blob: 171dee7a131f6959fe444405f280374be7eb25d0 [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>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080034#include <linux/uio.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>
David Herrmann9183df22014-08-08 14:25:29 -070069#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070070#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070071#include <uapi/linux/memfd.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070072
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <asm/pgtable.h>
75
Mel Gormandd56b042015-11-06 16:28:43 -080076#include "internal.h"
77
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030078#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
79#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/* Pretend that each entry is of this size in directory's i_size */
82#define BOGO_DIRENT_SIZE 20
83
Hugh Dickins69f07ec2011-08-03 16:21:26 -070084/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
85#define SHORT_SYMLINK_LEN 128
86
Hugh Dickins1aac1402012-05-29 15:06:42 -070087/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070088 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
89 * inode->i_private (with i_mutex making sure that it has only one user at
90 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070091 */
92struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -070093 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -070094 pgoff_t start; /* start of range currently being fallocated */
95 pgoff_t next; /* the next page offset to be fallocated */
96 pgoff_t nr_falloced; /* how many new pages have been fallocated */
97 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
98};
99
Hugh Dickins285b2c42011-08-03 16:21:20 -0700100/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 SGP_READ, /* don't exceed i_size, don't allocate page */
103 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700104 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
105 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106};
107
Andrew Mortonb76db732008-02-08 04:21:49 -0800108#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800109static unsigned long shmem_default_max_blocks(void)
110{
111 return totalram_pages / 2;
112}
113
114static unsigned long shmem_default_max_inodes(void)
115{
116 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117}
Andrew Mortonb76db732008-02-08 04:21:49 -0800118#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800119
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700120static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700123static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700124 struct page **pagep, enum sgp_type sgp,
125 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700126
127static inline int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700128 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700129{
130 return shmem_getpage_gfp(inode, index, pagep, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700131 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700132}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
135{
136 return sb->s_fs_info;
137}
138
139/*
140 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141 * for shared memory and for shared anonymous (/dev/zero) mappings
142 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143 * consistent with the pre-accounting of private mappings ...
144 */
145static inline int shmem_acct_size(unsigned long flags, loff_t size)
146{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000147 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000148 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151static inline void shmem_unacct_size(unsigned long flags, loff_t size)
152{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000153 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 vm_unacct_memory(VM_ACCT(size));
155}
156
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700157static inline int shmem_reacct_size(unsigned long flags,
158 loff_t oldsize, loff_t newsize)
159{
160 if (!(flags & VM_NORESERVE)) {
161 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
162 return security_vm_enough_memory_mm(current->mm,
163 VM_ACCT(newsize) - VM_ACCT(oldsize));
164 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
165 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
166 }
167 return 0;
168}
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/*
171 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700172 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
175 */
176static inline int shmem_acct_block(unsigned long flags)
177{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000178 return (flags & VM_NORESERVE) ?
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300179 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180}
181
182static inline void shmem_unacct_blocks(unsigned long flags, long pages)
183{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000184 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300185 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
Hugh Dickins759b9772007-03-05 00:30:28 -0800188static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700189static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800190static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800191static const struct inode_operations shmem_inode_operations;
192static const struct inode_operations shmem_dir_inode_operations;
193static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400194static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800197static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800199static int shmem_reserve_inode(struct super_block *sb)
200{
201 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
202 if (sbinfo->max_inodes) {
203 spin_lock(&sbinfo->stat_lock);
204 if (!sbinfo->free_inodes) {
205 spin_unlock(&sbinfo->stat_lock);
206 return -ENOSPC;
207 }
208 sbinfo->free_inodes--;
209 spin_unlock(&sbinfo->stat_lock);
210 }
211 return 0;
212}
213
214static void shmem_free_inode(struct super_block *sb)
215{
216 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
217 if (sbinfo->max_inodes) {
218 spin_lock(&sbinfo->stat_lock);
219 sbinfo->free_inodes++;
220 spin_unlock(&sbinfo->stat_lock);
221 }
222}
223
Randy Dunlap46711812008-03-19 17:00:41 -0700224/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700225 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 * @inode: inode to recalc
227 *
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
230 *
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
233 *
234 * It has to be called with the spinlock held.
235 */
236static void shmem_recalc_inode(struct inode *inode)
237{
238 struct shmem_inode_info *info = SHMEM_I(inode);
239 long freed;
240
241 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
242 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700243 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
244 if (sbinfo->max_blocks)
245 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700247 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
250}
251
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700252/*
253 * Replace item expected in radix tree by a new item, while holding tree lock.
254 */
255static int shmem_radix_tree_replace(struct address_space *mapping,
256 pgoff_t index, void *expected, void *replacement)
257{
258 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700259 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700260
261 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700262 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700263 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700264 if (!pslot)
265 return -ENOENT;
266 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700267 if (item != expected)
268 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700269 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700270 return 0;
271}
272
273/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
276 *
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
279 */
280static bool shmem_confirm_swap(struct address_space *mapping,
281 pgoff_t index, swp_entry_t swap)
282{
283 void *item;
284
285 rcu_read_lock();
286 item = radix_tree_lookup(&mapping->page_tree, index);
287 rcu_read_unlock();
288 return item == swp_to_radix_entry(swap);
289}
290
291/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700292 * Like add_to_page_cache_locked, but error if expected item has gone.
293 */
294static int shmem_add_to_page_cache(struct page *page,
295 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700296 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700297{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700298 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700299
Sasha Levin309381fea2014-01-23 15:52:54 -0800300 VM_BUG_ON_PAGE(!PageLocked(page), page);
301 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700302
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300303 get_page(page);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700304 page->mapping = mapping;
305 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700306
Hugh Dickinsb065b432012-07-11 14:02:48 -0700307 spin_lock_irq(&mapping->tree_lock);
308 if (!expected)
309 error = radix_tree_insert(&mapping->page_tree, index, page);
310 else
311 error = shmem_radix_tree_replace(mapping, index, expected,
312 page);
313 if (!error) {
314 mapping->nrpages++;
315 __inc_zone_page_state(page, NR_FILE_PAGES);
316 __inc_zone_page_state(page, NR_SHMEM);
317 spin_unlock_irq(&mapping->tree_lock);
318 } else {
319 page->mapping = NULL;
320 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300321 put_page(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700322 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700323 return error;
324}
325
326/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700327 * Like delete_from_page_cache, but substitutes swap for page.
328 */
329static void shmem_delete_from_page_cache(struct page *page, void *radswap)
330{
331 struct address_space *mapping = page->mapping;
332 int error;
333
334 spin_lock_irq(&mapping->tree_lock);
335 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
336 page->mapping = NULL;
337 mapping->nrpages--;
338 __dec_zone_page_state(page, NR_FILE_PAGES);
339 __dec_zone_page_state(page, NR_SHMEM);
340 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300341 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700342 BUG_ON(error);
343}
344
345/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700346 * Remove swap entry from radix tree, free the swap and its page cache.
347 */
348static int shmem_free_swap(struct address_space *mapping,
349 pgoff_t index, void *radswap)
350{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700351 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700352
353 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700354 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700355 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700356 if (old != radswap)
357 return -ENOENT;
358 free_swap_and_cache(radix_to_swp_entry(radswap));
359 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700360}
361
362/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800363 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800364 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800365 *
366 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
367 * as long as the inode doesn't go away and racy results are not a problem.
368 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800369unsigned long shmem_partial_swap_usage(struct address_space *mapping,
370 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800371{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800372 struct radix_tree_iter iter;
373 void **slot;
374 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800375 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800376
377 rcu_read_lock();
378
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800379 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
380 if (iter.index >= end)
381 break;
382
383 page = radix_tree_deref_slot(slot);
384
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700385 if (radix_tree_deref_retry(page)) {
386 slot = radix_tree_iter_retry(&iter);
387 continue;
388 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800389
390 if (radix_tree_exceptional_entry(page))
391 swapped++;
392
393 if (need_resched()) {
394 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700395 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800396 }
397 }
398
399 rcu_read_unlock();
400
401 return swapped << PAGE_SHIFT;
402}
403
404/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800405 * Determine (in bytes) how many of the shmem object's pages mapped by the
406 * given vma is swapped out.
407 *
408 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
409 * as long as the inode doesn't go away and racy results are not a problem.
410 */
411unsigned long shmem_swap_usage(struct vm_area_struct *vma)
412{
413 struct inode *inode = file_inode(vma->vm_file);
414 struct shmem_inode_info *info = SHMEM_I(inode);
415 struct address_space *mapping = inode->i_mapping;
416 unsigned long swapped;
417
418 /* Be careful as we don't hold info->lock */
419 swapped = READ_ONCE(info->swapped);
420
421 /*
422 * The easier cases are when the shmem object has nothing in swap, or
423 * the vma maps it whole. Then we can simply use the stats that we
424 * already track.
