blob: 9428c51ab2d6b026ee9a10afc7e21c70e0a33d80 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
80
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 Dickinsa0ee5ec2008-02-04 22:28:51 -0800104 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700105 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
106 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107};
108
Andrew Mortonb76db732008-02-08 04:21:49 -0800109#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800110static unsigned long shmem_default_max_blocks(void)
111{
112 return totalram_pages / 2;
113}
114
115static unsigned long shmem_default_max_inodes(void)
116{
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118}
Andrew Mortonb76db732008-02-08 04:21:49 -0800119#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800120
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700121static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
122static int shmem_replace_page(struct page **pagep, gfp_t gfp,
123 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700124static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
125 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
126
127static inline int shmem_getpage(struct inode *inode, pgoff_t index,
128 struct page **pagep, enum sgp_type sgp, int *fault_type)
129{
130 return shmem_getpage_gfp(inode, index, pagep, sgp,
131 mapping_gfp_mask(inode->i_mapping), fault_type);
132}
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
172 * pages are allocated, in order to allow huge sparse files.
173 * 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) ?
Al Viro191c5422012-02-13 03:58:52 +0000179 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
186}
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
Hugh Dickinsb065b432012-07-11 14:02:48 -0700303 page_cache_get(page);
304 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);
321 page_cache_release(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);
341 page_cache_release(page);
342 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);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700477 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700478 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
479 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
480 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_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;
531 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
532 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700533 unsigned int top = PAGE_CACHE_SIZE;
534 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);
541 page_cache_release(page);
542 }
543 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700544 if (partial_end) {
545 struct page *page = NULL;
546 shmem_getpage(inode, end, &page, SGP_READ, NULL);
547 if (page) {
548 zero_user_segment(page, 0, partial_end);
549 set_page_dirty(page);
550 unlock_page(page);
551 page_cache_release(page);
552 }
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);
836 page_cache_release(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
950#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800951#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700952static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800953{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700954 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800955
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700956 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700957 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800958
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800959 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800960
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700961 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800962}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700963
964static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
965{
966 struct mempolicy *mpol = NULL;
967 if (sbinfo->mpol) {
968 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
969 mpol = sbinfo->mpol;
970 mpol_get(mpol);
971 spin_unlock(&sbinfo->stat_lock);
972 }
973 return mpol;
974}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800975#endif /* CONFIG_TMPFS */
976
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700977static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
978 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800981 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800983 /* Create a pseudo vma that just contains the policy */
984 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700985 /* Bias interleave by inode number to distribute better across nodes */
986 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800987 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700988 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700989
Mel Gorman18a2f372012-12-05 14:01:41 -0800990 page = swapin_readahead(swap, gfp, &pvma, 0);
991
992 /* Drop reference taken by mpol_shared_policy_lookup() */
993 mpol_cond_put(pvma.vm_policy);
994
995 return page;
996}
997
998static struct page *shmem_alloc_page(gfp_t gfp,
999 struct shmem_inode_info *info, pgoff_t index)
1000{
1001 struct vm_area_struct pvma;
1002 struct page *page;
1003
1004 /* Create a pseudo vma that just contains the policy */
1005 pvma.vm_start = 0;
1006 /* Bias interleave by inode number to distribute better across nodes */
1007 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
1008 pvma.vm_ops = NULL;
1009 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1010
1011 page = alloc_page_vma(gfp, &pvma, 0);
1012
1013 /* Drop reference taken by mpol_shared_policy_lookup() */
1014 mpol_cond_put(pvma.vm_policy);
1015
1016 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001018#else /* !CONFIG_NUMA */
1019#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001020static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001021{
1022}
1023#endif /* CONFIG_TMPFS */
1024
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001025static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1026 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001028 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
Hugh Dickins02098fe2008-02-04 22:28:42 -08001031static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001032 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001034 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001036#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001038#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1039static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1040{
1041 return NULL;
1042}
1043#endif
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001046 * When a page is moved from swapcache to shmem filecache (either by the
1047 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1048 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1049 * ignorance of the mapping it belongs to. If that mapping has special
1050 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1051 * we may need to copy to a suitable page before moving to filecache.
1052 *
1053 * In a future release, this may well be extended to respect cpuset and
1054 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1055 * but for now it is a simple matter of zone.
1056 */
1057static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1058{
1059 return page_zonenum(page) > gfp_zone(gfp);
1060}
1061
1062static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1063 struct shmem_inode_info *info, pgoff_t index)
1064{
1065 struct page *oldpage, *newpage;
1066 struct address_space *swap_mapping;
1067 pgoff_t swap_index;
1068 int error;
1069
1070 oldpage = *pagep;
1071 swap_index = page_private(oldpage);
1072 swap_mapping = page_mapping(oldpage);
1073
1074 /*
1075 * We have arrived here because our zones are constrained, so don't
1076 * limit chance of success by further cpuset and node constraints.
1077 */
1078 gfp &= ~GFP_CONSTRAINT_MASK;
1079 newpage = shmem_alloc_page(gfp, info, index);
1080 if (!newpage)
1081 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001082
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001083 page_cache_get(newpage);
1084 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001085 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001086
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001087 __SetPageLocked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001088 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001089 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001090 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001091 SetPageSwapCache(newpage);
1092
1093 /*
1094 * Our caller will very soon move newpage out of swapcache, but it's
1095 * a nice clean interface for us to replace oldpage by newpage there.
1096 */
1097 spin_lock_irq(&swap_mapping->tree_lock);
1098 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1099 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001100 if (!error) {
1101 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1102 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1103 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001104 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001105
Hugh Dickins0142ef62012-06-07 14:21:09 -07001106 if (unlikely(error)) {
1107 /*
1108 * Is this possible? I think not, now that our callers check
1109 * both PageSwapCache and page_private after getting page lock;
1110 * but be defensive. Reverse old to newpage for clear and free.
1111 */
1112 oldpage = newpage;
1113 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001114 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001115 lru_cache_add_anon(newpage);
1116 *pagep = newpage;
1117 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001118
1119 ClearPageSwapCache(oldpage);
1120 set_page_private(oldpage, 0);
1121
1122 unlock_page(oldpage);
1123 page_cache_release(oldpage);
1124 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001125 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001126}
1127
1128/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001129 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 *
1131 * If we allocate a new one we do not mark it dirty. That's up to the
1132 * vm. If we swap it in we mark it dirty since we also free the swap
1133 * entry since a page cannot live in both the swap and page cache
1134 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001135static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001136 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
1138 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001139 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 struct shmem_sb_info *sbinfo;
Johannes Weiner00501b52014-08-08 14:19:20 -07001141 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001142 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 swp_entry_t swap;
1144 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001145 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001146 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001148 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001151 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001152 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001153 if (radix_tree_exceptional_entry(page)) {
1154 swap = radix_to_swp_entry(page);
1155 page = NULL;
1156 }
1157
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001158 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001159 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1160 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001161 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001162 }
1163
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001164 if (page && sgp == SGP_WRITE)
1165 mark_page_accessed(page);
1166
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001167 /* fallocated page? */
1168 if (page && !PageUptodate(page)) {
1169 if (sgp != SGP_READ)
1170 goto clear;
1171 unlock_page(page);
1172 page_cache_release(page);
1173 page = NULL;
1174 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001175 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001176 *pagep = page;
1177 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001181 * Fast cache lookup did not find it:
1182 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001184 info = SHMEM_I(inode);
1185 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 if (swap.val) {
1188 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001189 page = lookup_swap_cache(swap);
1190 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001191 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001192 if (fault_type)
1193 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001194 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001195 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001196 error = -ENOMEM;
1197 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 }
1200
1201 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001202 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001203 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001204 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001205 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001206 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001207 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001208 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001210 goto failed;
1211 }
1212 wait_on_page_writeback(page);
1213
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001214 if (shmem_should_replace_page(page, gfp)) {
1215 error = shmem_replace_page(&page, gfp, info, index);
1216 if (error)
1217 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
1219
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001220 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1221 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001222 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001223 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001224 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001225 /*
1226 * We already confirmed swap under page lock, and make
1227 * no memory allocation here, so usually no possibility
1228 * of error; but free_swap_and_cache() only trylocks a
1229 * page, so it is just possible that the entry has been
1230 * truncated or holepunched since swap was confirmed.
