0e072011-07-25 17:12:33 -07001652 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001653 error = shmem_getpage(inode, index, &page,
1654 SGP_CACHE, NULL);
1655 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001656 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001657 unlock_page(page);
1658 page_cache_release(spd.pages[page_nr]);
1659 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001660 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001661
1662 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001663 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1664 if (unlikely(!isize || index > end_index))
1665 break;
1666
Hugh Dickins708e3502011-07-25 17:12:32 -07001667 if (end_index == index) {
1668 unsigned int plen;
1669
Hugh Dickins708e3502011-07-25 17:12:32 -07001670 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1671 if (plen <= loff)
1672 break;
1673
Hugh Dickins708e3502011-07-25 17:12:32 -07001674 this_len = min(this_len, plen - loff);
1675 len = this_len;
1676 }
1677
1678 spd.partial[page_nr].offset = loff;
1679 spd.partial[page_nr].len = this_len;
1680 len -= this_len;
1681 loff = 0;
1682 spd.nr_pages++;
1683 index++;
1684 }
1685
Hugh Dickins708e3502011-07-25 17:12:32 -07001686 while (page_nr < nr_pages)
1687 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001688
1689 if (spd.nr_pages)
1690 error = splice_to_pipe(pipe, &spd);
1691
Eric Dumazet047fe362012-06-12 15:24:40 +02001692 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001693
1694 if (error > 0) {
1695 *ppos += error;
1696 file_accessed(in);
1697 }
1698 return error;
1699}
1700
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001701/*
1702 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1703 */
1704static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001705 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001706{
1707 struct page *page;
1708 struct pagevec pvec;
1709 pgoff_t indices[PAGEVEC_SIZE];
1710 bool done = false;
1711 int i;
1712
1713 pagevec_init(&pvec, 0);
1714 pvec.nr = 1; /* start small: we may be there already */
1715 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001716 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001717 pvec.nr, pvec.pages, indices);
1718 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001719 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001720 index = end;
1721 break;
1722 }
1723 for (i = 0; i < pvec.nr; i++, index++) {
1724 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001725 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001726 done = true;
1727 break;
1728 }
1729 index = indices[i];
1730 }
1731 page = pvec.pages[i];
1732 if (page && !radix_tree_exceptional_entry(page)) {
1733 if (!PageUptodate(page))
1734 page = NULL;
1735 }
1736 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001737 (page && whence == SEEK_DATA) ||
1738 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001739 done = true;
1740 break;
1741 }
1742 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001743 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001744 pagevec_release(&pvec);
1745 pvec.nr = PAGEVEC_SIZE;
1746 cond_resched();
1747 }
1748 return index;
1749}
1750
Andrew Morton965c8e52012-12-17 15:59:39 -08001751static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001752{
1753 struct address_space *mapping = file->f_mapping;
1754 struct inode *inode = mapping->host;
1755 pgoff_t start, end;
1756 loff_t new_offset;
1757
Andrew Morton965c8e52012-12-17 15:59:39 -08001758 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1759 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001760 MAX_LFS_FILESIZE, i_size_read(inode));
1761 mutex_lock(&inode->i_mutex);
1762 /* We're holding i_mutex so we can access i_size directly */
1763
1764 if (offset < 0)
1765 offset = -EINVAL;
1766 else if (offset >= inode->i_size)
1767 offset = -ENXIO;
1768 else {
1769 start = offset >> PAGE_CACHE_SHIFT;
1770 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001771 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001772 new_offset <<= PAGE_CACHE_SHIFT;
1773 if (new_offset > offset) {
1774 if (new_offset < inode->i_size)
1775 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001776 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001777 offset = -ENXIO;
1778 else
1779 offset = inode->i_size;
1780 }
1781 }
1782
Hugh Dickins387aae62013-08-04 11:30:25 -07001783 if (offset >= 0)
1784 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001785 mutex_unlock(&inode->i_mutex);
1786 return offset;
1787}
1788
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001789static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1790 loff_t len)
1791{
Al Viro496ad9a2013-01-23 17:07:38 -05001792 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001793 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001794 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001795 pgoff_t start, index, end;
1796 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001797
Hugh Dickins13ace4d2014-06-23 13:22:03 -07001798 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1799 return -EOPNOTSUPP;
1800
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001801 mutex_lock(&inode->i_mutex);
1802
1803 if (mode & FALLOC_FL_PUNCH_HOLE) {
1804 struct address_space *mapping = file->f_mapping;
1805 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1806 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001807 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001808
Hugh Dickins8e205f72014-07-23 14:00:10 -07001809 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001810 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1811 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1812 spin_lock(&inode->i_lock);
1813 inode->i_private = &shmem_falloc;
1814 spin_unlock(&inode->i_lock);
1815
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001816 if ((u64)unmap_end > (u64)unmap_start)
1817 unmap_mapping_range(mapping, unmap_start,
1818 1 + unmap_end - unmap_start, 0);
1819 shmem_truncate_range(inode, offset, offset + len - 1);
1820 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07001821
1822 spin_lock(&inode->i_lock);
1823 inode->i_private = NULL;
1824 wake_up_all(&shmem_falloc_waitq);
1825 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001826 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001827 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001828 }