425 */
426 if (!swapped)
427 return 0;
428
429 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
430 return swapped << PAGE_SHIFT;
431
432 /* Here comes the more involved part */
433 return shmem_partial_swap_usage(mapping,
434 linear_page_index(vma, vma->vm_start),
435 linear_page_index(vma, vma->vm_end));
436}
437
438/*
Hugh Dickins24513262012-01-20 14:34:21 -0800439 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
440 */
441void shmem_unlock_mapping(struct address_space *mapping)
442{
443 struct pagevec pvec;
444 pgoff_t indices[PAGEVEC_SIZE];
445 pgoff_t index = 0;
446
447 pagevec_init(&pvec, 0);
448 /*
449 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
450 */
451 while (!mapping_unevictable(mapping)) {
452 /*
453 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
454 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
455 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700456 pvec.nr = find_get_entries(mapping, index,
457 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800458 if (!pvec.nr)
459 break;
460 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700461 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800462 check_move_unevictable_pages(pvec.pages, pvec.nr);
463 pagevec_release(&pvec);
464 cond_resched();
465 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700466}
467
468/*
469 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700470 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700471 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700472static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
473 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700475 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300477 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
478 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
479 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
480 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700481 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700482 pgoff_t indices[PAGEVEC_SIZE];
483 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700484 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700485 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700487 if (lend == -1)
488 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700489
490 pagevec_init(&pvec, 0);
491 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700492 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700493 pvec.nr = find_get_entries(mapping, index,
494 min(end - index, (pgoff_t)PAGEVEC_SIZE),
495 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700496 if (!pvec.nr)
497 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700498 for (i = 0; i < pagevec_count(&pvec); i++) {
499 struct page *page = pvec.pages[i];
500
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700501 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700502 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700503 break;
504
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700505 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700506 if (unfalloc)
507 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700508 nr_swaps_freed += !shmem_free_swap(mapping,
509 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700510 continue;
511 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700512
513 if (!trylock_page(page))
514 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700515 if (!unfalloc || !PageUptodate(page)) {
516 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800517 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700518 truncate_inode_page(mapping, page);
519 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700520 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700521 unlock_page(page);
522 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700523 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800524 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700525 cond_resched();
526 index++;
527 }
528
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700529 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700530 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700531 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700532 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300533 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700534 if (start > end) {
535 top = partial_end;
536 partial_end = 0;
537 }
538 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700539 set_page_dirty(page);
540 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300541 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700542 }
543 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700544 if (partial_end) {
545 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700546 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700547 if (page) {
548 zero_user_segment(page, 0, partial_end);
549 set_page_dirty(page);
550 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300551 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700552 }
553 }
554 if (start >= end)
555 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700556
557 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700558 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700559 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700560
561 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700562 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700563 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700564 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700565 /* If all gone or hole-punch or unfalloc, we're done */
566 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700567 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700568 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700569 index = start;
570 continue;
571 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700572 for (i = 0; i < pagevec_count(&pvec); i++) {
573 struct page *page = pvec.pages[i];
574
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700575 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700576 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700577 break;
578
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700579 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700580 if (unfalloc)
581 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700582 if (shmem_free_swap(mapping, index, page)) {
583 /* Swap was replaced by page: retry */
584 index--;
585 break;
586 }
587 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700588 continue;
589 }
590
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700591 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700592 if (!unfalloc || !PageUptodate(page)) {
593 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800594 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700595 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700596 } else {
597 /* Page was replaced by swap: retry */
598 unlock_page(page);
599 index--;
600 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700601 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700602 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700603 unlock_page(page);
604 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700605 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800606 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700607 index++;
608 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700611 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 shmem_recalc_inode(inode);
613 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700614}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700616void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
617{
618 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700619 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700621EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Yu Zhao44a30222015-09-08 15:03:33 -0700623static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
624 struct kstat *stat)
625{
626 struct inode *inode = dentry->d_inode;
627 struct shmem_inode_info *info = SHMEM_I(inode);
628
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800629 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
630 spin_lock(&info->lock);
631 shmem_recalc_inode(inode);
632 spin_unlock(&info->lock);
633 }
Yu Zhao44a30222015-09-08 15:03:33 -0700634 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700635 return 0;
636}
637
Hugh Dickins94c1e622011-06-27 16:18:03 -0700638static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
David Howells75c3cfa2015-03-17 22:26:12 +0000640 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700641 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 int error;
643
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200644 error = inode_change_ok(inode, attr);
645 if (error)
646 return error;
647
Hugh Dickins94c1e622011-06-27 16:18:03 -0700648 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
649 loff_t oldsize = inode->i_size;
650 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000651
David Herrmann40e041a2014-08-08 14:25:27 -0700652 /* protected by i_mutex */
653 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
654 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
655 return -EPERM;
656
Hugh Dickins94c1e622011-06-27 16:18:03 -0700657 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700658 error = shmem_reacct_size(SHMEM_I(inode)->flags,
659 oldsize, newsize);
660 if (error)
661 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700662 i_size_write(inode, newsize);
663 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
664 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700665 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700666 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800667 if (oldsize > holebegin)
668 unmap_mapping_range(inode->i_mapping,
669 holebegin, 0, 1);
670 if (info->alloced)
671 shmem_truncate_range(inode,
672 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700673 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800674 if (oldsize > holebegin)
675 unmap_mapping_range(inode->i_mapping,
676 holebegin, 0, 1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200680 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200681 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -0800682 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 return error;
684}
685
Al Viro1f895f72010-06-05 19:10:41 -0400686static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 struct shmem_inode_info *info = SHMEM_I(inode);
689
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000690 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 shmem_unacct_size(info->flags, inode->i_size);
692 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000693 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800695 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800697 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
Al Viro3ed47db2016-01-22 18:08:52 -0500699 }
Eric Parisb09e0fa2011-05-24 17:12:39 -0700700
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400701 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800702 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800703 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200704 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705}
706
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700707/*
708 * If swap found in inode, free it and move page from swapcache to filecache.
709 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700710static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700711 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700713 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700714 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700715 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700716 gfp_t gfp;
717 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700719 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700720 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700721 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -0700722 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800723
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800724 /*
725 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400726 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800727 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700728 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800729 */
730 if (shmem_swaplist.next != &info->swaplist)
731 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800732
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700733 gfp = mapping_gfp_mask(mapping);
734 if (shmem_should_replace_page(*pagep, gfp)) {
735 mutex_unlock(&shmem_swaplist_mutex);
736 error = shmem_replace_page(pagep, gfp, info, index);
737 mutex_lock(&shmem_swaplist_mutex);
738 /*
739 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700740 * allocation, but the inode might have been freed while we
741 * dropped it: although a racing shmem_evict_inode() cannot
742 * complete without emptying the radix_tree, our page lock
743 * on this swapcache page is not enough to prevent that -
744 * free_swap_and_cache() of our swap entry will only
745 * trylock_page(), removing swap from radix_tree whatever.
746 *
747 * We must not proceed to shmem_add_to_page_cache() if the
748 * inode has been freed, but of course we cannot rely on
749 * inode or mapping or info to check that. However, we can
750 * safely check if our swap entry is still in use (and here
751 * it can't have got reused for another page): if it's still
752 * in use, then the inode cannot have been freed yet, and we
753 * can safely proceed (if it's no longer in use, that tells
754 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700755 */
756 if (!page_swapcount(*pagep))
757 error = -ENOENT;
758 }
759
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800760 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700761 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
762 * but also to hold up shmem_evict_inode(): so inode cannot be freed
763 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800764 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700765 if (!error)
766 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700767 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700768 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700769 /*
770 * Truncation and eviction use free_swap_and_cache(), which
771 * only does trylock page: if we raced, best clean up here.
772 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700773 delete_from_swap_cache(*pagep);
774 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700775 if (!error) {
776 spin_lock(&info->lock);
777 info->swapped--;
778 spin_unlock(&info->lock);
779 swap_free(swap);
780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800782 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783}
784
785/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700786 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700788int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700790 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -0700792 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700793 int error = 0;
794
795 /*
796 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700797 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700798 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700799 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700800 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700801
802 /*
803 * Charge page using GFP_KERNEL while we can wait, before taking
804 * the shmem_swaplist_mutex which might hold up shmem_writepage().
805 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700806 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800807 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
808 false);
Hugh Dickins778dd892011-05-11 15:13:37 -0700809 if (error)
810 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700811 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -0700812 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800814 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700815 list_for_each_safe(this, next, &shmem_swaplist) {
816 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700817 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -0700818 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700819 else
820 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800821 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -0700822 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -0700823 break;
Johannes Weiner00501b52014-08-08 14:19:20 -0700824 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800826 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700827
Johannes Weiner00501b52014-08-08 14:19:20 -0700828 if (error) {
829 if (error != -ENOMEM)
830 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800831 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700832 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800833 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -0700834out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800835 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300836 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700837 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838}
839
840/*
841 * Move the page from the page cache to the swap cache.
842 */
843static int shmem_writepage(struct page *page, struct writeback_control *wbc)
844{
845 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700848 swp_entry_t swap;
849 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
851 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 mapping = page->mapping;
853 index = page->index;
854 inode = mapping->host;
855 info = SHMEM_I(inode);
856 if (info->flags & VM_LOCKED)
857 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800858 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 goto redirty;
860
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800861 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100862 * Our capabilities prevent regular writeback or sync from ever calling
863 * shmem_writepage; but a stacking filesystem might use ->writepage of
864 * its underlying filesystem, in which case tmpfs should write out to
865 * swap only in response to memory pressure, and not for the writeback
866 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800867 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700868 if (!wbc->for_reclaim) {
869 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
870 goto redirty;
871 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700872
873 /*
874 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
875 * value into swapfile.c, the only way we can correctly account for a
876 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700877 *
878 * That's okay for a page already fallocated earlier, but if we have
879 * not yet completed the fallocation, then (a) we want to keep track
880 * of this page in case we have to undo it, and (b) it may not be a
881 * good idea to continue anyway, once we're pushing into swap. So
882 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700883 */
884 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700885 if (inode->i_private) {
886 struct shmem_falloc *shmem_falloc;
887 spin_lock(&inode->i_lock);
888 shmem_falloc = inode->i_private;
889 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -0700890 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -0700891 index >= shmem_falloc->start &&
892 index < shmem_falloc->next)
893 shmem_falloc->nr_unswapped++;
894 else
895 shmem_falloc = NULL;
896 spin_unlock(&inode->i_lock);
897 if (shmem_falloc)
898 goto redirty;
899 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700900 clear_highpage(page);
901 flush_dcache_page(page);
902 SetPageUptodate(page);
903 }
904
Hugh Dickins48f170f2011-07-25 17:12:37 -0700905 swap = get_swap_page();
906 if (!swap.val)
907 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800908
Vladimir Davydov37e84352016-01-20 15:02:56 -0800909 if (mem_cgroup_try_charge_swap(page, swap))
910 goto free_swap;
911
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700912 /*
913 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700914 * if it's not already there. Do it now before the page is
915 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700916 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700917 * we've incremented swapped, because shmem_unuse_inode() will
918 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700919 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700920 mutex_lock(&shmem_swaplist_mutex);
921 if (list_empty(&info->swaplist))
922 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700923
Hugh Dickins48f170f2011-07-25 17:12:37 -0700924 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins267a4c72015-12-11 13:40:55 -0800925 spin_lock(&info->lock);
926 shmem_recalc_inode(inode);
927 info->swapped++;
928 spin_unlock(&info->lock);
929
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800930 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700931 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
932
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700933 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800934 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700935 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return 0;
937 }
938
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700939 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -0800940free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700941 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942redirty:
943 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800944 if (wbc->for_reclaim)
945 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
946 unlock_page(page);
947 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948}
949
Hugh Dickins75edd342016-05-19 17:12:44 -0700950#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700951static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800952{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700953 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800954
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700955 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700956 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800957
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800958 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800959
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700960 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800961}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700962
963static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
964{
965 struct mempolicy *mpol = NULL;
966 if (sbinfo->mpol) {
967 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
968 mpol = sbinfo->mpol;
969 mpol_get(mpol);
970 spin_unlock(&sbinfo->stat_lock);
971 }
972 return mpol;
973}
Hugh Dickins75edd342016-05-19 17:12:44 -0700974#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
975static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
976{
977}
978static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
979{
980 return NULL;
981}
982#endif /* CONFIG_NUMA && CONFIG_TMPFS */
983#ifndef CONFIG_NUMA
984#define vm_policy vm_private_data
985#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800986
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700987static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
988 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800991 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800993 /* Create a pseudo vma that just contains the policy */
994 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700995 /* Bias interleave by inode number to distribute better across nodes */
996 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800997 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700998 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700999
Mel Gorman18a2f372012-12-05 14:01:41 -08001000 page = swapin_readahead(swap, gfp, &pvma, 0);
1001
1002 /* Drop reference taken by mpol_shared_policy_lookup() */
1003 mpol_cond_put(pvma.vm_policy);
1004
1005 return page;
1006}
1007
1008static struct page *shmem_alloc_page(gfp_t gfp,
1009 struct shmem_inode_info *info, pgoff_t index)
1010{
1011 struct vm_area_struct pvma;
1012 struct page *page;
1013
1014 /* Create a pseudo vma that just contains the policy */
1015 pvma.vm_start = 0;
1016 /* Bias interleave by inode number to distribute better across nodes */
1017 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
1018 pvma.vm_ops = NULL;
1019 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1020
Hugh Dickins75edd342016-05-19 17:12:44 -07001021 page = alloc_pages_vma(gfp, 0, &pvma, 0, numa_node_id(), false);
1022 if (page) {
1023 __SetPageLocked(page);
1024 __SetPageSwapBacked(page);
1025 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001026
1027 /* Drop reference taken by mpol_shared_policy_lookup() */
1028 mpol_cond_put(pvma.vm_policy);
1029
1030 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001034 * When a page is moved from swapcache to shmem filecache (either by the
1035 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1036 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1037 * ignorance of the mapping it belongs to. If that mapping has special
1038 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1039 * we may need to copy to a suitable page before moving to filecache.