1231 * shmem_undo_range() will have done some of the
1232 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001233 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001234 * Reset swap.val? No, leave it so "failed" goes back to
1235 * "repeat": reading a hole and writing should succeed.
1236 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001237 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001238 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001239 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001240 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001241 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001242 if (error)
1243 goto failed;
1244
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001245 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001246
Hugh Dickins54af60422011-08-03 16:21:24 -07001247 spin_lock(&info->lock);
1248 info->swapped--;
1249 shmem_recalc_inode(inode);
1250 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001251
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001252 if (sgp == SGP_WRITE)
1253 mark_page_accessed(page);
1254
Hugh Dickins27ab7002011-07-25 17:12:36 -07001255 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001256 set_page_dirty(page);
1257 swap_free(swap);
1258
Hugh Dickins54af60422011-08-03 16:21:24 -07001259 } else {
1260 if (shmem_acct_block(info->flags)) {
1261 error = -ENOSPC;
1262 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001264 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001265 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001266 sbinfo->max_blocks) >= 0) {
1267 error = -ENOSPC;
1268 goto unacct;
1269 }
Tim Chen7e496292010-08-09 17:19:05 -07001270 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
Hugh Dickins54af60422011-08-03 16:21:24 -07001273 page = shmem_alloc_page(gfp, info, index);
1274 if (!page) {
1275 error = -ENOMEM;
1276 goto decused;
1277 }
1278
Mel Gorman07a42782014-06-04 16:10:24 -07001279 __SetPageSwapBacked(page);
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001280 __SetPageLocked(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001281 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001282 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001283
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001284 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1285 false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001286 if (error)
1287 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001288 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001289 if (!error) {
1290 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001291 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001292 radix_tree_preload_end();
1293 }
1294 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001295 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001296 goto decused;
1297 }
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001298 mem_cgroup_commit_charge(page, memcg, false, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001299 lru_cache_add_anon(page);
1300
1301 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001303 inode->i_blocks += BLOCKS_PER_PAGE;
1304 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001306 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001307
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001308 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001309 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1310 */
1311 if (sgp == SGP_FALLOC)
1312 sgp = SGP_WRITE;
1313clear:
1314 /*
1315 * Let SGP_WRITE caller clear ends if write does not fill page;
1316 * but SGP_FALLOC on a page fallocated earlier must initialize
1317 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001318 */
1319 if (sgp != SGP_WRITE) {
1320 clear_highpage(page);
1321 flush_dcache_page(page);
1322 SetPageUptodate(page);
1323 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001324 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001325 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001326 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001327
Hugh Dickins54af60422011-08-03 16:21:24 -07001328 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001329 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001330 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001331 if (alloced) {
1332 ClearPageDirty(page);
1333 delete_from_page_cache(page);
1334 spin_lock(&info->lock);
1335 shmem_recalc_inode(inode);
1336 spin_unlock(&info->lock);
1337 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001338 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001339 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001340 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001341 *pagep = page;
1342 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001343
Nick Piggind0217ac2007-07-19 01:47:03 -07001344 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001345 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001347decused:
1348 if (sbinfo->max_blocks)
1349 percpu_counter_add(&sbinfo->used_blocks, -1);
1350unacct:
1351 shmem_unacct_blocks(info->flags, 1);
1352failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001353 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001354 error = -EEXIST;
1355unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001356 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001357 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001358 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001359 }
1360 if (error == -ENOSPC && !once++) {
1361 info = SHMEM_I(inode);
1362 spin_lock(&info->lock);
1363 shmem_recalc_inode(inode);
1364 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001365 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001366 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001367 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001368 goto repeat;
1369 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370}
1371
1372static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1373{
Al Viro496ad9a2013-01-23 17:07:38 -05001374 struct inode *inode = file_inode(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001376 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001378 /*
1379 * Trinity finds that probing a hole which tmpfs is punching can
1380 * prevent the hole-punch from ever completing: which in turn
1381 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001382 * faulting pages into the hole while it's being punched. Although
1383 * shmem_undo_range() does remove the additions, it may be unable to
1384 * keep up, as each new page needs its own unmap_mapping_range() call,
1385 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1386 *
1387 * It does not matter if we sometimes reach this check just before the
1388 * hole-punch begins, so that one fault then races with the punch:
1389 * we just need to make racing faults a rare case.
1390 *
1391 * The implementation below would be much simpler if we just used a
1392 * standard mutex or completion: but we cannot take i_mutex in fault,
1393 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001394 */
1395 if (unlikely(inode->i_private)) {
1396 struct shmem_falloc *shmem_falloc;
1397
1398 spin_lock(&inode->i_lock);
1399 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001400 if (shmem_falloc &&
1401 shmem_falloc->waitq &&
1402 vmf->pgoff >= shmem_falloc->start &&
1403 vmf->pgoff < shmem_falloc->next) {
1404 wait_queue_head_t *shmem_falloc_waitq;
1405 DEFINE_WAIT(shmem_fault_wait);
1406
1407 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001408 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1409 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001410 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001411 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001412 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001413 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001414
1415 shmem_falloc_waitq = shmem_falloc->waitq;
1416 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1417 TASK_UNINTERRUPTIBLE);
1418 spin_unlock(&inode->i_lock);
1419 schedule();
1420
1421 /*
1422 * shmem_falloc_waitq points into the shmem_fallocate()
1423 * stack of the hole-punching task: shmem_falloc_waitq
1424 * is usually invalid by the time we reach here, but
1425 * finish_wait() does not dereference it in that case;
1426 * though i_lock needed lest racing with wake_up_all().
1427 */
1428 spin_lock(&inode->i_lock);
1429 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1430 spin_unlock(&inode->i_lock);
1431 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001432 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001433 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001434 }
1435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1437 if (error)
1438 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001439
Ying Han456f9982011-05-26 16:25:38 -07001440 if (ret & VM_FAULT_MAJOR) {
1441 count_vm_event(PGMAJFAULT);
1442 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1443 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001444 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445}
1446
1447#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001448static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
Al Viro496ad9a2013-01-23 17:07:38 -05001450 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001451 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001454static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1455 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
Al Viro496ad9a2013-01-23 17:07:38 -05001457 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001458 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001460 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1461 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463#endif
1464
1465int shmem_lock(struct file *file, int lock, struct user_struct *user)
1466{
Al Viro496ad9a2013-01-23 17:07:38 -05001467 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 struct shmem_inode_info *info = SHMEM_I(inode);
1469 int retval = -ENOMEM;
1470
1471 spin_lock(&info->lock);
1472 if (lock && !(info->flags & VM_LOCKED)) {
1473 if (!user_shm_lock(inode->i_size, user))
1474 goto out_nomem;
1475 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001476 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 }
1478 if (!lock && (info->flags & VM_LOCKED) && user) {
1479 user_shm_unlock(inode->i_size, user);
1480 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001481 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001484
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485out_nomem:
1486 spin_unlock(&info->lock);
1487 return retval;
1488}
1489
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001490static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
1492 file_accessed(file);
1493 vma->vm_ops = &shmem_vm_ops;
1494 return 0;
1495}
1496
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001497static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001498 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
1500 struct inode *inode;
1501 struct shmem_inode_info *info;
1502 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1503
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001504 if (shmem_reserve_inode(sb))
1505 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
1507 inode = new_inode(sb);
1508 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001509 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001510 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001513 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 info = SHMEM_I(inode);
1515 memset(info, 0, (char *)inode - (char *)info);
1516 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07001517 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001518 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001520 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001521 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
1523 switch (mode & S_IFMT) {
1524 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001525 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 init_special_inode(inode, mode, dev);
1527 break;
1528 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001529 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 inode->i_op = &shmem_inode_operations;
1531 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001532 mpol_shared_policy_init(&info->policy,
1533 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 break;
1535 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001536 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 /* Some things misbehave if size == 0 on a directory */
1538 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1539 inode->i_op = &shmem_dir_inode_operations;
1540 inode->i_fop = &simple_dir_operations;
1541 break;
1542 case S_IFLNK:
1543 /*
1544 * Must not load anything in the rbtree,
1545 * mpol_free_shared_policy will not be called.