1829
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001830 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1831 error = inode_newsize_ok(inode, offset + len);
1832 if (error)
1833 goto out;
1834
1835 start = offset >> PAGE_CACHE_SHIFT;
1836 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1837 /* Try to avoid a swapstorm if len is impossible to satisfy */
1838 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1839 error = -ENOSPC;
1840 goto out;
1841 }
1842
Hugh Dickins8e205f72014-07-23 14:00:10 -07001843 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001844 shmem_falloc.start = start;
1845 shmem_falloc.next = start;
1846 shmem_falloc.nr_falloced = 0;
1847 shmem_falloc.nr_unswapped = 0;
1848 spin_lock(&inode->i_lock);
1849 inode->i_private = &shmem_falloc;
1850 spin_unlock(&inode->i_lock);
1851
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001852 for (index = start; index < end; index++) {
1853 struct page *page;
1854
1855 /*
1856 * Good, the fallocate(2) manpage permits EINTR: we may have
1857 * been interrupted because we are using up too much memory.
1858 */
1859 if (signal_pending(current))
1860 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001861 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1862 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001863 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001864 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001865 NULL);
1866 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001867 /* Remove the !PageUptodate pages we added */
1868 shmem_undo_range(inode,
1869 (loff_t)start << PAGE_CACHE_SHIFT,
1870 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001871 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001872 }
1873
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001874 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07001875 * Inform shmem_writepage() how far we have reached.
1876 * No need for lock or barrier: we have the page lock.
1877 */
1878 shmem_falloc.next++;
1879 if (!PageUptodate(page))
1880 shmem_falloc.nr_falloced++;
1881
1882 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001883 * If !PageUptodate, leave it that way so that freeable pages
1884 * can be recognized if we need to rollback on error later.
1885 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001886 * than free the pages we are allocating (and SGP_CACHE pages
1887 * might still be clean: we now need to mark those dirty too).
1888 */
1889 set_page_dirty(page);
1890 unlock_page(page);
1891 page_cache_release(page);
1892 cond_resched();
1893 }
1894
1895 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1896 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001897 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001898undone:
1899 spin_lock(&inode->i_lock);
1900 inode->i_private = NULL;
1901 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001902out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001903 mutex_unlock(&inode->i_mutex);
1904 return error;
1905}
1906
David Howells726c3342006-06-23 02:02:58 -07001907static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908{
David Howells726c3342006-06-23 02:02:58 -07001909 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
1911 buf->f_type = TMPFS_MAGIC;
1912 buf->f_bsize = PAGE_CACHE_SIZE;
1913 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001914 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001916 buf->f_bavail =
1917 buf->f_bfree = sbinfo->max_blocks -
1918 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001919 }
1920 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 buf->f_files = sbinfo->max_inodes;
1922 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 }
1924 /* else leave those fields 0 like simple_statfs */
1925 return 0;
1926}
1927
1928/*
1929 * File creation. Allocate an inode, and we're done..
1930 */
1931static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001932shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001934 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 int error = -ENOSPC;
1936
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001937 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001939 error = simple_acl_create(dir, inode);
1940 if (error)
1941 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05001942 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001943 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001944 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001945 if (error && error != -EOPNOTSUPP)
1946 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04001947
Al Viro718deb62009-12-16 19:35:36 -05001948 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 dir->i_size += BOGO_DIRENT_SIZE;
1950 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1951 d_instantiate(dentry, inode);
1952 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 }
1954 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001955out_iput:
1956 iput(inode);
1957 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958}
1959
Al Viro60545d02013-06-07 01:20:27 -04001960static int
1961shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1962{
1963 struct inode *inode;
1964 int error = -ENOSPC;
1965
1966 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1967 if (inode) {
1968 error = security_inode_init_security(inode, dir,
1969 NULL,
1970 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001971 if (error && error != -EOPNOTSUPP)
1972 goto out_iput;
1973 error = simple_acl_create(dir, inode);
1974 if (error)
1975 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04001976 d_tmpfile(dentry, inode);
1977 }
1978 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001979out_iput:
1980 iput(inode);
1981 return error;
Al Viro60545d02013-06-07 01:20:27 -04001982}
1983
Al Viro18bb1db2011-07-26 01:41:39 -04001984static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985{
1986 int error;
1987
1988 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1989 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001990 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 return 0;
1992}
1993
Al Viro4acdaf22011-07-26 01:42:34 -04001994static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04001995 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996{
1997 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1998}
1999
2000/*
2001 * Link a file..