1040 *
1041 * In a future release, this may well be extended to respect cpuset and
1042 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1043 * but for now it is a simple matter of zone.
1044 */
1045static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1046{
1047 return page_zonenum(page) > gfp_zone(gfp);
1048}
1049
1050static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1051 struct shmem_inode_info *info, pgoff_t index)
1052{
1053 struct page *oldpage, *newpage;
1054 struct address_space *swap_mapping;
1055 pgoff_t swap_index;
1056 int error;
1057
1058 oldpage = *pagep;
1059 swap_index = page_private(oldpage);
1060 swap_mapping = page_mapping(oldpage);
1061
1062 /*
1063 * We have arrived here because our zones are constrained, so don't
1064 * limit chance of success by further cpuset and node constraints.
1065 */
1066 gfp &= ~GFP_CONSTRAINT_MASK;
1067 newpage = shmem_alloc_page(gfp, info, index);
1068 if (!newpage)
1069 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001070
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001071 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001072 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001073 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001074
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001075 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001076 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001077 SetPageSwapCache(newpage);
1078
1079 /*
1080 * Our caller will very soon move newpage out of swapcache, but it's
1081 * a nice clean interface for us to replace oldpage by newpage there.
1082 */
1083 spin_lock_irq(&swap_mapping->tree_lock);
1084 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1085 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001086 if (!error) {
1087 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1088 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1089 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001090 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001091
Hugh Dickins0142ef62012-06-07 14:21:09 -07001092 if (unlikely(error)) {
1093 /*
1094 * Is this possible? I think not, now that our callers check
1095 * both PageSwapCache and page_private after getting page lock;
1096 * but be defensive. Reverse old to newpage for clear and free.
1097 */
1098 oldpage = newpage;
1099 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001100 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001101 lru_cache_add_anon(newpage);
1102 *pagep = newpage;
1103 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001104
1105 ClearPageSwapCache(oldpage);
1106 set_page_private(oldpage, 0);
1107
1108 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001109 put_page(oldpage);
1110 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001111 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001112}
1113
1114/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001115 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 *
1117 * If we allocate a new one we do not mark it dirty. That's up to the
1118 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001119 * entry since a page cannot live in both the swap and page cache.
1120 *
1121 * fault_mm and fault_type are only supplied by shmem_fault:
1122 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001124static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001125 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1126 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127{
1128 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001129 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001131 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001132 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001133 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 swp_entry_t swap;
1135 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001136 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001137 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001139 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001142 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001143 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001144 if (radix_tree_exceptional_entry(page)) {
1145 swap = radix_to_swp_entry(page);
1146 page = NULL;
1147 }
1148
Hugh Dickins75edd342016-05-19 17:12:44 -07001149 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001150 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001151 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001152 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001153 }
1154
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001155 if (page && sgp == SGP_WRITE)
1156 mark_page_accessed(page);
1157
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001158 /* fallocated page? */
1159 if (page && !PageUptodate(page)) {
1160 if (sgp != SGP_READ)
1161 goto clear;
1162 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001163 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001164 page = NULL;
1165 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001166 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001167 *pagep = page;
1168 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001172 * Fast cache lookup did not find it:
1173 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001175 info = SHMEM_I(inode);
1176 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001177 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (swap.val) {
1180 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001181 page = lookup_swap_cache(swap);
1182 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001183 /* Or update major stats only when swapin succeeds?? */
1184 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001185 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001186 count_vm_event(PGMAJFAULT);
1187 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1188 }
1189 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001190 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001191 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001192 error = -ENOMEM;
1193 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
1196
1197 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001198 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001199 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001200 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001201 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001202 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001203 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001204 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001206 goto failed;
1207 }
1208 wait_on_page_writeback(page);
1209
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001210 if (shmem_should_replace_page(page, gfp)) {
1211 error = shmem_replace_page(&page, gfp, info, index);
1212 if (error)
1213 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001216 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001217 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001218 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001219 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001220 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001221 /*
1222 * We already confirmed swap under page lock, and make
1223 * no memory allocation here, so usually no possibility
1224 * of error; but free_swap_and_cache() only trylocks a
1225 * page, so it is just possible that the entry has been
1226 * truncated or holepunched since swap was confirmed.
1227 * shmem_undo_range() will have done some of the
1228 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001229 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001230 * Reset swap.val? No, leave it so "failed" goes back to
1231 * "repeat": reading a hole and writing should succeed.
1232 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001233 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001234 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001235 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001236 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001237 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001238 if (error)
1239 goto failed;
1240
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001241 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001242
Hugh Dickins54af60422011-08-03 16:21:24 -07001243 spin_lock(&info->lock);
1244 info->swapped--;
1245 shmem_recalc_inode(inode);
1246 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001247
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001248 if (sgp == SGP_WRITE)
1249 mark_page_accessed(page);
1250
Hugh Dickins27ab7002011-07-25 17:12:36 -07001251 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001252 set_page_dirty(page);
1253 swap_free(swap);
1254
Hugh Dickins54af60422011-08-03 16:21:24 -07001255 } else {
1256 if (shmem_acct_block(info->flags)) {
1257 error = -ENOSPC;
1258 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001260 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001261 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001262 sbinfo->max_blocks) >= 0) {
1263 error = -ENOSPC;
1264 goto unacct;
1265 }
Tim Chen7e496292010-08-09 17:19:05 -07001266 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001267 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Hugh Dickins54af60422011-08-03 16:21:24 -07001269 page = shmem_alloc_page(gfp, info, index);
1270 if (!page) {
1271 error = -ENOMEM;
1272 goto decused;
1273 }
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001274 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001275 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001276
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001277 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001278 false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001279 if (error)
1280 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001281 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001282 if (!error) {
1283 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001284 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001285 radix_tree_preload_end();
1286 }
1287 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001288 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001289 goto decused;
1290 }
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001291 mem_cgroup_commit_charge(page, memcg, false, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001292 lru_cache_add_anon(page);
1293
1294 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001296 inode->i_blocks += BLOCKS_PER_PAGE;
1297 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001299 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001300
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001301 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001302 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1303 */
1304 if (sgp == SGP_FALLOC)
1305 sgp = SGP_WRITE;
1306clear:
1307 /*
1308 * Let SGP_WRITE caller clear ends if write does not fill page;
1309 * but SGP_FALLOC on a page fallocated earlier must initialize
1310 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001311 */
1312 if (sgp != SGP_WRITE) {
1313 clear_highpage(page);
1314 flush_dcache_page(page);
1315 SetPageUptodate(page);
1316 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001317 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001318
Hugh Dickins54af60422011-08-03 16:21:24 -07001319 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001320 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001321 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001322 if (alloced) {
1323 ClearPageDirty(page);
1324 delete_from_page_cache(page);
1325 spin_lock(&info->lock);
1326 shmem_recalc_inode(inode);
1327 spin_unlock(&info->lock);
1328 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001329 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001330 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001331 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001332 *pagep = page;
1333 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001334
Nick Piggind0217ac2007-07-19 01:47:03 -07001335 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001336 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001338decused:
1339 if (sbinfo->max_blocks)
1340 percpu_counter_add(&sbinfo->used_blocks, -1);
1341unacct:
1342 shmem_unacct_blocks(info->flags, 1);
1343failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001344 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001345 error = -EEXIST;
1346unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001347 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001348 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001349 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001350 }
1351 if (error == -ENOSPC && !once++) {
1352 info = SHMEM_I(inode);
1353 spin_lock(&info->lock);
1354 shmem_recalc_inode(inode);
1355 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001356 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001357 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001358 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001359 goto repeat;
1360 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361}
1362
1363static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1364{
Al Viro496ad9a2013-01-23 17:07:38 -05001365 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001366 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001368 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001370 /*
1371 * Trinity finds that probing a hole which tmpfs is punching can
1372 * prevent the hole-punch from ever completing: which in turn
1373 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001374 * faulting pages into the hole while it's being punched. Although
1375 * shmem_undo_range() does remove the additions, it may be unable to
1376 * keep up, as each new page needs its own unmap_mapping_range() call,
1377 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1378 *
1379 * It does not matter if we sometimes reach this check just before the
1380 * hole-punch begins, so that one fault then races with the punch:
1381 * we just need to make racing faults a rare case.
1382 *
1383 * The implementation below would be much simpler if we just used a
1384 * standard mutex or completion: but we cannot take i_mutex in fault,
1385 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001386 */
1387 if (unlikely(inode->i_private)) {
1388 struct shmem_falloc *shmem_falloc;
1389
1390 spin_lock(&inode->i_lock);
1391 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001392 if (shmem_falloc &&
1393 shmem_falloc->waitq &&
1394 vmf->pgoff >= shmem_falloc->start &&
1395 vmf->pgoff < shmem_falloc->next) {
1396 wait_queue_head_t *shmem_falloc_waitq;
1397 DEFINE_WAIT(shmem_fault_wait);
1398
1399 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001400 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1401 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001402 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001403 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001404 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001405 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001406
1407 shmem_falloc_waitq = shmem_falloc->waitq;
1408 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1409 TASK_UNINTERRUPTIBLE);
1410 spin_unlock(&inode->i_lock);
1411 schedule();
1412
1413 /*
1414 * shmem_falloc_waitq points into the shmem_fallocate()
1415 * stack of the hole-punching task: shmem_falloc_waitq
1416 * is usually invalid by the time we reach here, but
1417 * finish_wait() does not dereference it in that case;
1418 * though i_lock needed lest racing with wake_up_all().