1546 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001547 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 break;
1549 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001550 } else
1551 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 return inode;
1553}
1554
Johannes Weiner0cd61442014-04-03 14:47:46 -07001555bool shmem_mapping(struct address_space *mapping)
1556{
Sasha Levinf0774d82015-02-23 05:38:00 -05001557 if (!mapping->host)
1558 return false;
1559
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001560 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001561}
1562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001564static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001565static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001567#ifdef CONFIG_TMPFS_XATTR
1568static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1569#else
1570#define shmem_initxattrs NULL
1571#endif
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001574shmem_write_begin(struct file *file, struct address_space *mapping,
1575 loff_t pos, unsigned len, unsigned flags,
1576 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
Nick Piggin800d15a2007-10-16 01:25:03 -07001578 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07001579 struct shmem_inode_info *info = SHMEM_I(inode);
Nick Piggin800d15a2007-10-16 01:25:03 -07001580 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07001581
1582 /* i_mutex is held by caller */
1583 if (unlikely(info->seals)) {
1584 if (info->seals & F_SEAL_WRITE)
1585 return -EPERM;
1586 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1587 return -EPERM;
1588 }
1589
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001590 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
Nick Piggin800d15a2007-10-16 01:25:03 -07001591}
1592
1593static int
1594shmem_write_end(struct file *file, struct address_space *mapping,
1595 loff_t pos, unsigned len, unsigned copied,
1596 struct page *page, void *fsdata)
1597{
1598 struct inode *inode = mapping->host;
1599
Hugh Dickinsd3602442008-02-04 22:28:44 -08001600 if (pos + copied > inode->i_size)
1601 i_size_write(inode, pos + copied);
1602
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001603 if (!PageUptodate(page)) {
1604 if (copied < PAGE_CACHE_SIZE) {
1605 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1606 zero_user_segments(page, 0, from,
1607 from + copied, PAGE_CACHE_SIZE);
1608 }
1609 SetPageUptodate(page);
1610 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001611 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001612 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001613 page_cache_release(page);
1614
Nick Piggin800d15a2007-10-16 01:25:03 -07001615 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616}
1617
Al Viro2ba5bbe2014-04-02 20:00:02 -04001618static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
Al Viro6e58e792014-02-03 17:07:03 -05001620 struct file *file = iocb->ki_filp;
1621 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001623 pgoff_t index;
1624 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001625 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02001626 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05001627 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05001628 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001629
1630 /*
1631 * Might this read be for a stacking filesystem? Then when reading
1632 * holes of a sparse file, we actually need to allocate those pages,
1633 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1634 */
Al Viro777eda22014-12-17 04:46:46 -05001635 if (!iter_is_iovec(to))
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001636 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
1638 index = *ppos >> PAGE_CACHE_SHIFT;
1639 offset = *ppos & ~PAGE_CACHE_MASK;
1640
1641 for (;;) {
1642 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001643 pgoff_t end_index;
1644 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 loff_t i_size = i_size_read(inode);
1646
1647 end_index = i_size >> PAGE_CACHE_SHIFT;
1648 if (index > end_index)
1649 break;
1650 if (index == end_index) {
1651 nr = i_size & ~PAGE_CACHE_MASK;
1652 if (nr <= offset)
1653 break;
1654 }
1655
Al Viro6e58e792014-02-03 17:07:03 -05001656 error = shmem_getpage(inode, index, &page, sgp, NULL);
1657 if (error) {
1658 if (error == -EINVAL)
1659 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 break;
1661 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001662 if (page)
1663 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665 /*
1666 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001667 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 */
1669 nr = PAGE_CACHE_SIZE;
1670 i_size = i_size_read(inode);
1671 end_index = i_size >> PAGE_CACHE_SHIFT;
1672 if (index == end_index) {
1673 nr = i_size & ~PAGE_CACHE_MASK;
1674 if (nr <= offset) {
1675 if (page)
1676 page_cache_release(page);
1677 break;
1678 }
1679 }
1680 nr -= offset;
1681
1682 if (page) {
1683 /*
1684 * If users can be writing to this page using arbitrary
1685 * virtual addresses, take care about potential aliasing
1686 * before reading the page on the kernel side.
1687 */
1688 if (mapping_writably_mapped(mapping))
1689 flush_dcache_page(page);
1690 /*
1691 * Mark the page accessed if we read the beginning.
1692 */
1693 if (!offset)
1694 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001695 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001697 page_cache_get(page);
1698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
1700 /*
1701 * Ok, we have the page, and it's up-to-date, so
1702 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04001704 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05001705 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 offset += ret;
1707 index += offset >> PAGE_CACHE_SHIFT;
1708 offset &= ~PAGE_CACHE_MASK;
1709
1710 page_cache_release(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04001711 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 break;
Al Viro6e58e792014-02-03 17:07:03 -05001713 if (ret < nr) {
1714 error = -EFAULT;
1715 break;
1716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 cond_resched();
1718 }
1719
1720 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05001721 file_accessed(file);
1722 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723}
1724
Hugh Dickins708e3502011-07-25 17:12:32 -07001725static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1726 struct pipe_inode_info *pipe, size_t len,
1727 unsigned int flags)
1728{
1729 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001730 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001731 unsigned int loff, nr_pages, req_pages;
1732 struct page *pages[PIPE_DEF_BUFFERS];
1733 struct partial_page partial[PIPE_DEF_BUFFERS];
1734 struct page *page;
1735 pgoff_t index, end_index;
1736 loff_t isize, left;
1737 int error, page_nr;
1738 struct splice_pipe_desc spd = {
1739 .pages = pages,
1740 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001741 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001742 .flags = flags,
1743 .ops = &page_cache_pipe_buf_ops,
1744 .spd_release = spd_release_page,
1745 };
1746
Hugh Dickins71f0e072011-07-25 17:12:33 -07001747 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001748 if (unlikely(*ppos >= isize))
1749 return 0;
1750
1751 left = isize - *ppos;
1752 if (unlikely(left < len))
1753 len = left;
1754
1755 if (splice_grow_spd(pipe, &spd))
1756 return -ENOMEM;
1757
1758 index = *ppos >> PAGE_CACHE_SHIFT;
1759 loff = *ppos & ~PAGE_CACHE_MASK;
1760 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04001761 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07001762
Hugh Dickins708e3502011-07-25 17:12:32 -07001763 spd.nr_pages = find_get_pages_contig(mapping, index,
1764 nr_pages, spd.pages);
1765 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001766 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001767
Hugh Dickins708e3502011-07-25 17:12:32 -07001768 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001769 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1770 if (error)
1771 break;
1772 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001773 spd.pages[spd.nr_pages++] = page;
1774 index++;
1775 }
1776
Hugh Dickins708e3502011-07-25 17:12:32 -07001777 index = *ppos >> PAGE_CACHE_SHIFT;
1778 nr_pages = spd.nr_pages;
1779 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001780
Hugh Dickins708e3502011-07-25 17:12:32 -07001781 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1782 unsigned int this_len;
1783
1784 if (!len)
1785 break;
1786
Hugh Dickins708e3502011-07-25 17:12:32 -07001787 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1788 page = spd.pages[page_nr];
1789
Hugh Dickins71f0e072011-07-25 17:12:33 -07001790 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001791 error = shmem_getpage(inode, index, &page,
1792 SGP_CACHE, NULL);
1793 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001794 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001795 unlock_page(page);
1796 page_cache_release(spd.pages[page_nr]);
1797 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001798 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001799
1800 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001801 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1802 if (unlikely(!isize || index > end_index))