2002 */
2003static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2004{
2005 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002006 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
2008 /*
2009 * No ordinary (disk based) filesystem counts links as inodes;
2010 * but each new link needs a new dentry, pinning lowmem, and
2011 * tmpfs dentries cannot be pruned until they are unlinked.
2012 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002013 ret = shmem_reserve_inode(inode->i_sb);
2014 if (ret)
2015 goto out;
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 Hansend8c76e62006-09-30 23:29:04 -07002019 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002020 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 dget(dentry); /* Extra pinning count for the created dentry */
2022 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002023out:
2024 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025}
2026
2027static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2028{
2029 struct inode *inode = dentry->d_inode;
2030
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002031 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2032 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
2034 dir->i_size -= BOGO_DIRENT_SIZE;
2035 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002036 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 dput(dentry); /* Undo the count from "create" - this does all the work */
2038 return 0;
2039}
2040
2041static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2042{
2043 if (!simple_empty(dentry))
2044 return -ENOTEMPTY;
2045
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002046 drop_nlink(dentry->d_inode);
2047 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return shmem_unlink(dir, dentry);
2049}
2050
2051/*
2052 * The VFS layer already does all the dentry stuff for rename,
2053 * we just have to decrement the usage count for the target if
2054 * it exists so that the VFS layer correctly free's it when it
2055 * gets overwritten.
2056 */
2057static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2058{
2059 struct inode *inode = old_dentry->d_inode;
2060 int they_are_dirs = S_ISDIR(inode->i_mode);
2061
2062 if (!simple_empty(new_dentry))
2063 return -ENOTEMPTY;
2064
2065 if (new_dentry->d_inode) {
2066 (void) shmem_unlink(new_dir, new_dentry);
2067 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002068 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002070 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002071 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073
2074 old_dir->i_size -= BOGO_DIRENT_SIZE;
2075 new_dir->i_size += BOGO_DIRENT_SIZE;
2076 old_dir->i_ctime = old_dir->i_mtime =
2077 new_dir->i_ctime = new_dir->i_mtime =
2078 inode->i_ctime = CURRENT_TIME;
2079 return 0;
2080}
2081
2082static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2083{
2084 int error;
2085 int len;
2086 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002087 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 char *kaddr;
2089 struct shmem_inode_info *info;
2090
2091 len = strlen(symname) + 1;
2092 if (len > PAGE_CACHE_SIZE)
2093 return -ENAMETOOLONG;
2094
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002095 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 if (!inode)
2097 return -ENOSPC;
2098
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002099 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002100 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002101 if (error) {
2102 if (error != -EOPNOTSUPP) {
2103 iput(inode);
2104 return error;
2105 }
2106 error = 0;
2107 }
2108
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 info = SHMEM_I(inode);
2110 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002111 if (len <= SHORT_SYMLINK_LEN) {
2112 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2113 if (!info->symlink) {
2114 iput(inode);
2115 return -ENOMEM;
2116 }
2117 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 } else {
2119 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2120 if (error) {
2121 iput(inode);
2122 return error;
2123 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002124 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002126 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002128 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002129 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002131 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 page_cache_release(page);
2133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 dir->i_size += BOGO_DIRENT_SIZE;
2135 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2136 d_instantiate(dentry, inode);
2137 dget(dentry);
2138 return 0;
2139}
2140
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002141static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002143 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002144 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145}
2146
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002147static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148{
2149 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002150 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2151 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002152 if (page)
2153 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002154 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155}
2156
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002157static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158{
2159 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002160 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 kunmap(page);
2162 mark_page_accessed(page);
2163 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 }
2165}
2166
Eric Parisb09e0fa2011-05-24 17:12:39 -07002167#ifdef CONFIG_TMPFS_XATTR
2168/*
2169 * Superblocks without xattr inode operations may get some security.* xattr
2170 * support from the LSM "for free". As soon as we have any other xattrs
2171 * like ACLs, we also need to implement the security.* handlers at
2172 * filesystem level, though.