1419 */
1420 spin_lock(&inode->i_lock);
1421 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1422 spin_unlock(&inode->i_lock);
1423 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001424 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001425 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001426 }
1427
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001428 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
1429 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (error)
1431 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001432 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433}
1434
1435#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001436static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Al Viro496ad9a2013-01-23 17:07:38 -05001438 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001439 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440}
1441
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001442static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1443 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
Al Viro496ad9a2013-01-23 17:07:38 -05001445 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001446 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001448 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1449 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450}
1451#endif
1452
1453int shmem_lock(struct file *file, int lock, struct user_struct *user)
1454{
Al Viro496ad9a2013-01-23 17:07:38 -05001455 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 struct shmem_inode_info *info = SHMEM_I(inode);
1457 int retval = -ENOMEM;
1458
1459 spin_lock(&info->lock);
1460 if (lock && !(info->flags & VM_LOCKED)) {
1461 if (!user_shm_lock(inode->i_size, user))
1462 goto out_nomem;
1463 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001464 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 }
1466 if (!lock && (info->flags & VM_LOCKED) && user) {
1467 user_shm_unlock(inode->i_size, user);
1468 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001469 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 }
1471 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473out_nomem:
1474 spin_unlock(&info->lock);
1475 return retval;
1476}
1477
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001478static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479{
1480 file_accessed(file);
1481 vma->vm_ops = &shmem_vm_ops;
1482 return 0;
1483}
1484
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001485static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001486 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
1488 struct inode *inode;
1489 struct shmem_inode_info *info;
1490 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1491
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001492 if (shmem_reserve_inode(sb))
1493 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 inode = new_inode(sb);
1496 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001497 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001498 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001501 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 info = SHMEM_I(inode);
1503 memset(info, 0, (char *)inode - (char *)info);
1504 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07001505 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001506 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001508 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001509 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
1511 switch (mode & S_IFMT) {
1512 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001513 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 init_special_inode(inode, mode, dev);
1515 break;
1516 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001517 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 inode->i_op = &shmem_inode_operations;
1519 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001520 mpol_shared_policy_init(&info->policy,
1521 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 break;
1523 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001524 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 /* Some things misbehave if size == 0 on a directory */
1526 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1527 inode->i_op = &shmem_dir_inode_operations;
1528 inode->i_fop = &simple_dir_operations;
1529 break;
1530 case S_IFLNK:
1531 /*
1532 * Must not load anything in the rbtree,
1533 * mpol_free_shared_policy will not be called.
1534 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001535 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 break;
1537 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001538 } else
1539 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 return inode;
1541}
1542
Johannes Weiner0cd61442014-04-03 14:47:46 -07001543bool shmem_mapping(struct address_space *mapping)
1544{
Sasha Levinf0774d82015-02-23 05:38:00 -05001545 if (!mapping->host)
1546 return false;
1547
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001548 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001549}
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001552static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001553static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001555#ifdef CONFIG_TMPFS_XATTR
1556static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1557#else
1558#define shmem_initxattrs NULL
1559#endif
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001562shmem_write_begin(struct file *file, struct address_space *mapping,
1563 loff_t pos, unsigned len, unsigned flags,
1564 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565{
Nick Piggin800d15a2007-10-16 01:25:03 -07001566 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07001567 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001568 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07001569
1570 /* i_mutex is held by caller */
1571 if (unlikely(info->seals)) {
1572 if (info->seals & F_SEAL_WRITE)
1573 return -EPERM;
1574 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1575 return -EPERM;
1576 }
1577
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001578 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07001579}
1580
1581static int
1582shmem_write_end(struct file *file, struct address_space *mapping,
1583 loff_t pos, unsigned len, unsigned copied,
1584 struct page *page, void *fsdata)
1585{
1586 struct inode *inode = mapping->host;
1587
Hugh Dickinsd3602442008-02-04 22:28:44 -08001588 if (pos + copied > inode->i_size)
1589 i_size_write(inode, pos + copied);
1590
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001591 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001592 if (copied < PAGE_SIZE) {
1593 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001594 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001595 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001596 }
1597 SetPageUptodate(page);
1598 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001599 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001600 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001601 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001602
Nick Piggin800d15a2007-10-16 01:25:03 -07001603 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604}
1605
Al Viro2ba5bbe2014-04-02 20:00:02 -04001606static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607{
Al Viro6e58e792014-02-03 17:07:03 -05001608 struct file *file = iocb->ki_filp;
1609 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001611 pgoff_t index;
1612 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001613 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02001614 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05001615 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05001616 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001617
1618 /*
1619 * Might this read be for a stacking filesystem? Then when reading
1620 * holes of a sparse file, we actually need to allocate those pages,
1621 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1622 */
Al Viro777eda22014-12-17 04:46:46 -05001623 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07001624 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001626 index = *ppos >> PAGE_SHIFT;
1627 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 for (;;) {
1630 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001631 pgoff_t end_index;
1632 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 loff_t i_size = i_size_read(inode);
1634
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001635 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (index > end_index)
1637 break;
1638 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001639 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 if (nr <= offset)
1641 break;
1642 }
1643
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001644 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05001645 if (error) {
1646 if (error == -EINVAL)
1647 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 break;
1649 }
Hugh Dickins75edd342016-05-19 17:12:44 -07001650 if (page) {
1651 if (sgp == SGP_CACHE)
1652 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08001653 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07001654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
1656 /*
1657 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001658 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001660 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001662 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001664 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 if (nr <= offset) {
1666 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001667 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 break;
1669 }
1670 }
1671 nr -= offset;
1672
1673 if (page) {
1674 /*
1675 * If users can be writing to this page using arbitrary
1676 * virtual addresses, take care about potential aliasing
1677 * before reading the page on the kernel side.
1678 */
1679 if (mapping_writably_mapped(mapping))
1680 flush_dcache_page(page);
1681 /*
1682 * Mark the page accessed if we read the beginning.
1683 */
1684 if (!offset)
1685 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001686 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001688 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 /*
1692 * Ok, we have the page, and it's up-to-date, so
1693 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04001695 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05001696 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001698 index += offset >> PAGE_SHIFT;
1699 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001701 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04001702 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 break;
Al Viro6e58e792014-02-03 17:07:03 -05001704 if (ret < nr) {
1705 error = -EFAULT;
1706 break;
1707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 cond_resched();
1709 }
1710
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001711 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05001712 file_accessed(file);
1713 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714}
1715
Hugh Dickins708e3502011-07-25 17:12:32 -07001716static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1717 struct pipe_inode_info *pipe, size_t len,
1718 unsigned int flags)
1719{
1720 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001721 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001722 unsigned int loff, nr_pages, req_pages;
1723 struct page *pages[PIPE_DEF_BUFFERS];
1724 struct partial_page partial[PIPE_DEF_BUFFERS];
1725 struct page *page;
1726 pgoff_t index, end_index;
1727 loff_t isize, left;
1728 int error, page_nr;
1729 struct splice_pipe_desc spd = {
1730 .pages = pages,
1731 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001732 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001733 .flags = flags,
1734 .ops = &page_cache_pipe_buf_ops,
1735 .spd_release = spd_release_page,
1736 };
1737
Hugh Dickins71f0e072011-07-25 17:12:33 -07001738 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001739 if (unlikely(*ppos >= isize))
1740 return 0;
1741
1742 left = isize - *ppos;
1743 if (unlikely(left < len))
1744 len = left;
1745
1746 if (splice_grow_spd(pipe, &spd))
1747 return -ENOMEM;
1748
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001749 index = *ppos >> PAGE_SHIFT;
1750 loff = *ppos & ~PAGE_MASK;
1751 req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04001752 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07001753
Hugh Dickins708e3502011-07-25 17:12:32 -07001754 spd.nr_pages = find_get_pages_contig(mapping, index,
1755 nr_pages, spd.pages);
1756 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001757 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001758
Hugh Dickins708e3502011-07-25 17:12:32 -07001759 while (spd.nr_pages < nr_pages) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001760 error = shmem_getpage(inode, index, &page, SGP_CACHE);
Hugh Dickins71f0e072011-07-25 17:12:33 -07001761 if (error)
1762 break;
1763 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001764 spd.pages[spd.nr_pages++] = page;
1765 index++;
1766 }
1767
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001768 index = *ppos >> PAGE_SHIFT;
Hugh Dickins708e3502011-07-25 17:12:32 -07001769 nr_pages = spd.nr_pages;
1770 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001771
Hugh Dickins708e3502011-07-25 17:12:32 -07001772 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1773 unsigned int this_len;
1774
1775 if (!len)
1776 break;
1777
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001778 this_len = min_t(unsigned long, len, PAGE_SIZE - loff);
Hugh Dickins708e3502011-07-25 17:12:32 -07001779 page = spd.pages[page_nr];
1780
Hugh Dickins71f0e072011-07-25 17:12:33 -07001781 if (!PageUptodate(page) || page->mapping != mapping) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001782 error = shmem_getpage(inode, index, &page, SGP_CACHE);
Hugh Dickins71f0e072011-07-25 17:12:33 -07001783 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001784 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001785 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001786 put_page(spd.pages[page_nr]);
Hugh Dickins71f0e072011-07-25 17:12:33 -07001787 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001788 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001789
1790 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001791 end_index = (isize - 1) >> PAGE_SHIFT;