1803 break;
1804
Hugh Dickins708e3502011-07-25 17:12:32 -07001805 if (end_index == index) {
1806 unsigned int plen;
1807
Hugh Dickins708e3502011-07-25 17:12:32 -07001808 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1809 if (plen <= loff)
1810 break;
1811
Hugh Dickins708e3502011-07-25 17:12:32 -07001812 this_len = min(this_len, plen - loff);
1813 len = this_len;
1814 }
1815
1816 spd.partial[page_nr].offset = loff;
1817 spd.partial[page_nr].len = this_len;
1818 len -= this_len;
1819 loff = 0;
1820 spd.nr_pages++;
1821 index++;
1822 }
1823
Hugh Dickins708e3502011-07-25 17:12:32 -07001824 while (page_nr < nr_pages)
1825 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001826
1827 if (spd.nr_pages)
1828 error = splice_to_pipe(pipe, &spd);
1829
Eric Dumazet047fe362012-06-12 15:24:40 +02001830 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001831
1832 if (error > 0) {
1833 *ppos += error;
1834 file_accessed(in);
1835 }
1836 return error;
1837}
1838
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001839/*
1840 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1841 */
1842static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001843 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001844{
1845 struct page *page;
1846 struct pagevec pvec;
1847 pgoff_t indices[PAGEVEC_SIZE];
1848 bool done = false;
1849 int i;
1850
1851 pagevec_init(&pvec, 0);
1852 pvec.nr = 1; /* start small: we may be there already */
1853 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001854 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001855 pvec.nr, pvec.pages, indices);
1856 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001857 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001858 index = end;
1859 break;
1860 }
1861 for (i = 0; i < pvec.nr; i++, index++) {
1862 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001863 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001864 done = true;
1865 break;
1866 }
1867 index = indices[i];
1868 }
1869 page = pvec.pages[i];
1870 if (page && !radix_tree_exceptional_entry(page)) {
1871 if (!PageUptodate(page))
1872 page = NULL;
1873 }
1874 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001875 (page && whence == SEEK_DATA) ||
1876 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001877 done = true;
1878 break;
1879 }
1880 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001881 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001882 pagevec_release(&pvec);
1883 pvec.nr = PAGEVEC_SIZE;
1884 cond_resched();
1885 }
1886 return index;
1887}
1888
Andrew Morton965c8e52012-12-17 15:59:39 -08001889static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001890{
1891 struct address_space *mapping = file->f_mapping;
1892 struct inode *inode = mapping->host;
1893 pgoff_t start, end;
1894 loff_t new_offset;
1895
Andrew Morton965c8e52012-12-17 15:59:39 -08001896 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1897 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001898 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05001899 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001900 /* We're holding i_mutex so we can access i_size directly */
1901
1902 if (offset < 0)
1903 offset = -EINVAL;
1904 else if (offset >= inode->i_size)
1905 offset = -ENXIO;
1906 else {
1907 start = offset >> PAGE_CACHE_SHIFT;
1908 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001909 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001910 new_offset <<= PAGE_CACHE_SHIFT;
1911 if (new_offset > offset) {
1912 if (new_offset < inode->i_size)
1913 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001914 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001915 offset = -ENXIO;
1916 else
1917 offset = inode->i_size;
1918 }
1919 }
1920
Hugh Dickins387aae62013-08-04 11:30:25 -07001921 if (offset >= 0)
1922 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05001923 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001924 return offset;
1925}
1926
David Herrmann05f65b52014-08-08 14:25:36 -07001927/*
1928 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1929 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1930 */
1931#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1932#define LAST_SCAN 4 /* about 150ms max */
1933
1934static void shmem_tag_pins(struct address_space *mapping)
1935{
1936 struct radix_tree_iter iter;
1937 void **slot;
1938 pgoff_t start;
1939 struct page *page;
1940
1941 lru_add_drain();
1942 start = 0;
1943 rcu_read_lock();
1944
David Herrmann05f65b52014-08-08 14:25:36 -07001945 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1946 page = radix_tree_deref_slot(slot);
1947 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001948 if (radix_tree_deref_retry(page)) {
1949 slot = radix_tree_iter_retry(&iter);
1950 continue;
1951 }
David Herrmann05f65b52014-08-08 14:25:36 -07001952 } else if (page_count(page) - page_mapcount(page) > 1) {
1953 spin_lock_irq(&mapping->tree_lock);
1954 radix_tree_tag_set(&mapping->page_tree, iter.index,
1955 SHMEM_TAG_PINNED);
1956 spin_unlock_irq(&mapping->tree_lock);
1957 }
1958
1959 if (need_resched()) {
1960 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07001961 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07001962 }
1963 }
1964 rcu_read_unlock();
1965}
1966
1967/*
1968 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1969 * via get_user_pages(), drivers might have some pending I/O without any active
1970 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1971 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1972 * them to be dropped.
1973 * The caller must guarantee that no new user will acquire writable references
1974 * to those pages to avoid races.
1975 */
David Herrmann40e041a2014-08-08 14:25:27 -07001976static int shmem_wait_for_pins(struct address_space *mapping)
1977{
David Herrmann05f65b52014-08-08 14:25:36 -07001978 struct radix_tree_iter iter;
1979 void **slot;
1980 pgoff_t start;
1981 struct page *page;
1982 int error, scan;
1983
1984 shmem_tag_pins(mapping);
1985
1986 error = 0;
1987 for (scan = 0; scan <= LAST_SCAN; scan++) {
1988 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1989 break;
1990
1991 if (!scan)
1992 lru_add_drain_all();
1993 else if (schedule_timeout_killable((HZ << scan) / 200))
1994 scan = LAST_SCAN;
1995
1996 start = 0;
1997 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07001998 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
1999 start, SHMEM_TAG_PINNED) {
2000
2001 page = radix_tree_deref_slot(slot);
2002 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002003 if (radix_tree_deref_retry(page)) {
2004 slot = radix_tree_iter_retry(&iter);
2005 continue;
2006 }
David Herrmann05f65b52014-08-08 14:25:36 -07002007
2008 page = NULL;
2009 }
2010
2011 if (page &&
2012 page_count(page) - page_mapcount(page) != 1) {
2013 if (scan < LAST_SCAN)
2014 goto continue_resched;
2015
2016 /*
2017 * On the last scan, we clean up all those tags
2018 * we inserted; but make a note that we still
2019 * found pages pinned.
2020 */
2021 error = -EBUSY;
2022 }
2023
2024 spin_lock_irq(&mapping->tree_lock);
2025 radix_tree_tag_clear(&mapping->page_tree,
2026 iter.index, SHMEM_TAG_PINNED);
2027 spin_unlock_irq(&mapping->tree_lock);
2028continue_resched:
2029 if (need_resched()) {
2030 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002031 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002032 }
2033 }
2034 rcu_read_unlock();
2035 }
2036
2037 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002038}
2039
2040#define F_ALL_SEALS (F_SEAL_SEAL | \
2041 F_SEAL_SHRINK | \
2042 F_SEAL_GROW | \
2043 F_SEAL_WRITE)
2044
2045int shmem_add_seals(struct file *file, unsigned int seals)
2046{
2047 struct inode *inode = file_inode(file);
2048 struct shmem_inode_info *info = SHMEM_I(inode);
2049 int error;
2050
2051 /*
2052 * SEALING
2053 * Sealing allows multiple parties to share a shmem-file but restrict
2054 * access to a specific subset of file operations. Seals can only be
2055 * added, but never removed. This way, mutually untrusted parties can
2056 * share common memory regions with a well-defined policy. A malicious
2057 * peer can thus never perform unwanted operations on a shared object.
2058 *
2059 * Seals are only supported on special shmem-files and always affect
2060 * the whole underlying inode. Once a seal is set, it may prevent some
2061 * kinds of access to the file. Currently, the following seals are
2062 * defined:
2063 * SEAL_SEAL: Prevent further seals from being set on this file
2064 * SEAL_SHRINK: Prevent the file from shrinking
2065 * SEAL_GROW: Prevent the file from growing
2066 * SEAL_WRITE: Prevent write access to the file
2067 *
2068 * As we don't require any trust relationship between two parties, we
2069 * must prevent seals from being removed. Therefore, sealing a file
2070 * only adds a given set of seals to the file, it never touches
2071 * existing seals. Furthermore, the "setting seals"-operation can be
2072 * sealed itself, which basically prevents any further seal from being
2073 * added.
2074 *
2075 * Semantics of sealing are only defined on volatile files. Only
2076 * anonymous shmem files support sealing. More importantly, seals are
2077 * never written to disk. Therefore, there's no plan to support it on
2078 * other file types.