2173 */
2174
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002175/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002176 * Callback for security_inode_init_security() for acquiring xattrs.
2177 */
2178static int shmem_initxattrs(struct inode *inode,
2179 const struct xattr *xattr_array,
2180 void *fs_info)
2181{
2182 struct shmem_inode_info *info = SHMEM_I(inode);
2183 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002184 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002185 size_t len;
2186
2187 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002188 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002189 if (!new_xattr)
2190 return -ENOMEM;
2191
2192 len = strlen(xattr->name) + 1;
2193 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2194 GFP_KERNEL);
2195 if (!new_xattr->name) {
2196 kfree(new_xattr);
2197 return -ENOMEM;
2198 }
2199
2200 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2201 XATTR_SECURITY_PREFIX_LEN);
2202 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2203 xattr->name, len);
2204
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002205 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002206 }
2207
2208 return 0;
2209}
2210
Eric Parisb09e0fa2011-05-24 17:12:39 -07002211static const struct xattr_handler *shmem_xattr_handlers[] = {
2212#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002213 &posix_acl_access_xattr_handler,
2214 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002215#endif
2216 NULL
2217};
2218
2219static int shmem_xattr_validate(const char *name)
2220{
2221 struct { const char *prefix; size_t len; } arr[] = {
2222 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2223 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2224 };
2225 int i;
2226
2227 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2228 size_t preflen = arr[i].len;
2229 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2230 if (!name[preflen])
2231 return -EINVAL;
2232 return 0;
2233 }
2234 }
2235 return -EOPNOTSUPP;
2236}
2237
2238static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2239 void *buffer, size_t size)
2240{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002241 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002242 int err;
2243
2244 /*
2245 * If this is a request for a synthetic attribute in the system.*
2246 * namespace use the generic infrastructure to resolve a handler
2247 * for it via sb->s_xattr.
2248 */
2249 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2250 return generic_getxattr(dentry, name, buffer, size);
2251
2252 err = shmem_xattr_validate(name);
2253 if (err)
2254 return err;
2255
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002256 return simple_xattr_get(&info->xattrs, name, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002257}
2258
2259static int shmem_setxattr(struct dentry *dentry, const char *name,
2260 const void *value, size_t size, int flags)
2261{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002262 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002263 int err;
2264
2265 /*
2266 * If this is a request for a synthetic attribute in the system.*
2267 * namespace use the generic infrastructure to resolve a handler
2268 * for it via sb->s_xattr.
2269 */
2270 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2271 return generic_setxattr(dentry, name, value, size, flags);
2272
2273 err = shmem_xattr_validate(name);
2274 if (err)
2275 return err;
2276
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002277 return simple_xattr_set(&info->xattrs, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002278}
2279
2280static int shmem_removexattr(struct dentry *dentry, const char *name)
2281{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002282 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002283 int err;
2284
2285 /*
2286 * If this is a request for a synthetic attribute in the system.*
2287 * namespace use the generic infrastructure to resolve a handler
2288 * for it via sb->s_xattr.