Hugh Dickins708e3502011-07-25 17:12:32 -07001792 if (unlikely(!isize || index > end_index))
1793 break;
1794
Hugh Dickins708e3502011-07-25 17:12:32 -07001795 if (end_index == index) {
1796 unsigned int plen;
1797
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001798 plen = ((isize - 1) & ~PAGE_MASK) + 1;
Hugh Dickins708e3502011-07-25 17:12:32 -07001799 if (plen <= loff)
1800 break;
1801
Hugh Dickins708e3502011-07-25 17:12:32 -07001802 this_len = min(this_len, plen - loff);
1803 len = this_len;
1804 }
1805
1806 spd.partial[page_nr].offset = loff;
1807 spd.partial[page_nr].len = this_len;
1808 len -= this_len;
1809 loff = 0;
1810 spd.nr_pages++;
1811 index++;
1812 }
1813
Hugh Dickins708e3502011-07-25 17:12:32 -07001814 while (page_nr < nr_pages)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001815 put_page(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001816
1817 if (spd.nr_pages)
1818 error = splice_to_pipe(pipe, &spd);
1819
Eric Dumazet047fe362012-06-12 15:24:40 +02001820 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001821
1822 if (error > 0) {
1823 *ppos += error;
1824 file_accessed(in);
1825 }
1826 return error;
1827}
1828
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001829/*
1830 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1831 */
1832static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001833 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001834{
1835 struct page *page;
1836 struct pagevec pvec;
1837 pgoff_t indices[PAGEVEC_SIZE];
1838 bool done = false;
1839 int i;
1840
1841 pagevec_init(&pvec, 0);
1842 pvec.nr = 1; /* start small: we may be there already */
1843 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001844 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001845 pvec.nr, pvec.pages, indices);
1846 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001847 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001848 index = end;
1849 break;
1850 }
1851 for (i = 0; i < pvec.nr; i++, index++) {
1852 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001853 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001854 done = true;
1855 break;
1856 }
1857 index = indices[i];
1858 }
1859 page = pvec.pages[i];
1860 if (page && !radix_tree_exceptional_entry(page)) {
1861 if (!PageUptodate(page))
1862 page = NULL;
1863 }
1864 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001865 (page && whence == SEEK_DATA) ||
1866 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001867 done = true;
1868 break;
1869 }
1870 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001871 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001872 pagevec_release(&pvec);
1873 pvec.nr = PAGEVEC_SIZE;
1874 cond_resched();
1875 }
1876 return index;
1877}
1878
Andrew Morton965c8e52012-12-17 15:59:39 -08001879static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001880{
1881 struct address_space *mapping = file->f_mapping;
1882 struct inode *inode = mapping->host;
1883 pgoff_t start, end;
1884 loff_t new_offset;
1885
Andrew Morton965c8e52012-12-17 15:59:39 -08001886 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1887 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001888 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05001889 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001890 /* We're holding i_mutex so we can access i_size directly */
1891
1892 if (offset < 0)
1893 offset = -EINVAL;
1894 else if (offset >= inode->i_size)
1895 offset = -ENXIO;
1896 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001897 start = offset >> PAGE_SHIFT;
1898 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001899 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001900 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001901 if (new_offset > offset) {
1902 if (new_offset < inode->i_size)
1903 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001904 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001905 offset = -ENXIO;
1906 else
1907 offset = inode->i_size;
1908 }
1909 }
1910
Hugh Dickins387aae62013-08-04 11:30:25 -07001911 if (offset >= 0)
1912 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05001913 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001914 return offset;
1915}
1916
David Herrmann05f65b52014-08-08 14:25:36 -07001917/*
1918 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1919 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1920 */
1921#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1922#define LAST_SCAN 4 /* about 150ms max */
1923
1924static void shmem_tag_pins(struct address_space *mapping)
1925{
1926 struct radix_tree_iter iter;
1927 void **slot;
1928 pgoff_t start;
1929 struct page *page;
1930
1931 lru_add_drain();
1932 start = 0;
1933 rcu_read_lock();
1934
David Herrmann05f65b52014-08-08 14:25:36 -07001935 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1936 page = radix_tree_deref_slot(slot);
1937 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001938 if (radix_tree_deref_retry(page)) {
1939 slot = radix_tree_iter_retry(&iter);
1940 continue;
1941 }
David Herrmann05f65b52014-08-08 14:25:36 -07001942 } else if (page_count(page) - page_mapcount(page) > 1) {
1943 spin_lock_irq(&mapping->tree_lock);
1944 radix_tree_tag_set(&mapping->page_tree, iter.index,
1945 SHMEM_TAG_PINNED);
1946 spin_unlock_irq(&mapping->tree_lock);
1947 }
1948
1949 if (need_resched()) {
1950 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07001951 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07001952 }
1953 }
1954 rcu_read_unlock();
1955}
1956
1957/*
1958 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1959 * via get_user_pages(), drivers might have some pending I/O without any active
1960 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1961 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1962 * them to be dropped.
1963 * The caller must guarantee that no new user will acquire writable references
1964 * to those pages to avoid races.
1965 */
David Herrmann40e041a2014-08-08 14:25:27 -07001966static int shmem_wait_for_pins(struct address_space *mapping)
1967{
David Herrmann05f65b52014-08-08 14:25:36 -07001968 struct radix_tree_iter iter;
1969 void **slot;
1970 pgoff_t start;
1971 struct page *page;
1972 int error, scan;
1973
1974 shmem_tag_pins(mapping);
1975
1976 error = 0;
1977 for (scan = 0; scan <= LAST_SCAN; scan++) {
1978 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1979 break;
1980
1981 if (!scan)
1982 lru_add_drain_all();
1983 else if (schedule_timeout_killable((HZ << scan) / 200))
1984 scan = LAST_SCAN;
1985
1986 start = 0;
1987 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07001988 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
1989 start, SHMEM_TAG_PINNED) {
1990
1991 page = radix_tree_deref_slot(slot);
1992 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001993 if (radix_tree_deref_retry(page)) {
1994 slot = radix_tree_iter_retry(&iter);
1995 continue;
1996 }
David Herrmann05f65b52014-08-08 14:25:36 -07001997
1998 page = NULL;
1999 }
2000
2001 if (page &&
2002 page_count(page) - page_mapcount(page) != 1) {
2003 if (scan < LAST_SCAN)
2004 goto continue_resched;
2005
2006 /*
2007 * On the last scan, we clean up all those tags
2008 * we inserted; but make a note that we still
2009 * found pages pinned.
2010 */
2011 error = -EBUSY;
2012 }
2013
2014 spin_lock_irq(&mapping->tree_lock);
2015 radix_tree_tag_clear(&mapping->page_tree,
2016 iter.index, SHMEM_TAG_PINNED);
2017 spin_unlock_irq(&mapping->tree_lock);
2018continue_resched:
2019 if (need_resched()) {
2020 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002021 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002022 }
2023 }
2024 rcu_read_unlock();
2025 }
2026
2027 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002028}
2029
2030#define F_ALL_SEALS (F_SEAL_SEAL | \
2031 F_SEAL_SHRINK | \
2032 F_SEAL_GROW | \
2033 F_SEAL_WRITE)
2034
2035int shmem_add_seals(struct file *file, unsigned int seals)
2036{
2037 struct inode *inode = file_inode(file);
2038 struct shmem_inode_info *info = SHMEM_I(inode);
2039 int error;
2040
2041 /*
2042 * SEALING
2043 * Sealing allows multiple parties to share a shmem-file but restrict
2044 * access to a specific subset of file operations. Seals can only be
2045 * added, but never removed. This way, mutually untrusted parties can
2046 * share common memory regions with a well-defined policy. A malicious
2047 * peer can thus never perform unwanted operations on a shared object.
2048 *
2049 * Seals are only supported on special shmem-files and always affect
2050 * the whole underlying inode. Once a seal is set, it may prevent some
2051 * kinds of access to the file. Currently, the following seals are
2052 * defined:
2053 * SEAL_SEAL: Prevent further seals from being set on this file
2054 * SEAL_SHRINK: Prevent the file from shrinking
2055 * SEAL_GROW: Prevent the file from growing
2056 * SEAL_WRITE: Prevent write access to the file
2057 *
2058 * As we don't require any trust relationship between two parties, we
2059 * must prevent seals from being removed. Therefore, sealing a file
2060 * only adds a given set of seals to the file, it never touches
2061 * existing seals. Furthermore, the "setting seals"-operation can be
2062 * sealed itself, which basically prevents any further seal from being
2063 * added.
2064 *
2065 * Semantics of sealing are only defined on volatile files. Only
2066 * anonymous shmem files support sealing. More importantly, seals are
2067 * never written to disk. Therefore, there's no plan to support it on
2068 * other file types.
2069 */
2070
2071 if (file->f_op != &shmem_file_operations)
2072 return -EINVAL;
2073 if (!(file->f_mode & FMODE_WRITE))
2074 return -EPERM;
2075 if (seals & ~(unsigned int)F_ALL_SEALS)
2076 return -EINVAL;
2077
Al Viro59551022016-01-22 15:40:57 -05002078 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002079
2080 if (info->seals & F_SEAL_SEAL) {
2081 error = -EPERM;
2082 goto unlock;
2083 }
2084
2085 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2086 error = mapping_deny_writable(file->f_mapping);
2087 if (error)
2088 goto unlock;
2089
2090 error = shmem_wait_for_pins(file->f_mapping);
2091 if (error) {
2092 mapping_allow_writable(file->f_mapping);
2093 goto unlock;
2094 }
2095 }
2096
2097 info->seals |= seals;
2098 error = 0;
2099
2100unlock:
Al Viro59551022016-01-22 15:40:57 -05002101 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002102 return error;
2103}
2104EXPORT_SYMBOL_GPL(shmem_add_seals);
2105
2106int shmem_get_seals(struct file *file)
2107{
2108 if (file->f_op != &shmem_file_operations)
2109 return -EINVAL;
2110
2111 return SHMEM_I(file_inode(file))->seals;
2112}
2113EXPORT_SYMBOL_GPL(shmem_get_seals);
2114
2115long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2116{
2117 long error;
2118
2119 switch (cmd) {
2120 case F_ADD_SEALS:
2121 /* disallow upper 32bit */
2122 if (arg > UINT_MAX)
2123 return -EINVAL;
2124
2125 error = shmem_add_seals(file, arg);
2126 break;
2127 case F_GET_SEALS:
2128 error = shmem_get_seals(file);
2129 break;
2130 default:
2131 error = -EINVAL;
2132 break;
2133 }
2134
2135 return error;
2136}
2137
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002138static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2139 loff_t len)
2140{
Al Viro496ad9a2013-01-23 17:07:38 -05002141 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002142 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002143 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002144 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002145 pgoff_t start, index, end;
2146 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002147
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002148 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2149 return -EOPNOTSUPP;
2150
Al Viro59551022016-01-22 15:40:57 -05002151 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002152
2153 if (mode & FALLOC_FL_PUNCH_HOLE) {
2154 struct address_space *mapping = file->f_mapping;
2155 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2156 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002157 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002158
David Herrmann40e041a2014-08-08 14:25:27 -07002159 /* protected by i_mutex */
2160 if (info->seals & F_SEAL_WRITE) {
2161 error = -EPERM;
2162 goto out;
2163 }
2164
Hugh Dickins8e205f72014-07-23 14:00:10 -07002165 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002166 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2167 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2168 spin_lock(&inode->i_lock);
2169 inode->i_private = &shmem_falloc;
2170 spin_unlock(&inode->i_lock);
2171
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002172 if ((u64)unmap_end > (u64)unmap_start)
2173 unmap_mapping_range(mapping, unmap_start,
2174 1 + unmap_end - unmap_start, 0);
2175 shmem_truncate_range(inode, offset, offset + len - 1);
2176 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002177
2178 spin_lock(&inode->i_lock);
2179 inode->i_private = NULL;
2180 wake_up_all(&shmem_falloc_waitq);
2181 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002182 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002183 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002184 }
2185
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002186 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2187 error = inode_newsize_ok(inode, offset + len);
2188 if (error)
2189 goto out;
2190
David Herrmann40e041a2014-08-08 14:25:27 -07002191 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2192 error = -EPERM;
2193 goto out;
2194 }
2195
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002196 start = offset >> PAGE_SHIFT;
2197 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002198 /* Try to avoid a swapstorm if len is impossible to satisfy */
2199 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2200 error = -ENOSPC;
2201 goto out;
2202 }
2203
Hugh Dickins8e205f72014-07-23 14:00:10 -07002204 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002205 shmem_falloc.start = start;
2206 shmem_falloc.next = start;
2207 shmem_falloc.nr_falloced = 0;
2208 shmem_falloc.nr_unswapped = 0;
2209 spin_lock(&inode->i_lock);
2210 inode->i_private = &shmem_falloc;
2211 spin_unlock(&inode->i_lock);
2212
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002213 for (index = start; index < end; index++) {
2214 struct page *page;
2215
2216 /*
2217 * Good, the fallocate(2) manpage permits EINTR: we may have
2218 * been interrupted because we are using up too much memory.