2079 */
2080
2081 if (file->f_op != &shmem_file_operations)
2082 return -EINVAL;
2083 if (!(file->f_mode & FMODE_WRITE))
2084 return -EPERM;
2085 if (seals & ~(unsigned int)F_ALL_SEALS)
2086 return -EINVAL;
2087
Al Viro59551022016-01-22 15:40:57 -05002088 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002089
2090 if (info->seals & F_SEAL_SEAL) {
2091 error = -EPERM;
2092 goto unlock;
2093 }
2094
2095 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2096 error = mapping_deny_writable(file->f_mapping);
2097 if (error)
2098 goto unlock;
2099
2100 error = shmem_wait_for_pins(file->f_mapping);
2101 if (error) {
2102 mapping_allow_writable(file->f_mapping);
2103 goto unlock;
2104 }
2105 }
2106
2107 info->seals |= seals;
2108 error = 0;
2109
2110unlock:
Al Viro59551022016-01-22 15:40:57 -05002111 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002112 return error;
2113}
2114EXPORT_SYMBOL_GPL(shmem_add_seals);
2115
2116int shmem_get_seals(struct file *file)
2117{
2118 if (file->f_op != &shmem_file_operations)
2119 return -EINVAL;
2120
2121 return SHMEM_I(file_inode(file))->seals;
2122}
2123EXPORT_SYMBOL_GPL(shmem_get_seals);
2124
2125long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2126{
2127 long error;
2128
2129 switch (cmd) {
2130 case F_ADD_SEALS:
2131 /* disallow upper 32bit */
2132 if (arg > UINT_MAX)
2133 return -EINVAL;
2134
2135 error = shmem_add_seals(file, arg);
2136 break;
2137 case F_GET_SEALS:
2138 error = shmem_get_seals(file);
2139 break;
2140 default:
2141 error = -EINVAL;
2142 break;
2143 }
2144
2145 return error;
2146}
2147
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002148static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2149 loff_t len)
2150{
Al Viro496ad9a2013-01-23 17:07:38 -05002151 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002152 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002153 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002154 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002155 pgoff_t start, index, end;
2156 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002157
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002158 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2159 return -EOPNOTSUPP;
2160
Al Viro59551022016-01-22 15:40:57 -05002161 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002162
2163 if (mode & FALLOC_FL_PUNCH_HOLE) {
2164 struct address_space *mapping = file->f_mapping;
2165 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2166 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002167 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002168
David Herrmann40e041a2014-08-08 14:25:27 -07002169 /* protected by i_mutex */
2170 if (info->seals & F_SEAL_WRITE) {
2171 error = -EPERM;
2172 goto out;
2173 }
2174
Hugh Dickins8e205f72014-07-23 14:00:10 -07002175 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002176 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2177 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2178 spin_lock(&inode->i_lock);
2179 inode->i_private = &shmem_falloc;
2180 spin_unlock(&inode->i_lock);
2181
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002182 if ((u64)unmap_end > (u64)unmap_start)
2183 unmap_mapping_range(mapping, unmap_start,
2184 1 + unmap_end - unmap_start, 0);
2185 shmem_truncate_range(inode, offset, offset + len - 1);
2186 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002187
2188 spin_lock(&inode->i_lock);
2189 inode->i_private = NULL;
2190 wake_up_all(&shmem_falloc_waitq);
2191 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002192 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002193 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002194 }
2195
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002196 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2197 error = inode_newsize_ok(inode, offset + len);
2198 if (error)
2199 goto out;
2200
David Herrmann40e041a2014-08-08 14:25:27 -07002201 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2202 error = -EPERM;
2203 goto out;
2204 }
2205
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002206 start = offset >> PAGE_CACHE_SHIFT;
2207 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2208 /* Try to avoid a swapstorm if len is impossible to satisfy */
2209 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2210 error = -ENOSPC;
2211 goto out;
2212 }
2213
Hugh Dickins8e205f72014-07-23 14:00:10 -07002214 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002215 shmem_falloc.start = start;
2216 shmem_falloc.next = start;
2217 shmem_falloc.nr_falloced = 0;
2218 shmem_falloc.nr_unswapped = 0;
2219 spin_lock(&inode->i_lock);
2220 inode->i_private = &shmem_falloc;
2221 spin_unlock(&inode->i_lock);
2222
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002223 for (index = start; index < end; index++) {
2224 struct page *page;
2225
2226 /*
2227 * Good, the fallocate(2) manpage permits EINTR: we may have
2228 * been interrupted because we are using up too much memory.
2229 */
2230 if (signal_pending(current))
2231 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002232 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2233 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002234 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002235 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002236 NULL);
2237 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002238 /* Remove the !PageUptodate pages we added */
2239 shmem_undo_range(inode,
2240 (loff_t)start << PAGE_CACHE_SHIFT,
2241 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002242 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002243 }
2244
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002245 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002246 * Inform shmem_writepage() how far we have reached.
2247 * No need for lock or barrier: we have the page lock.
2248 */
2249 shmem_falloc.next++;
2250 if (!PageUptodate(page))
2251 shmem_falloc.nr_falloced++;
2252
2253 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002254 * If !PageUptodate, leave it that way so that freeable pages
2255 * can be recognized if we need to rollback on error later.
2256 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002257 * than free the pages we are allocating (and SGP_CACHE pages
2258 * might still be clean: we now need to mark those dirty too).
2259 */
2260 set_page_dirty(page);
2261 unlock_page(page);
2262 page_cache_release(page);
2263 cond_resched();
2264 }
2265
2266 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2267 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002268 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002269undone:
2270 spin_lock(&inode->i_lock);
2271 inode->i_private = NULL;
2272 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002273out:
Al Viro59551022016-01-22 15:40:57 -05002274 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002275 return error;
2276}
2277
David Howells726c3342006-06-23 02:02:58 -07002278static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279{
David Howells726c3342006-06-23 02:02:58 -07002280 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
2282 buf->f_type = TMPFS_MAGIC;
2283 buf->f_bsize = PAGE_CACHE_SIZE;
2284 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002285 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002287 buf->f_bavail =
2288 buf->f_bfree = sbinfo->max_blocks -
2289 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002290 }
2291 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 buf->f_files = sbinfo->max_inodes;
2293 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 }
2295 /* else leave those fields 0 like simple_statfs */
2296 return 0;
2297}
2298
2299/*
2300 * File creation. Allocate an inode, and we're done..
2301 */
2302static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002303shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002305 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 int error = -ENOSPC;
2307
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002308 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002310 error = simple_acl_create(dir, inode);
2311 if (error)
2312 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002313 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002314 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002315 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002316 if (error && error != -EOPNOTSUPP)
2317 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002318
Al Viro718deb62009-12-16 19:35:36 -05002319 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 dir->i_size += BOGO_DIRENT_SIZE;
2321 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2322 d_instantiate(dentry, inode);
2323 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 }
2325 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002326out_iput:
2327 iput(inode);
2328 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329}
2330
Al Viro60545d02013-06-07 01:20:27 -04002331static int
2332shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2333{
2334 struct inode *inode;
2335 int error = -ENOSPC;
2336
2337 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2338 if (inode) {
2339 error = security_inode_init_security(inode, dir,
2340 NULL,
2341 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002342 if (error && error != -EOPNOTSUPP)
2343 goto out_iput;
2344 error = simple_acl_create(dir, inode);
2345 if (error)
2346 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002347 d_tmpfile(dentry, inode);
2348 }
2349 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002350out_iput:
2351 iput(inode);
2352 return error;
Al Viro60545d02013-06-07 01:20:27 -04002353}
2354
Al Viro18bb1db2011-07-26 01:41:39 -04002355static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356{
2357 int error;
2358
2359 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2360 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002361 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 return 0;
2363}
2364
Al Viro4acdaf22011-07-26 01:42:34 -04002365static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002366 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367{
2368 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2369}
2370
2371/*
2372 * Link a file..
2373 */
2374static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2375{
David Howells75c3cfa2015-03-17 22:26:12 +00002376 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002377 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
2379 /*
2380 * No ordinary (disk based) filesystem counts links as inodes;
2381 * but each new link needs a new dentry, pinning lowmem, and
2382 * tmpfs dentries cannot be pruned until they are unlinked.