2289 */
2290 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2291 return generic_removexattr(dentry, name);
2292
2293 err = shmem_xattr_validate(name);
2294 if (err)
2295 return err;
2296
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002297 return simple_xattr_remove(&info->xattrs, name);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002298}
2299
2300static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2301{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002302 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2303 return simple_xattr_list(&info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002304}
2305#endif /* CONFIG_TMPFS_XATTR */
2306
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002307static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002309 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002310#ifdef CONFIG_TMPFS_XATTR
2311 .setxattr = shmem_setxattr,
2312 .getxattr = shmem_getxattr,
2313 .listxattr = shmem_listxattr,
2314 .removexattr = shmem_removexattr,
2315#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316};
2317
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002318static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 .readlink = generic_readlink,
2320 .follow_link = shmem_follow_link,
2321 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002322#ifdef CONFIG_TMPFS_XATTR
2323 .setxattr = shmem_setxattr,
2324 .getxattr = shmem_getxattr,
2325 .listxattr = shmem_listxattr,
2326 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002327#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002328};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002329
David M. Grimes91828a42006-10-17 00:09:45 -07002330static struct dentry *shmem_get_parent(struct dentry *child)
2331{
2332 return ERR_PTR(-ESTALE);
2333}
2334
2335static int shmem_match(struct inode *ino, void *vfh)
2336{
2337 __u32 *fh = vfh;
2338 __u64 inum = fh[2];
2339 inum = (inum << 32) | fh[1];
2340 return ino->i_ino == inum && fh[0] == ino->i_generation;
2341}
2342
Christoph Hellwig480b1162007-10-21 16:42:13 -07002343static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2344 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002345{
David M. Grimes91828a42006-10-17 00:09:45 -07002346 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002347 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002348 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002349
Christoph Hellwig480b1162007-10-21 16:42:13 -07002350 if (fh_len < 3)
2351 return NULL;
2352
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002353 inum = fid->raw[2];
2354 inum = (inum << 32) | fid->raw[1];
2355
Christoph Hellwig480b1162007-10-21 16:42:13 -07002356 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2357 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002358 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002359 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002360 iput(inode);
2361 }
2362
Christoph Hellwig480b1162007-10-21 16:42:13 -07002363 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002364}
2365
Al Virob0b03822012-04-02 14:34:06 -04002366static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2367 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002368{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302369 if (*len < 3) {
2370 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002371 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302372 }
David M. Grimes91828a42006-10-17 00:09:45 -07002373
Al Viro1d3382cb2010-10-23 15:19:20 -04002374 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002375 /* Unfortunately insert_inode_hash is not idempotent,
2376 * so as we hash inodes here rather than at creation
2377 * time, we need a lock to ensure we only try
2378 * to do it once
2379 */
2380 static DEFINE_SPINLOCK(lock);
2381 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002382 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002383 __insert_inode_hash(inode,
2384 inode->i_ino + inode->i_generation);
2385 spin_unlock(&lock);
2386 }
2387
2388 fh[0] = inode->i_generation;
2389 fh[1] = inode->i_ino;
2390 fh[2] = ((__u64)inode->i_ino) >> 32;
2391
2392 *len = 3;
2393 return 1;
2394}
2395
Christoph Hellwig39655162007-10-21 16:42:17 -07002396static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002397 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002398 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002399 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002400};
2401
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002402static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2403 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404{
2405 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002406 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002407 uid_t uid;
2408 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002410 while (options != NULL) {
2411 this_char = options;
2412 for (;;) {
2413 /*
2414 * NUL-terminate this option: unfortunately,
2415 * mount options form a comma-separated list,
2416 * but mpol's nodelist may also contain commas.
2417 */
2418 options = strchr(options, ',');
2419 if (options == NULL)
2420 break;
2421 options++;
2422 if (!isdigit(*options)) {
2423 options[-1] = '\0';
2424 break;
2425 }
2426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 if (!*this_char)
2428 continue;
2429 if ((value = strchr(this_char,'=')) != NULL) {
2430 *value++ = 0;
2431 } else {
2432 printk(KERN_ERR
2433 "tmpfs: No value for mount option '%s'\n",
2434 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002435 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 }
2437
2438 if (!strcmp(this_char,"size")) {
2439 unsigned long long size;
2440 size = memparse(value,&rest);
2441 if (*rest == '%') {
2442 size <<= PAGE_SHIFT;
2443 size *= totalram_pages;
2444 do_div(size, 100);
2445 rest++;
2446 }
2447 if (*rest)
2448 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002449 sbinfo->max_blocks =
2450 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002452 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 if (*rest)
2454 goto bad_val;
2455 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002456 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 if (*rest)
2458 goto bad_val;
2459 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002460 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002462 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 if (*rest)
2464 goto bad_val;
2465 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002466 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002468 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 if (*rest)
2470 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002471 sbinfo->uid = make_kuid(current_user_ns(), uid);
2472 if (!uid_valid(sbinfo->uid))
2473 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002475 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002477 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 if (*rest)
2479 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002480 sbinfo->gid = make_kgid(current_user_ns(), gid);
2481 if (!gid_valid(sbinfo->gid))
2482 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002483 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002484 mpol_put(mpol);
2485 mpol = NULL;
2486 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002487 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 } else {
2489 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2490 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002491 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 }
2493 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002494 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 return 0;
2496
2497bad_val:
2498 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2499 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002500error:
2501 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 return 1;
2503
2504}
2505
2506static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2507{
2508 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002509 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002510 unsigned long inodes;
2511 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
Greg Thelen5f001102013-02-22 16:36:01 -08002513 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002514 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002515 return error;
2516
2517 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002518 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002519 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002520 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002521 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002522 goto out;
2523 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002524 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002525 * but we must separately disallow unlimited->limited, because
2526 * in that case we have no record of how much is already in use.