2219 */
2220 if (signal_pending(current))
2221 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002222 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2223 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002224 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002225 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002226 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002227 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002228 if (index > start) {
2229 shmem_undo_range(inode,
2230 (loff_t)start << PAGE_SHIFT,
2231 ((loff_t)index << PAGE_SHIFT) - 1, true);
2232 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002233 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002234 }
2235
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002236 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002237 * Inform shmem_writepage() how far we have reached.
2238 * No need for lock or barrier: we have the page lock.
2239 */
2240 shmem_falloc.next++;
2241 if (!PageUptodate(page))
2242 shmem_falloc.nr_falloced++;
2243
2244 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002245 * If !PageUptodate, leave it that way so that freeable pages
2246 * can be recognized if we need to rollback on error later.
2247 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002248 * than free the pages we are allocating (and SGP_CACHE pages
2249 * might still be clean: we now need to mark those dirty too).
2250 */
2251 set_page_dirty(page);
2252 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002253 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002254 cond_resched();
2255 }
2256
2257 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2258 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002259 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002260undone:
2261 spin_lock(&inode->i_lock);
2262 inode->i_private = NULL;
2263 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002264out:
Al Viro59551022016-01-22 15:40:57 -05002265 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002266 return error;
2267}
2268
David Howells726c3342006-06-23 02:02:58 -07002269static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270{
David Howells726c3342006-06-23 02:02:58 -07002271 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
2273 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002274 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002276 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002278 buf->f_bavail =
2279 buf->f_bfree = sbinfo->max_blocks -
2280 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002281 }
2282 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 buf->f_files = sbinfo->max_inodes;
2284 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 }
2286 /* else leave those fields 0 like simple_statfs */
2287 return 0;
2288}
2289
2290/*
2291 * File creation. Allocate an inode, and we're done..
2292 */
2293static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002294shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002296 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 int error = -ENOSPC;
2298
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002299 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002301 error = simple_acl_create(dir, inode);
2302 if (error)
2303 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002304 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002305 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002306 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002307 if (error && error != -EOPNOTSUPP)
2308 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002309
Al Viro718deb62009-12-16 19:35:36 -05002310 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 dir->i_size += BOGO_DIRENT_SIZE;
2312 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2313 d_instantiate(dentry, inode);
2314 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
2316 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002317out_iput:
2318 iput(inode);
2319 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320}
2321
Al Viro60545d02013-06-07 01:20:27 -04002322static int
2323shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2324{
2325 struct inode *inode;
2326 int error = -ENOSPC;
2327
2328 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2329 if (inode) {
2330 error = security_inode_init_security(inode, dir,
2331 NULL,
2332 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002333 if (error && error != -EOPNOTSUPP)
2334 goto out_iput;
2335 error = simple_acl_create(dir, inode);
2336 if (error)
2337 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002338 d_tmpfile(dentry, inode);
2339 }
2340 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002341out_iput:
2342 iput(inode);
2343 return error;
Al Viro60545d02013-06-07 01:20:27 -04002344}
2345
Al Viro18bb1db2011-07-26 01:41:39 -04002346static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347{
2348 int error;
2349
2350 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2351 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002352 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 return 0;
2354}
2355
Al Viro4acdaf22011-07-26 01:42:34 -04002356static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002357 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358{
2359 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2360}
2361
2362/*
2363 * Link a file..
2364 */
2365static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2366{
David Howells75c3cfa2015-03-17 22:26:12 +00002367 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002368 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
2370 /*
2371 * No ordinary (disk based) filesystem counts links as inodes;
2372 * but each new link needs a new dentry, pinning lowmem, and
2373 * tmpfs dentries cannot be pruned until they are unlinked.
2374 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002375 ret = shmem_reserve_inode(inode->i_sb);
2376 if (ret)
2377 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
2379 dir->i_size += BOGO_DIRENT_SIZE;
2380 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002381 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002382 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 dget(dentry); /* Extra pinning count for the created dentry */
2384 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002385out:
2386 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387}
2388
2389static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2390{
David Howells75c3cfa2015-03-17 22:26:12 +00002391 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002393 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2394 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
2396 dir->i_size -= BOGO_DIRENT_SIZE;
2397 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002398 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 dput(dentry); /* Undo the count from "create" - this does all the work */
2400 return 0;
2401}
2402
2403static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2404{
2405 if (!simple_empty(dentry))
2406 return -ENOTEMPTY;
2407
David Howells75c3cfa2015-03-17 22:26:12 +00002408 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002409 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 return shmem_unlink(dir, dentry);
2411}
2412
Miklos Szeredi37456772014-07-23 15:15:34 +02002413static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2414{
David Howellse36cb0b2015-01-29 12:02:35 +00002415 bool old_is_dir = d_is_dir(old_dentry);
2416 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002417
2418 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2419 if (old_is_dir) {
2420 drop_nlink(old_dir);
2421 inc_nlink(new_dir);
2422 } else {
2423 drop_nlink(new_dir);
2424 inc_nlink(old_dir);
2425 }
2426 }
2427 old_dir->i_ctime = old_dir->i_mtime =
2428 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002429 d_inode(old_dentry)->i_ctime =
2430 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
Miklos Szeredi37456772014-07-23 15:15:34 +02002431
2432 return 0;
2433}
2434
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002435static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2436{
2437 struct dentry *whiteout;
2438 int error;
2439
2440 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2441 if (!whiteout)
2442 return -ENOMEM;
2443
2444 error = shmem_mknod(old_dir, whiteout,
2445 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2446 dput(whiteout);
2447 if (error)
2448 return error;
2449
2450 /*
2451 * Cheat and hash the whiteout while the old dentry is still in
2452 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2453 *
2454 * d_lookup() will consistently find one of them at this point,
2455 * not sure which one, but that isn't even important.
2456 */
2457 d_rehash(whiteout);
2458 return 0;
2459}
2460
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461/*
2462 * The VFS layer already does all the dentry stuff for rename,
2463 * we just have to decrement the usage count for the target if
2464 * it exists so that the VFS layer correctly free's it when it
2465 * gets overwritten.
2466 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002467static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468{
David Howells75c3cfa2015-03-17 22:26:12 +00002469 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 int they_are_dirs = S_ISDIR(inode->i_mode);
2471
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002472 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002473 return -EINVAL;
2474
Miklos Szeredi37456772014-07-23 15:15:34 +02002475 if (flags & RENAME_EXCHANGE)
2476 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2477
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 if (!simple_empty(new_dentry))
2479 return -ENOTEMPTY;
2480
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002481 if (flags & RENAME_WHITEOUT) {
2482 int error;
2483
2484 error = shmem_whiteout(old_dir, old_dentry);
2485 if (error)
2486 return error;
2487 }
2488
David Howells75c3cfa2015-03-17 22:26:12 +00002489 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002491 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002492 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002493 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002496 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002497 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 }
2499
2500 old_dir->i_size -= BOGO_DIRENT_SIZE;
2501 new_dir->i_size += BOGO_DIRENT_SIZE;
2502 old_dir->i_ctime = old_dir->i_mtime =
2503 new_dir->i_ctime = new_dir->i_mtime =
2504 inode->i_ctime = CURRENT_TIME;
2505 return 0;
2506}
2507
2508static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2509{
2510 int error;
2511 int len;
2512 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002513 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 struct shmem_inode_info *info;
2515
2516 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002517 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 return -ENAMETOOLONG;
2519
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002520 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 if (!inode)
2522 return -ENOSPC;
2523
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002524 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002525 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002526 if (error) {
2527 if (error != -EOPNOTSUPP) {
2528 iput(inode);
2529 return error;
2530 }
2531 error = 0;
2532 }
2533
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 info = SHMEM_I(inode);
2535 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002536 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05002537 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
2538 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002539 iput(inode);
2540 return -ENOMEM;
2541 }
2542 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05002544 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002545 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 if (error) {
2547 iput(inode);
2548 return error;
2549 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002550 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05002552 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002553 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002555 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002556 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 dir->i_size += BOGO_DIRENT_SIZE;
2559 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2560 d_instantiate(dentry, inode);
2561 dget(dentry);
2562 return 0;
2563}
2564
Al Virofceef392015-12-29 15:58:39 -05002565static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566{
Al Virofceef392015-12-29 15:58:39 -05002567 mark_page_accessed(arg);
2568 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569}
2570
Al Viro6b255392015-11-17 10:20:54 -05002571static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05002572 struct inode *inode,
2573 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05002576 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05002577 if (!dentry) {
2578 page = find_get_page(inode->i_mapping, 0);
2579 if (!page)
2580 return ERR_PTR(-ECHILD);
2581 if (!PageUptodate(page)) {
2582 put_page(page);
2583 return ERR_PTR(-ECHILD);
2584 }
2585 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002586 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05002587 if (error)
2588 return ERR_PTR(error);
2589 unlock_page(page);
2590 }
Al Virofceef392015-12-29 15:58:39 -05002591 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05002592 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593}
2594
Eric Parisb09e0fa2011-05-24 17:12:39 -07002595#ifdef CONFIG_TMPFS_XATTR
2596/*
2597 * Superblocks without xattr inode operations may get some security.* xattr
2598 * support from the LSM "for free". As soon as we have any other xattrs
2599 * like ACLs, we also need to implement the security.* handlers at
2600 * filesystem level, though.