2383 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002384 ret = shmem_reserve_inode(inode->i_sb);
2385 if (ret)
2386 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
2388 dir->i_size += BOGO_DIRENT_SIZE;
2389 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002390 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002391 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 dget(dentry); /* Extra pinning count for the created dentry */
2393 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002394out:
2395 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396}
2397
2398static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2399{
David Howells75c3cfa2015-03-17 22:26:12 +00002400 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002402 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2403 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
2405 dir->i_size -= BOGO_DIRENT_SIZE;
2406 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002407 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 dput(dentry); /* Undo the count from "create" - this does all the work */
2409 return 0;
2410}
2411
2412static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2413{
2414 if (!simple_empty(dentry))
2415 return -ENOTEMPTY;
2416
David Howells75c3cfa2015-03-17 22:26:12 +00002417 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002418 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 return shmem_unlink(dir, dentry);
2420}
2421
Miklos Szeredi37456772014-07-23 15:15:34 +02002422static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2423{
David Howellse36cb0b2015-01-29 12:02:35 +00002424 bool old_is_dir = d_is_dir(old_dentry);
2425 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002426
2427 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2428 if (old_is_dir) {
2429 drop_nlink(old_dir);
2430 inc_nlink(new_dir);
2431 } else {
2432 drop_nlink(new_dir);
2433 inc_nlink(old_dir);
2434 }
2435 }
2436 old_dir->i_ctime = old_dir->i_mtime =
2437 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002438 d_inode(old_dentry)->i_ctime =
2439 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
Miklos Szeredi37456772014-07-23 15:15:34 +02002440
2441 return 0;
2442}
2443
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002444static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2445{
2446 struct dentry *whiteout;
2447 int error;
2448
2449 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2450 if (!whiteout)
2451 return -ENOMEM;
2452
2453 error = shmem_mknod(old_dir, whiteout,
2454 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2455 dput(whiteout);
2456 if (error)
2457 return error;
2458
2459 /*
2460 * Cheat and hash the whiteout while the old dentry is still in
2461 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2462 *
2463 * d_lookup() will consistently find one of them at this point,
2464 * not sure which one, but that isn't even important.
2465 */
2466 d_rehash(whiteout);
2467 return 0;
2468}
2469
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470/*
2471 * The VFS layer already does all the dentry stuff for rename,
2472 * we just have to decrement the usage count for the target if
2473 * it exists so that the VFS layer correctly free's it when it
2474 * gets overwritten.
2475 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002476static 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 -07002477{
David Howells75c3cfa2015-03-17 22:26:12 +00002478 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 int they_are_dirs = S_ISDIR(inode->i_mode);
2480
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002481 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002482 return -EINVAL;
2483
Miklos Szeredi37456772014-07-23 15:15:34 +02002484 if (flags & RENAME_EXCHANGE)
2485 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2486
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 if (!simple_empty(new_dentry))
2488 return -ENOTEMPTY;
2489
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002490 if (flags & RENAME_WHITEOUT) {
2491 int error;
2492
2493 error = shmem_whiteout(old_dir, old_dentry);
2494 if (error)
2495 return error;
2496 }
2497
David Howells75c3cfa2015-03-17 22:26:12 +00002498 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002500 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002501 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002502 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002505 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002506 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 }
2508
2509 old_dir->i_size -= BOGO_DIRENT_SIZE;
2510 new_dir->i_size += BOGO_DIRENT_SIZE;
2511 old_dir->i_ctime = old_dir->i_mtime =
2512 new_dir->i_ctime = new_dir->i_mtime =
2513 inode->i_ctime = CURRENT_TIME;
2514 return 0;
2515}
2516
2517static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2518{
2519 int error;
2520 int len;
2521 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002522 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 struct shmem_inode_info *info;
2524
2525 len = strlen(symname) + 1;
2526 if (len > PAGE_CACHE_SIZE)
2527 return -ENAMETOOLONG;
2528
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002529 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (!inode)
2531 return -ENOSPC;
2532
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002533 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002534 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002535 if (error) {
2536 if (error != -EOPNOTSUPP) {
2537 iput(inode);
2538 return error;
2539 }
2540 error = 0;
2541 }
2542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 info = SHMEM_I(inode);
2544 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002545 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05002546 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
2547 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002548 iput(inode);
2549 return -ENOMEM;
2550 }
2551 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05002553 inode_nohighmem(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2555 if (error) {
2556 iput(inode);
2557 return error;
2558 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002559 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05002561 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002562 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002564 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 page_cache_release(page);
2566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 dir->i_size += BOGO_DIRENT_SIZE;
2568 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2569 d_instantiate(dentry, inode);
2570 dget(dentry);
2571 return 0;
2572}
2573
Al Virofceef392015-12-29 15:58:39 -05002574static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575{
Al Virofceef392015-12-29 15:58:39 -05002576 mark_page_accessed(arg);
2577 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578}
2579
Al Viro6b255392015-11-17 10:20:54 -05002580static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05002581 struct inode *inode,
2582 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05002585 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05002586 if (!dentry) {
2587 page = find_get_page(inode->i_mapping, 0);
2588 if (!page)
2589 return ERR_PTR(-ECHILD);
2590 if (!PageUptodate(page)) {
2591 put_page(page);
2592 return ERR_PTR(-ECHILD);
2593 }
2594 } else {
2595 error = shmem_getpage(inode, 0, &page, SGP_READ, NULL);
2596 if (error)
2597 return ERR_PTR(error);
2598 unlock_page(page);
2599 }
Al Virofceef392015-12-29 15:58:39 -05002600 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05002601 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602}
2603
Eric Parisb09e0fa2011-05-24 17:12:39 -07002604#ifdef CONFIG_TMPFS_XATTR
2605/*
2606 * Superblocks without xattr inode operations may get some security.* xattr
2607 * support from the LSM "for free". As soon as we have any other xattrs
2608 * like ACLs, we also need to implement the security.* handlers at
2609 * filesystem level, though.
2610 */
2611
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002612/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002613 * Callback for security_inode_init_security() for acquiring xattrs.
2614 */
2615static int shmem_initxattrs(struct inode *inode,
2616 const struct xattr *xattr_array,
2617 void *fs_info)
2618{
2619 struct shmem_inode_info *info = SHMEM_I(inode);
2620 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002621 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002622 size_t len;
2623
2624 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002625 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002626 if (!new_xattr)
2627 return -ENOMEM;
2628
2629 len = strlen(xattr->name) + 1;
2630 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2631 GFP_KERNEL);
2632 if (!new_xattr->name) {
2633 kfree(new_xattr);
2634 return -ENOMEM;
2635 }
2636
2637 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2638 XATTR_SECURITY_PREFIX_LEN);
2639 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2640 xattr->name, len);
2641
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002642 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002643 }
2644
2645 return 0;
2646}
2647
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002648static int shmem_xattr_handler_get(const struct xattr_handler *handler,
2649 struct dentry *dentry, const char *name,
2650 void *buffer, size_t size)
2651{
2652 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2653
2654 name = xattr_full_name(handler, name);
2655 return simple_xattr_get(&info->xattrs, name, buffer, size);
2656}
2657
2658static int shmem_xattr_handler_set(const struct xattr_handler *handler,
2659 struct dentry *dentry, const char *name,
2660 const void *value, size_t size, int flags)
2661{
2662 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2663
2664 name = xattr_full_name(handler, name);
2665 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2666}
2667
2668static const struct xattr_handler shmem_security_xattr_handler = {
2669 .prefix = XATTR_SECURITY_PREFIX,
2670 .get = shmem_xattr_handler_get,
2671 .set = shmem_xattr_handler_set,
2672};
2673
2674static const struct xattr_handler shmem_trusted_xattr_handler = {
2675 .prefix = XATTR_TRUSTED_PREFIX,
2676 .get = shmem_xattr_handler_get,
2677 .set = shmem_xattr_handler_set,
2678};
2679
Eric Parisb09e0fa2011-05-24 17:12:39 -07002680static const struct xattr_handler *shmem_xattr_handlers[] = {
2681#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002682 &posix_acl_access_xattr_handler,
2683 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002684#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002685 &shmem_security_xattr_handler,
2686 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002687 NULL
2688};
2689
Eric Parisb09e0fa2011-05-24 17:12:39 -07002690static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2691{
David Howells75c3cfa2015-03-17 22:26:12 +00002692 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01002693 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002694}
2695#endif /* CONFIG_TMPFS_XATTR */
2696
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002697static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05002699 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002700#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002701 .setxattr = generic_setxattr,
2702 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002703 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002704 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002705#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706};
2707
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002708static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05002710 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002711#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002712 .setxattr = generic_setxattr,
2713 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002714 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002715 .removexattr = generic_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002716#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002717};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002718
David M. Grimes91828a42006-10-17 00:09:45 -07002719static struct dentry *shmem_get_parent(struct dentry *child)
2720{
2721 return ERR_PTR(-ESTALE);
2722}
2723
2724static int shmem_match(struct inode *ino, void *vfh)
2725{
2726 __u32 *fh = vfh;
2727 __u64 inum = fh[2];
2728 inum = (inum << 32) | fh[1];
2729 return ino->i_ino == inum && fh[0] == ino->i_generation;
2730}
2731
Christoph Hellwig480b1162007-10-21 16:42:13 -07002732static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2733 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002734{
David M. Grimes91828a42006-10-17 00:09:45 -07002735 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002736 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002737 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002738
Christoph Hellwig480b1162007-10-21 16:42:13 -07002739 if (fh_len < 3)
2740 return NULL;
2741
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002742 inum = fid->raw[2];
2743 inum = (inum << 32) | fid->raw[1];
2744
Christoph Hellwig480b1162007-10-21 16:42:13 -07002745 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2746 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002747 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002748 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002749 iput(inode);
2750 }
2751
Christoph Hellwig480b1162007-10-21 16:42:13 -07002752 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002753}
2754
Al Virob0b03822012-04-02 14:34:06 -04002755static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2756 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002757{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302758 if (*len < 3) {
2759 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002760 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302761 }
David M. Grimes91828a42006-10-17 00:09:45 -07002762
Al Viro1d3382cb2010-10-23 15:19:20 -04002763 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002764 /* Unfortunately insert_inode_hash is not idempotent,
2765 * so as we hash inodes here rather than at creation
2766 * time, we need a lock to ensure we only try
2767 * to do it once
2768 */
2769 static DEFINE_SPINLOCK(lock);
2770 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002771 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002772 __insert_inode_hash(inode,
2773 inode->i_ino + inode->i_generation);
2774 spin_unlock(&lock);
2775 }
2776
2777 fh[0] = inode->i_generation;
2778 fh[1] = inode->i_ino;
2779 fh[2] = ((__u64)inode->i_ino) >> 32;
2780
2781 *len = 3;
2782 return 1;
2783}
2784
Christoph Hellwig39655162007-10-21 16:42:17 -07002785static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002786 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002787 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002788 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002789};
2790
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002791static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2792 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793{
2794 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002795 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002796 uid_t uid;
2797 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002799 while (options != NULL) {
2800 this_char = options;
2801 for (;;) {
2802 /*
2803 * NUL-terminate this option: unfortunately,
2804 * mount options form a comma-separated list,
2805 * but mpol's nodelist may also contain commas.