2527 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002528 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002529 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002530 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002531 goto out;
2532
2533 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002534 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002535 sbinfo->max_inodes = config.max_inodes;
2536 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002537
Greg Thelen5f001102013-02-22 16:36:01 -08002538 /*
2539 * Preserve previous mempolicy unless mpol remount option was specified.
2540 */
2541 if (config.mpol) {
2542 mpol_put(sbinfo->mpol);
2543 sbinfo->mpol = config.mpol; /* transfers initial ref */
2544 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002545out:
2546 spin_unlock(&sbinfo->stat_lock);
2547 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002549
Al Viro34c80b12011-12-08 21:32:45 -05002550static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002551{
Al Viro34c80b12011-12-08 21:32:45 -05002552 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002553
2554 if (sbinfo->max_blocks != shmem_default_max_blocks())
2555 seq_printf(seq, ",size=%luk",
2556 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2557 if (sbinfo->max_inodes != shmem_default_max_inodes())
2558 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2559 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002560 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002561 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2562 seq_printf(seq, ",uid=%u",
2563 from_kuid_munged(&init_user_ns, sbinfo->uid));
2564 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2565 seq_printf(seq, ",gid=%u",
2566 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002567 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002568 return 0;
2569}
2570#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
2572static void shmem_put_super(struct super_block *sb)
2573{
Hugh Dickins602586a2010-08-17 15:23:56 -07002574 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2575
2576 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002577 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07002578 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 sb->s_fs_info = NULL;
2580}
2581
Kay Sievers2b2af542009-04-30 15:23:42 +02002582int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583{
2584 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002585 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002586 int err = -ENOMEM;
2587
2588 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002589 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002590 L1_CACHE_BYTES), GFP_KERNEL);
2591 if (!sbinfo)
2592 return -ENOMEM;
2593
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002594 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002595 sbinfo->uid = current_fsuid();
2596 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002597 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002599#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 /*
2601 * Per default we only allow half of the physical ram per
2602 * tmpfs instance, limiting inodes to one per page of lowmem;
2603 * but the internal instance is left unlimited.
2604 */
Al Viroca4e0512013-08-31 12:57:10 -04002605 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002606 sbinfo->max_blocks = shmem_default_max_blocks();
2607 sbinfo->max_inodes = shmem_default_max_inodes();
2608 if (shmem_parse_options(data, sbinfo, false)) {
2609 err = -EINVAL;
2610 goto failed;
2611 }
Al Viroca4e0512013-08-31 12:57:10 -04002612 } else {
2613 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 }
David M. Grimes91828a42006-10-17 00:09:45 -07002615 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07002616 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617#else
2618 sb->s_flags |= MS_NOUSER;
2619#endif
2620
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002621 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002622 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2623 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002624 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002625
Hugh Dickins285b2c42011-08-03 16:21:20 -07002626 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 sb->s_blocksize = PAGE_CACHE_SIZE;
2628 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2629 sb->s_magic = TMPFS_MAGIC;
2630 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002631 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002632#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002633 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002634#endif
2635#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002636 sb->s_flags |= MS_POSIXACL;
2637#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002638
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002639 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 if (!inode)
2641 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002642 inode->i_uid = sbinfo->uid;
2643 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002644 sb->s_root = d_make_root(inode);
2645 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002646 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 return 0;
2648
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649failed:
2650 shmem_put_super(sb);
2651 return err;
2652}
2653
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002654static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
2656static struct inode *shmem_alloc_inode(struct super_block *sb)
2657{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002658 struct shmem_inode_info *info;
2659 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2660 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002662 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663}
2664
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002665static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002666{
2667 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002668 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2669}
2670
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671static void shmem_destroy_inode(struct inode *inode)
2672{
Al Viro09208d12011-07-26 03:15:03 -04002673 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002675 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676}
2677
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002678static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002680 struct shmem_inode_info *info = foo;
2681 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682}
2683
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002684static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685{