2601 */
2602
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002603/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002604 * Callback for security_inode_init_security() for acquiring xattrs.
2605 */
2606static int shmem_initxattrs(struct inode *inode,
2607 const struct xattr *xattr_array,
2608 void *fs_info)
2609{
2610 struct shmem_inode_info *info = SHMEM_I(inode);
2611 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002612 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002613 size_t len;
2614
2615 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002616 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002617 if (!new_xattr)
2618 return -ENOMEM;
2619
2620 len = strlen(xattr->name) + 1;
2621 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2622 GFP_KERNEL);
2623 if (!new_xattr->name) {
2624 kfree(new_xattr);
2625 return -ENOMEM;
2626 }
2627
2628 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2629 XATTR_SECURITY_PREFIX_LEN);
2630 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2631 xattr->name, len);
2632
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002633 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002634 }
2635
2636 return 0;
2637}
2638
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002639static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04002640 struct dentry *unused, struct inode *inode,
2641 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002642{
Al Virob2968212016-04-10 20:48:24 -04002643 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002644
2645 name = xattr_full_name(handler, name);
2646 return simple_xattr_get(&info->xattrs, name, buffer, size);
2647}
2648
2649static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04002650 struct dentry *unused, struct inode *inode,
2651 const char *name, const void *value,
2652 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002653{
Al Viro59301222016-05-27 10:19:30 -04002654 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002655
2656 name = xattr_full_name(handler, name);
2657 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2658}
2659
2660static const struct xattr_handler shmem_security_xattr_handler = {
2661 .prefix = XATTR_SECURITY_PREFIX,
2662 .get = shmem_xattr_handler_get,
2663 .set = shmem_xattr_handler_set,
2664};
2665
2666static const struct xattr_handler shmem_trusted_xattr_handler = {
2667 .prefix = XATTR_TRUSTED_PREFIX,
2668 .get = shmem_xattr_handler_get,
2669 .set = shmem_xattr_handler_set,
2670};
2671
Eric Parisb09e0fa2011-05-24 17:12:39 -07002672static const struct xattr_handler *shmem_xattr_handlers[] = {
2673#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002674 &posix_acl_access_xattr_handler,
2675 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002676#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002677 &shmem_security_xattr_handler,
2678 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002679 NULL
2680};
2681
Eric Parisb09e0fa2011-05-24 17:12:39 -07002682static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2683{
David Howells75c3cfa2015-03-17 22:26:12 +00002684 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01002685 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002686}
2687#endif /* CONFIG_TMPFS_XATTR */
2688
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002689static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05002691 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002692#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002693 .setxattr = generic_setxattr,
2694 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002695 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002696 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002697#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698};
2699
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002700static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05002702 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002703#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002704 .setxattr = generic_setxattr,
2705 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002706 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002707 .removexattr = generic_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002708#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002709};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002710
David M. Grimes91828a42006-10-17 00:09:45 -07002711static struct dentry *shmem_get_parent(struct dentry *child)
2712{
2713 return ERR_PTR(-ESTALE);
2714}
2715
2716static int shmem_match(struct inode *ino, void *vfh)
2717{
2718 __u32 *fh = vfh;
2719 __u64 inum = fh[2];
2720 inum = (inum << 32) | fh[1];
2721 return ino->i_ino == inum && fh[0] == ino->i_generation;
2722}
2723
Christoph Hellwig480b1162007-10-21 16:42:13 -07002724static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2725 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002726{
David M. Grimes91828a42006-10-17 00:09:45 -07002727 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002728 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002729 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002730
Christoph Hellwig480b1162007-10-21 16:42:13 -07002731 if (fh_len < 3)
2732 return NULL;
2733
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002734 inum = fid->raw[2];
2735 inum = (inum << 32) | fid->raw[1];
2736
Christoph Hellwig480b1162007-10-21 16:42:13 -07002737 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2738 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002739 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002740 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002741 iput(inode);
2742 }
2743
Christoph Hellwig480b1162007-10-21 16:42:13 -07002744 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002745}
2746
Al Virob0b03822012-04-02 14:34:06 -04002747static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2748 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002749{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302750 if (*len < 3) {
2751 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002752 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302753 }
David M. Grimes91828a42006-10-17 00:09:45 -07002754
Al Viro1d3382cb2010-10-23 15:19:20 -04002755 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002756 /* Unfortunately insert_inode_hash is not idempotent,
2757 * so as we hash inodes here rather than at creation
2758 * time, we need a lock to ensure we only try
2759 * to do it once
2760 */
2761 static DEFINE_SPINLOCK(lock);
2762 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002763 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002764 __insert_inode_hash(inode,
2765 inode->i_ino + inode->i_generation);
2766 spin_unlock(&lock);
2767 }
2768
2769 fh[0] = inode->i_generation;
2770 fh[1] = inode->i_ino;
2771 fh[2] = ((__u64)inode->i_ino) >> 32;
2772
2773 *len = 3;
2774 return 1;
2775}
2776
Christoph Hellwig39655162007-10-21 16:42:17 -07002777static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002778 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002779 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002780 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002781};
2782
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002783static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2784 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785{
2786 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002787 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002788 uid_t uid;
2789 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002791 while (options != NULL) {
2792 this_char = options;
2793 for (;;) {
2794 /*
2795 * NUL-terminate this option: unfortunately,
2796 * mount options form a comma-separated list,
2797 * but mpol's nodelist may also contain commas.
2798 */
2799 options = strchr(options, ',');
2800 if (options == NULL)
2801 break;
2802 options++;
2803 if (!isdigit(*options)) {
2804 options[-1] = '\0';
2805 break;
2806 }
2807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 if (!*this_char)
2809 continue;
2810 if ((value = strchr(this_char,'=')) != NULL) {
2811 *value++ = 0;
2812 } else {
Joe Perches11705322016-03-17 14:19:50 -07002813 pr_err("tmpfs: No value for mount option '%s'\n",
2814 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002815 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 }
2817
2818 if (!strcmp(this_char,"size")) {
2819 unsigned long long size;
2820 size = memparse(value,&rest);
2821 if (*rest == '%') {
2822 size <<= PAGE_SHIFT;
2823 size *= totalram_pages;
2824 do_div(size, 100);
2825 rest++;
2826 }
2827 if (*rest)
2828 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002829 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002830 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002832 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 if (*rest)
2834 goto bad_val;
2835 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002836 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 if (*rest)
2838 goto bad_val;
2839 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002840 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002842 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 if (*rest)
2844 goto bad_val;
2845 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002846 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002848 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 if (*rest)
2850 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002851 sbinfo->uid = make_kuid(current_user_ns(), uid);
2852 if (!uid_valid(sbinfo->uid))
2853 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002855 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002857 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 if (*rest)
2859 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002860 sbinfo->gid = make_kgid(current_user_ns(), gid);
2861 if (!gid_valid(sbinfo->gid))
2862 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002863 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002864 mpol_put(mpol);
2865 mpol = NULL;
2866 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002867 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 } else {
Joe Perches11705322016-03-17 14:19:50 -07002869 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002870 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 }
2872 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002873 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 return 0;
2875
2876bad_val:
Joe Perches11705322016-03-17 14:19:50 -07002877 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002879error:
2880 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 return 1;
2882
2883}
2884
2885static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2886{
2887 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002888 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002889 unsigned long inodes;
2890 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
Greg Thelen5f001102013-02-22 16:36:01 -08002892 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002893 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002894 return error;
2895
2896 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002897 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002898 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002899 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002900 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002901 goto out;
2902 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002903 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002904 * but we must separately disallow unlimited->limited, because
2905 * in that case we have no record of how much is already in use.
2906 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002907 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002908 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002909 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002910 goto out;
2911
2912 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002913 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002914 sbinfo->max_inodes = config.max_inodes;
2915 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002916
Greg Thelen5f001102013-02-22 16:36:01 -08002917 /*
2918 * Preserve previous mempolicy unless mpol remount option was specified.
2919 */
2920 if (config.mpol) {
2921 mpol_put(sbinfo->mpol);
2922 sbinfo->mpol = config.mpol; /* transfers initial ref */
2923 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002924out:
2925 spin_unlock(&sbinfo->stat_lock);
2926 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002928
Al Viro34c80b12011-12-08 21:32:45 -05002929static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002930{
Al Viro34c80b12011-12-08 21:32:45 -05002931 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002932
2933 if (sbinfo->max_blocks != shmem_default_max_blocks())
2934 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002935 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002936 if (sbinfo->max_inodes != shmem_default_max_inodes())
2937 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2938 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002939 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002940 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2941 seq_printf(seq, ",uid=%u",
2942 from_kuid_munged(&init_user_ns, sbinfo->uid));
2943 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2944 seq_printf(seq, ",gid=%u",
2945 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002946 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002947 return 0;
2948}
David Herrmann9183df22014-08-08 14:25:29 -07002949
2950#define MFD_NAME_PREFIX "memfd:"
2951#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2952#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2953
2954#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2955
2956SYSCALL_DEFINE2(memfd_create,
2957 const char __user *, uname,
2958 unsigned int, flags)
2959{
2960 struct shmem_inode_info *info;
2961 struct file *file;
2962 int fd, error;
2963 char *name;
2964 long len;
2965
2966 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2967 return -EINVAL;
2968
2969 /* length includes terminating zero */
2970 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2971 if (len <= 0)
2972 return -EFAULT;
2973 if (len > MFD_NAME_MAX_LEN + 1)
2974 return -EINVAL;
2975
2976 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2977 if (!name)
2978 return -ENOMEM;
2979
2980 strcpy(name, MFD_NAME_PREFIX);
2981 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2982 error = -EFAULT;
2983 goto err_name;
2984 }
2985
2986 /* terminating-zero may have changed after strnlen_user() returned */
2987 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
2988 error = -EFAULT;
2989 goto err_name;
2990 }
2991
2992 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
2993 if (fd < 0) {
2994 error = fd;
2995 goto err_name;
2996 }
2997
2998 file = shmem_file_setup(name, 0, VM_NORESERVE);
2999 if (IS_ERR(file)) {
3000 error = PTR_ERR(file);
3001 goto err_fd;
3002 }
3003 info = SHMEM_I(file_inode(file));
3004 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3005 file->f_flags |= O_RDWR | O_LARGEFILE;
3006 if (flags & MFD_ALLOW_SEALING)
3007 info->seals &= ~F_SEAL_SEAL;
3008
3009 fd_install(fd, file);
3010 kfree(name);
3011 return fd;
3012
3013err_fd:
3014 put_unused_fd(fd);
3015err_name:
3016 kfree(name);
3017 return error;
3018}
3019
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003020#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021
3022static void shmem_put_super(struct super_block *sb)
3023{
Hugh Dickins602586a2010-08-17 15:23:56 -07003024 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3025
3026 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003027 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003028 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 sb->s_fs_info = NULL;
3030}
3031
Kay Sievers2b2af542009-04-30 15:23:42 +02003032int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033{
3034 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003035 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003036 int err = -ENOMEM;
3037
3038 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003039 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003040 L1_CACHE_BYTES), GFP_KERNEL);
3041 if (!sbinfo)
3042 return -ENOMEM;
3043
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003044 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003045 sbinfo->uid = current_fsuid();
3046 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003047 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003049#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 /*
3051 * Per default we only allow half of the physical ram per
3052 * tmpfs instance, limiting inodes to one per page of lowmem;
3053 * but the internal instance is left unlimited.