2806 */
2807 options = strchr(options, ',');
2808 if (options == NULL)
2809 break;
2810 options++;
2811 if (!isdigit(*options)) {
2812 options[-1] = '\0';
2813 break;
2814 }
2815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (!*this_char)
2817 continue;
2818 if ((value = strchr(this_char,'=')) != NULL) {
2819 *value++ = 0;
2820 } else {
Joe Perches11705322016-03-17 14:19:50 -07002821 pr_err("tmpfs: No value for mount option '%s'\n",
2822 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002823 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 }
2825
2826 if (!strcmp(this_char,"size")) {
2827 unsigned long long size;
2828 size = memparse(value,&rest);
2829 if (*rest == '%') {
2830 size <<= PAGE_SHIFT;
2831 size *= totalram_pages;
2832 do_div(size, 100);
2833 rest++;
2834 }
2835 if (*rest)
2836 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002837 sbinfo->max_blocks =
2838 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002840 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 if (*rest)
2842 goto bad_val;
2843 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002844 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 if (*rest)
2846 goto bad_val;
2847 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002848 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002850 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 if (*rest)
2852 goto bad_val;
2853 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002854 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002856 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 if (*rest)
2858 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002859 sbinfo->uid = make_kuid(current_user_ns(), uid);
2860 if (!uid_valid(sbinfo->uid))
2861 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002863 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002865 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 if (*rest)
2867 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002868 sbinfo->gid = make_kgid(current_user_ns(), gid);
2869 if (!gid_valid(sbinfo->gid))
2870 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002871 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002872 mpol_put(mpol);
2873 mpol = NULL;
2874 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002875 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 } else {
Joe Perches11705322016-03-17 14:19:50 -07002877 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002878 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 }
2880 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002881 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 return 0;
2883
2884bad_val:
Joe Perches11705322016-03-17 14:19:50 -07002885 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002887error:
2888 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 return 1;
2890
2891}
2892
2893static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2894{
2895 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002896 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002897 unsigned long inodes;
2898 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899
Greg Thelen5f001102013-02-22 16:36:01 -08002900 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002901 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002902 return error;
2903
2904 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002905 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002906 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002907 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002908 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002909 goto out;
2910 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002911 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002912 * but we must separately disallow unlimited->limited, because
2913 * in that case we have no record of how much is already in use.
2914 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002915 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002916 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002917 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002918 goto out;
2919
2920 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002921 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002922 sbinfo->max_inodes = config.max_inodes;
2923 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002924
Greg Thelen5f001102013-02-22 16:36:01 -08002925 /*
2926 * Preserve previous mempolicy unless mpol remount option was specified.
2927 */
2928 if (config.mpol) {
2929 mpol_put(sbinfo->mpol);
2930 sbinfo->mpol = config.mpol; /* transfers initial ref */
2931 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002932out:
2933 spin_unlock(&sbinfo->stat_lock);
2934 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002936
Al Viro34c80b12011-12-08 21:32:45 -05002937static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002938{
Al Viro34c80b12011-12-08 21:32:45 -05002939 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002940
2941 if (sbinfo->max_blocks != shmem_default_max_blocks())
2942 seq_printf(seq, ",size=%luk",
2943 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2944 if (sbinfo->max_inodes != shmem_default_max_inodes())
2945 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2946 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002947 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002948 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2949 seq_printf(seq, ",uid=%u",
2950 from_kuid_munged(&init_user_ns, sbinfo->uid));
2951 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2952 seq_printf(seq, ",gid=%u",
2953 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002954 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002955 return 0;
2956}
David Herrmann9183df22014-08-08 14:25:29 -07002957
2958#define MFD_NAME_PREFIX "memfd:"
2959#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2960#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2961
2962#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2963
2964SYSCALL_DEFINE2(memfd_create,
2965 const char __user *, uname,
2966 unsigned int, flags)
2967{
2968 struct shmem_inode_info *info;
2969 struct file *file;
2970 int fd, error;
2971 char *name;
2972 long len;
2973
2974 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2975 return -EINVAL;
2976
2977 /* length includes terminating zero */
2978 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2979 if (len <= 0)
2980 return -EFAULT;
2981 if (len > MFD_NAME_MAX_LEN + 1)
2982 return -EINVAL;
2983
2984 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2985 if (!name)
2986 return -ENOMEM;
2987
2988 strcpy(name, MFD_NAME_PREFIX);
2989 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2990 error = -EFAULT;
2991 goto err_name;
2992 }
2993
2994 /* terminating-zero may have changed after strnlen_user() returned */
2995 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
2996 error = -EFAULT;
2997 goto err_name;
2998 }
2999
3000 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3001 if (fd < 0) {
3002 error = fd;
3003 goto err_name;
3004 }
3005
3006 file = shmem_file_setup(name, 0, VM_NORESERVE);
3007 if (IS_ERR(file)) {
3008 error = PTR_ERR(file);
3009 goto err_fd;
3010 }
3011 info = SHMEM_I(file_inode(file));
3012 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3013 file->f_flags |= O_RDWR | O_LARGEFILE;
3014 if (flags & MFD_ALLOW_SEALING)
3015 info->seals &= ~F_SEAL_SEAL;
3016
3017 fd_install(fd, file);
3018 kfree(name);
3019 return fd;
3020
3021err_fd:
3022 put_unused_fd(fd);
3023err_name:
3024 kfree(name);
3025 return error;
3026}
3027
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003028#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029
3030static void shmem_put_super(struct super_block *sb)
3031{
Hugh Dickins602586a2010-08-17 15:23:56 -07003032 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3033
3034 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003035 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003036 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 sb->s_fs_info = NULL;
3038}
3039
Kay Sievers2b2af542009-04-30 15:23:42 +02003040int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041{
3042 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003043 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003044 int err = -ENOMEM;
3045
3046 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003047 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003048 L1_CACHE_BYTES), GFP_KERNEL);
3049 if (!sbinfo)
3050 return -ENOMEM;
3051
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003052 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003053 sbinfo->uid = current_fsuid();
3054 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003055 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003057#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 /*
3059 * Per default we only allow half of the physical ram per
3060 * tmpfs instance, limiting inodes to one per page of lowmem;
3061 * but the internal instance is left unlimited.