2686 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2687 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002688 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 return 0;
2690}
2691
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002692static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002694 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695}
2696
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002697static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002699 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002701 .write_begin = shmem_write_begin,
2702 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002704 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002705 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706};
2707
Helge Deller15ad7cd2006-12-06 20:40:36 -08002708static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 .mmap = shmem_mmap,
2710#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002711 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04002712 .read = new_sync_read,
Al Viro81742022014-04-03 03:17:43 -04002713 .write = new_sync_write,
Al Viro2ba5bbe2014-04-02 20:00:02 -04002714 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04002715 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002716 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002717 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04002718 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002719 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720#endif
2721};
2722
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002723static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002724 .setattr = shmem_setattr,
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,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002730 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002731#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732};
2733
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002734static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735#ifdef CONFIG_TMPFS
2736 .create = shmem_create,
2737 .lookup = simple_lookup,
2738 .link = shmem_link,
2739 .unlink = shmem_unlink,
2740 .symlink = shmem_symlink,
2741 .mkdir = shmem_mkdir,
2742 .rmdir = shmem_rmdir,
2743 .mknod = shmem_mknod,
2744 .rename = shmem_rename,
Al Viro60545d02013-06-07 01:20:27 -04002745 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002747#ifdef CONFIG_TMPFS_XATTR
2748 .setxattr = shmem_setxattr,
2749 .getxattr = shmem_getxattr,
2750 .listxattr = shmem_listxattr,
2751 .removexattr = shmem_removexattr,
2752#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002753#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002754 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002755 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002756#endif
2757};
2758
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002759static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002760#ifdef CONFIG_TMPFS_XATTR
2761 .setxattr = shmem_setxattr,
2762 .getxattr = shmem_getxattr,
2763 .listxattr = shmem_listxattr,
2764 .removexattr = shmem_removexattr,
2765#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002766#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002767 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002768 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002769#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770};
2771
Hugh Dickins759b9772007-03-05 00:30:28 -08002772static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 .alloc_inode = shmem_alloc_inode,
2774 .destroy_inode = shmem_destroy_inode,
2775#ifdef CONFIG_TMPFS
2776 .statfs = shmem_statfs,
2777 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002778 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779#endif
Al Viro1f895f72010-06-05 19:10:41 -04002780 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 .drop_inode = generic_delete_inode,
2782 .put_super = shmem_put_super,
2783};
2784
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002785static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002786 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07002787 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788#ifdef CONFIG_NUMA
2789 .set_policy = shmem_set_policy,
2790 .get_policy = shmem_get_policy,
2791#endif
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07002792 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793};
2794
Al Viro3c26ff62010-07-25 11:46:36 +04002795static struct dentry *shmem_mount(struct file_system_type *fs_type,
2796 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797{
Al Viro3c26ff62010-07-25 11:46:36 +04002798 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799}
2800
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002801static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 .owner = THIS_MODULE,
2803 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002804 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08002806 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002809int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810{
2811 int error;
2812
Rob Landley16203a72013-09-11 14:26:12 -07002813 /* If rootfs called this, don't re-init */
2814 if (shmem_inode_cachep)
2815 return 0;
2816
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002817 error = bdi_init(&shmem_backing_dev_info);
2818 if (error)
2819 goto out4;
2820
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002821 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 if (error)
2823 goto out3;
2824
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002825 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 if (error) {
2827 printk(KERN_ERR "Could not register tmpfs\n");
2828 goto out2;
2829 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002830
Al Viroca4e0512013-08-31 12:57:10 -04002831 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 if (IS_ERR(shm_mnt)) {
2833 error = PTR_ERR(shm_mnt);
2834 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2835 goto out1;
2836 }
2837 return 0;
2838
2839out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002840 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002842 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002844 bdi_destroy(&shmem_backing_dev_info);
2845out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 shm_mnt = ERR_PTR(error);
2847 return error;
2848}
Matt Mackall853ac432009-01-06 14:40:20 -08002849
2850#else /* !CONFIG_SHMEM */
2851
2852/*
2853 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2854 *
2855 * This is intended for small system where the benefits of the full
2856 * shmem code (swap-backed and resource-limited) are outweighed by
2857 * their complexity. On systems without swap this code should be
2858 * effectively equivalent, but much lighter weight.