3054 */
Al Viroca4e0512013-08-31 12:57:10 -04003055 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003056 sbinfo->max_blocks = shmem_default_max_blocks();
3057 sbinfo->max_inodes = shmem_default_max_inodes();
3058 if (shmem_parse_options(data, sbinfo, false)) {
3059 err = -EINVAL;
3060 goto failed;
3061 }
Al Viroca4e0512013-08-31 12:57:10 -04003062 } else {
3063 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 }
David M. Grimes91828a42006-10-17 00:09:45 -07003065 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003066 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067#else
3068 sb->s_flags |= MS_NOUSER;
3069#endif
3070
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003071 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003072 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003073 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003074 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003075
Hugh Dickins285b2c42011-08-03 16:21:20 -07003076 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003077 sb->s_blocksize = PAGE_SIZE;
3078 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 sb->s_magic = TMPFS_MAGIC;
3080 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003081 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003082#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003083 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003084#endif
3085#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003086 sb->s_flags |= MS_POSIXACL;
3087#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003088
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003089 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 if (!inode)
3091 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003092 inode->i_uid = sbinfo->uid;
3093 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003094 sb->s_root = d_make_root(inode);
3095 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003096 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 return 0;
3098
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099failed:
3100 shmem_put_super(sb);
3101 return err;
3102}
3103
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003104static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105
3106static struct inode *shmem_alloc_inode(struct super_block *sb)
3107{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003108 struct shmem_inode_info *info;
3109 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3110 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003112 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113}
3114
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003115static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003116{
3117 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003118 if (S_ISLNK(inode->i_mode))
3119 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003120 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3121}
3122
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123static void shmem_destroy_inode(struct inode *inode)
3124{
Al Viro09208d12011-07-26 03:15:03 -04003125 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003127 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128}
3129
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003130static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003132 struct shmem_inode_info *info = foo;
3133 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134}
3135
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003136static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137{
3138 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3139 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003140 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 return 0;
3142}
3143
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003144static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003146 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147}
3148
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003149static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003151 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003153 .write_begin = shmem_write_begin,
3154 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003156#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003157 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003158#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003159 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160};
3161
Helge Deller15ad7cd2006-12-06 20:40:36 -08003162static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 .mmap = shmem_mmap,
3164#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003165 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003166 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003167 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003168 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003169 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003170 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003171 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172#endif
3173};
3174
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003175static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003176 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003177 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003178#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003179 .setxattr = generic_setxattr,
3180 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003181 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003182 .removexattr = generic_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003183 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003184#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185};
3186
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003187static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188#ifdef CONFIG_TMPFS
3189 .create = shmem_create,
3190 .lookup = simple_lookup,
3191 .link = shmem_link,
3192 .unlink = shmem_unlink,
3193 .symlink = shmem_symlink,
3194 .mkdir = shmem_mkdir,
3195 .rmdir = shmem_rmdir,
3196 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003197 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003198 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003200#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003201 .setxattr = generic_setxattr,
3202 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003203 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003204 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003205#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003206#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003207 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003208 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003209#endif
3210};
3211
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003212static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003213#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003214 .setxattr = generic_setxattr,
3215 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003216 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003217 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003218#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003219#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003220 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003221 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003222#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223};
3224
Hugh Dickins759b9772007-03-05 00:30:28 -08003225static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 .alloc_inode = shmem_alloc_inode,
3227 .destroy_inode = shmem_destroy_inode,
3228#ifdef CONFIG_TMPFS
3229 .statfs = shmem_statfs,
3230 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003231 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232#endif
Al Viro1f895f72010-06-05 19:10:41 -04003233 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 .drop_inode = generic_delete_inode,
3235 .put_super = shmem_put_super,
3236};
3237
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003238static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003239 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003240 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241#ifdef CONFIG_NUMA
3242 .set_policy = shmem_set_policy,
3243 .get_policy = shmem_get_policy,
3244#endif
3245};
3246
Al Viro3c26ff62010-07-25 11:46:36 +04003247static struct dentry *shmem_mount(struct file_system_type *fs_type,
3248 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249{
Al Viro3c26ff62010-07-25 11:46:36 +04003250 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251}
3252
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003253static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 .owner = THIS_MODULE,
3255 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003256 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003258 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003261int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262{
3263 int error;
3264
Rob Landley16203a72013-09-11 14:26:12 -07003265 /* If rootfs called this, don't re-init */
3266 if (shmem_inode_cachep)
3267 return 0;
3268
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003269 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 if (error)
3271 goto out3;
3272
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003273 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003275 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 goto out2;
3277 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003278
Al Viroca4e0512013-08-31 12:57:10 -04003279 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 if (IS_ERR(shm_mnt)) {
3281 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003282 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 goto out1;
3284 }
3285 return 0;
3286
3287out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003288 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003290 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291out3:
3292 shm_mnt = ERR_PTR(error);
3293 return error;
3294}
Matt Mackall853ac432009-01-06 14:40:20 -08003295
3296#else /* !CONFIG_SHMEM */
3297
3298/*
3299 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3300 *
3301 * This is intended for small system where the benefits of the full
3302 * shmem code (swap-backed and resource-limited) are outweighed by
3303 * their complexity. On systems without swap this code should be
3304 * effectively equivalent, but much lighter weight.
3305 */
3306
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003307static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003308 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003309 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003310 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003311 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003312};
3313
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003314int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003315{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003316 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003317
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003318 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003319 BUG_ON(IS_ERR(shm_mnt));
3320
3321 return 0;
3322}
3323
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003324int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003325{
3326 return 0;
3327}
3328
Hugh Dickins3f96b792009-09-21 17:03:37 -07003329int shmem_lock(struct file *file, int lock, struct user_struct *user)
3330{
3331 return 0;
3332}
3333
Hugh Dickins24513262012-01-20 14:34:21 -08003334void shmem_unlock_mapping(struct address_space *mapping)
3335{
3336}
3337
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003338void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003339{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003340 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003341}
3342EXPORT_SYMBOL_GPL(shmem_truncate_range);
3343
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003344#define shmem_vm_ops generic_file_vm_ops
3345#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003346#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003347#define shmem_acct_size(flags, size) 0
3348#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003349
3350#endif /* CONFIG_SHMEM */
3351
3352/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353
Al Viro34515382013-02-14 22:38:02 -05003354static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003355 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003356};
3357
Eric Parisc7277092013-12-02 11:24:19 +00003358static struct file *__shmem_file_setup(const char *name, loff_t size,
3359 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360{
Al Viro6b4d0b22013-02-14 21:37:26 -05003361 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003363 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003364 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 struct qstr this;
3366
3367 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003368 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369
Hugh Dickins285b2c42011-08-03 16:21:20 -07003370 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 return ERR_PTR(-EINVAL);
3372
3373 if (shmem_acct_size(flags, size))
3374 return ERR_PTR(-ENOMEM);
3375
Al Viro6b4d0b22013-02-14 21:37:26 -05003376 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 this.name = name;
3378 this.len = strlen(name);
3379 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003380 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003381 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003382 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003383 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003385 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386
Al Viro6b4d0b22013-02-14 21:37:26 -05003387 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05003388 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003390 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391
Eric Parisc7277092013-12-02 11:24:19 +00003392 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04003393 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003395 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05003396 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3397 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003398 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003399
Al Viro6b4d0b22013-02-14 21:37:26 -05003400 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04003401 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05003402 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003403 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003404
Al Viro6b4d0b22013-02-14 21:37:26 -05003405 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407put_memory:
3408 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003409put_path:
3410 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05003411 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412}
Eric Parisc7277092013-12-02 11:24:19 +00003413
3414/**
3415 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3416 * kernel internal. There will be NO LSM permission checks against the
3417 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07003418 * higher layer. The users are the big_key and shm implementations. LSM
3419 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00003420 * @name: name for dentry (to be seen in /proc/<pid>/maps
3421 * @size: size to be set for the file
3422 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3423 */
3424struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3425{
3426 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3427}
3428
3429/**
3430 * shmem_file_setup - get an unlinked file living in tmpfs
3431 * @name: name for dentry (to be seen in /proc/<pid>/maps
3432 * @size: size to be set for the file
3433 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3434 */
3435struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3436{
3437 return __shmem_file_setup(name, size, flags, 0);
3438}
Keith Packard395e0dd2008-06-20 00:08:06 -07003439EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440
Randy Dunlap46711812008-03-19 17:00:41 -07003441/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3444 */
3445int shmem_zero_setup(struct vm_area_struct *vma)
3446{
3447 struct file *file;
3448 loff_t size = vma->vm_end - vma->vm_start;
3449
Hugh Dickins66fc1302015-06-14 09:48:09 -07003450 /*
3451 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3452 * between XFS directory reading and selinux: since this file is only
3453 * accessible to the user through its mapping, use S_PRIVATE flag to
3454 * bypass file security, in the same way as shmem_kernel_file_setup().
3455 */
3456 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 if (IS_ERR(file))
3458 return PTR_ERR(file);
3459
3460 if (vma->vm_file)
3461 fput(vma->vm_file);
3462 vma->vm_file = file;
3463 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 return 0;
3465}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003466
3467/**
3468 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3469 * @mapping: the page's address_space
3470 * @index: the page index
3471 * @gfp: the page allocator flags to use if allocating
3472 *
3473 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3474 * with any new page allocations done using the specified allocation flags.
3475 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3476 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3477 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3478 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003479 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3480 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003481 */
3482struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3483 pgoff_t index, gfp_t gfp)
3484{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003485#ifdef CONFIG_SHMEM
3486 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003487 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003488 int error;
3489
3490 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003491 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
3492 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003493 if (error)
3494 page = ERR_PTR(error);
3495 else
3496 unlock_page(page);
3497 return page;
3498#else
3499 /*
3500 * The tiny !SHMEM case uses ramfs without swap
3501 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003502 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003503#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003504}
3505EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);