2859 */
2860
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002861static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002862 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002863 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002864 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08002865 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08002866};
2867
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002868int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002869{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002870 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002871
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002872 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002873 BUG_ON(IS_ERR(shm_mnt));
2874
2875 return 0;
2876}
2877
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002878int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002879{
2880 return 0;
2881}
2882
Hugh Dickins3f96b792009-09-21 17:03:37 -07002883int shmem_lock(struct file *file, int lock, struct user_struct *user)
2884{
2885 return 0;
2886}
2887
Hugh Dickins24513262012-01-20 14:34:21 -08002888void shmem_unlock_mapping(struct address_space *mapping)
2889{
2890}
2891
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002892void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002893{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002894 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002895}
2896EXPORT_SYMBOL_GPL(shmem_truncate_range);
2897
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002898#define shmem_vm_ops generic_file_vm_ops
2899#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002900#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002901#define shmem_acct_size(flags, size) 0
2902#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002903
2904#endif /* CONFIG_SHMEM */
2905
2906/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
Al Viro34515382013-02-14 22:38:02 -05002908static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04002909 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05002910};
2911
Eric Parisc7277092013-12-02 11:24:19 +00002912static struct file *__shmem_file_setup(const char *name, loff_t size,
2913 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914{
Al Viro6b4d0b22013-02-14 21:37:26 -05002915 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002917 struct path path;
Al Viro34515382013-02-14 22:38:02 -05002918 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 struct qstr this;
2920
2921 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05002922 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
Hugh Dickins285b2c42011-08-03 16:21:20 -07002924 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 return ERR_PTR(-EINVAL);
2926
2927 if (shmem_acct_size(flags, size))
2928 return ERR_PTR(-ENOMEM);
2929
Al Viro6b4d0b22013-02-14 21:37:26 -05002930 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 this.name = name;
2932 this.len = strlen(name);
2933 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05002934 sb = shm_mnt->mnt_sb;
2935 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04002936 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05002938 d_set_d_op(path.dentry, &anon_ops);
Al Viro2c48b9c2009-08-09 00:52:35 +04002939 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940
Al Viro6b4d0b22013-02-14 21:37:26 -05002941 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05002942 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002944 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945
Eric Parisc7277092013-12-02 11:24:19 +00002946 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04002947 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002949 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05002950 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2951 if (IS_ERR(res))
Al Viro4b42af82009-08-05 18:25:56 +04002952 goto put_dentry;
Al Viro4b42af82009-08-05 18:25:56 +04002953
Al Viro6b4d0b22013-02-14 21:37:26 -05002954 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002955 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05002956 if (IS_ERR(res))
Al Viro4b42af82009-08-05 18:25:56 +04002957 goto put_dentry;
2958
Al Viro6b4d0b22013-02-14 21:37:26 -05002959 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002962 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963put_memory:
2964 shmem_unacct_size(flags, size);
Al Viro6b4d0b22013-02-14 21:37:26 -05002965 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966}
Eric Parisc7277092013-12-02 11:24:19 +00002967
2968/**
2969 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2970 * kernel internal. There will be NO LSM permission checks against the
2971 * underlying inode. So users of this interface must do LSM checks at a
2972 * higher layer. The one user is the big_key implementation. LSM checks
2973 * are provided at the key level rather than the inode level.
2974 * @name: name for dentry (to be seen in /proc/<pid>/maps
2975 * @size: size to be set for the file
2976 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2977 */
2978struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2979{
2980 return __shmem_file_setup(name, size, flags, S_PRIVATE);
2981}
2982
2983/**
2984 * shmem_file_setup - get an unlinked file living in tmpfs
2985 * @name: name for dentry (to be seen in /proc/<pid>/maps
2986 * @size: size to be set for the file
2987 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2988 */
2989struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2990{
2991 return __shmem_file_setup(name, size, flags, 0);
2992}
Keith Packard395e0dd2008-06-20 00:08:06 -07002993EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994
Randy Dunlap46711812008-03-19 17:00:41 -07002995/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2998 */
2999int shmem_zero_setup(struct vm_area_struct *vma)
3000{
3001 struct file *file;
3002 loff_t size = vma->vm_end - vma->vm_start;
3003
3004 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3005 if (IS_ERR(file))
3006 return PTR_ERR(file);
3007
3008 if (vma->vm_file)
3009 fput(vma->vm_file);
3010 vma->vm_file = file;
3011 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 return 0;
3013}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003014
3015/**
3016 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3017 * @mapping: the page's address_space
3018 * @index: the page index
3019 * @gfp: the page allocator flags to use if allocating
3020 *
3021 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3022 * with any new page allocations done using the specified allocation flags.
3023 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3024 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3025 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3026 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003027 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3028 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003029 */
3030struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3031 pgoff_t index, gfp_t gfp)
3032{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003033#ifdef CONFIG_SHMEM
3034 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003035 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003036 int error;
3037
3038 BUG_ON(mapping->a_ops != &shmem_aops);
3039 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3040 if (error)
3041 page = ERR_PTR(error);
3042 else
3043 unlock_page(page);
3044 return page;
3045#else
3046 /*
3047 * The tiny !SHMEM case uses ramfs without swap
3048 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003049 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003050#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003051}
3052EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);