3062 */
Al Viroca4e0512013-08-31 12:57:10 -04003063 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003064 sbinfo->max_blocks = shmem_default_max_blocks();
3065 sbinfo->max_inodes = shmem_default_max_inodes();
3066 if (shmem_parse_options(data, sbinfo, false)) {
3067 err = -EINVAL;
3068 goto failed;
3069 }
Al Viroca4e0512013-08-31 12:57:10 -04003070 } else {
3071 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 }
David M. Grimes91828a42006-10-17 00:09:45 -07003073 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003074 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075#else
3076 sb->s_flags |= MS_NOUSER;
3077#endif
3078
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003079 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003080 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003081 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003082 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003083
Hugh Dickins285b2c42011-08-03 16:21:20 -07003084 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 sb->s_blocksize = PAGE_CACHE_SIZE;
3086 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
3087 sb->s_magic = TMPFS_MAGIC;
3088 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003089 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003090#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003091 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003092#endif
3093#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003094 sb->s_flags |= MS_POSIXACL;
3095#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003096
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003097 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 if (!inode)
3099 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003100 inode->i_uid = sbinfo->uid;
3101 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003102 sb->s_root = d_make_root(inode);
3103 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003104 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 return 0;
3106
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107failed:
3108 shmem_put_super(sb);
3109 return err;
3110}
3111
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003112static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113
3114static struct inode *shmem_alloc_inode(struct super_block *sb)
3115{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003116 struct shmem_inode_info *info;
3117 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3118 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003120 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121}
3122
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003123static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003124{
3125 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro3ed47db2016-01-22 18:08:52 -05003126 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003127 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3128}
3129
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130static void shmem_destroy_inode(struct inode *inode)
3131{
Al Viro09208d12011-07-26 03:15:03 -04003132 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003134 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135}
3136
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003137static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003139 struct shmem_inode_info *info = foo;
3140 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141}
3142
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003143static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144{
3145 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3146 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003147 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 return 0;
3149}
3150
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003151static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003153 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154}
3155
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003156static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003158 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003160 .write_begin = shmem_write_begin,
3161 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003163#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003164 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003165#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003166 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167};
3168
Helge Deller15ad7cd2006-12-06 20:40:36 -08003169static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 .mmap = shmem_mmap,
3171#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003172 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003173 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003174 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003175 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003176 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003177 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003178 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179#endif
3180};
3181
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003182static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003183 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003184 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003185#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003186 .setxattr = generic_setxattr,
3187 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003188 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003189 .removexattr = generic_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003190 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003191#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192};
3193
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003194static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195#ifdef CONFIG_TMPFS
3196 .create = shmem_create,
3197 .lookup = simple_lookup,
3198 .link = shmem_link,
3199 .unlink = shmem_unlink,
3200 .symlink = shmem_symlink,
3201 .mkdir = shmem_mkdir,
3202 .rmdir = shmem_rmdir,
3203 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003204 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003205 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003207#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003208 .setxattr = generic_setxattr,
3209 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003210 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003211 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003212#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003213#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003214 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003215 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003216#endif
3217};
3218
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003219static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003220#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003221 .setxattr = generic_setxattr,
3222 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003223 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003224 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003225#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003226#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003227 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003228 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003229#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230};
3231
Hugh Dickins759b9772007-03-05 00:30:28 -08003232static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 .alloc_inode = shmem_alloc_inode,
3234 .destroy_inode = shmem_destroy_inode,
3235#ifdef CONFIG_TMPFS
3236 .statfs = shmem_statfs,
3237 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003238 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239#endif
Al Viro1f895f72010-06-05 19:10:41 -04003240 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 .drop_inode = generic_delete_inode,
3242 .put_super = shmem_put_super,
3243};
3244
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003245static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003246 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003247 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248#ifdef CONFIG_NUMA
3249 .set_policy = shmem_set_policy,
3250 .get_policy = shmem_get_policy,
3251#endif
3252};
3253
Al Viro3c26ff62010-07-25 11:46:36 +04003254static struct dentry *shmem_mount(struct file_system_type *fs_type,
3255 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256{
Al Viro3c26ff62010-07-25 11:46:36 +04003257 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258}
3259
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003260static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 .owner = THIS_MODULE,
3262 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003263 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003265 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003268int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269{
3270 int error;
3271
Rob Landley16203a72013-09-11 14:26:12 -07003272 /* If rootfs called this, don't re-init */
3273 if (shmem_inode_cachep)
3274 return 0;
3275
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003276 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 if (error)
3278 goto out3;
3279
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003280 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003282 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 goto out2;
3284 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003285
Al Viroca4e0512013-08-31 12:57:10 -04003286 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 if (IS_ERR(shm_mnt)) {
3288 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003289 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 goto out1;
3291 }
3292 return 0;
3293
3294out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003295 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003297 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298out3:
3299 shm_mnt = ERR_PTR(error);
3300 return error;
3301}
Matt Mackall853ac432009-01-06 14:40:20 -08003302
3303#else /* !CONFIG_SHMEM */
3304
3305/*
3306 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3307 *
3308 * This is intended for small system where the benefits of the full
3309 * shmem code (swap-backed and resource-limited) are outweighed by
3310 * their complexity. On systems without swap this code should be
3311 * effectively equivalent, but much lighter weight.
3312 */
3313
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003314static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003315 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003316 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003317 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003318 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003319};
3320
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003321int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003322{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003323 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003324
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003325 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003326 BUG_ON(IS_ERR(shm_mnt));
3327
3328 return 0;
3329}
3330
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003331int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003332{
3333 return 0;
3334}
3335
Hugh Dickins3f96b792009-09-21 17:03:37 -07003336int shmem_lock(struct file *file, int lock, struct user_struct *user)
3337{
3338 return 0;
3339}
3340
Hugh Dickins24513262012-01-20 14:34:21 -08003341void shmem_unlock_mapping(struct address_space *mapping)
3342{
3343}
3344
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003345void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003346{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003347 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003348}
3349EXPORT_SYMBOL_GPL(shmem_truncate_range);
3350
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003351#define shmem_vm_ops generic_file_vm_ops
3352#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003353#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003354#define shmem_acct_size(flags, size) 0
3355#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003356
3357#endif /* CONFIG_SHMEM */
3358
3359/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360
Al Viro34515382013-02-14 22:38:02 -05003361static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003362 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003363};
3364
Eric Parisc7277092013-12-02 11:24:19 +00003365static struct file *__shmem_file_setup(const char *name, loff_t size,
3366 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367{
Al Viro6b4d0b22013-02-14 21:37:26 -05003368 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003370 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003371 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 struct qstr this;
3373
3374 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003375 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376
Hugh Dickins285b2c42011-08-03 16:21:20 -07003377 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 return ERR_PTR(-EINVAL);
3379
3380 if (shmem_acct_size(flags, size))
3381 return ERR_PTR(-ENOMEM);
3382
Al Viro6b4d0b22013-02-14 21:37:26 -05003383 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 this.name = name;
3385 this.len = strlen(name);
3386 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003387 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003388 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003389 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003390 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003392 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393
Al Viro6b4d0b22013-02-14 21:37:26 -05003394 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05003395 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003397 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398
Eric Parisc7277092013-12-02 11:24:19 +00003399 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04003400 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003402 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05003403 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3404 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003405 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003406
Al Viro6b4d0b22013-02-14 21:37:26 -05003407 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04003408 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05003409 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003410 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003411
Al Viro6b4d0b22013-02-14 21:37:26 -05003412 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414put_memory:
3415 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003416put_path:
3417 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05003418 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419}
Eric Parisc7277092013-12-02 11:24:19 +00003420
3421/**
3422 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3423 * kernel internal. There will be NO LSM permission checks against the
3424 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07003425 * higher layer. The users are the big_key and shm implementations. LSM
3426 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00003427 * @name: name for dentry (to be seen in /proc/<pid>/maps
3428 * @size: size to be set for the file
3429 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3430 */
3431struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3432{
3433 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3434}
3435
3436/**
3437 * shmem_file_setup - get an unlinked file living in tmpfs
3438 * @name: name for dentry (to be seen in /proc/<pid>/maps
3439 * @size: size to be set for the file
3440 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3441 */
3442struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3443{
3444 return __shmem_file_setup(name, size, flags, 0);
3445}
Keith Packard395e0dd2008-06-20 00:08:06 -07003446EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447
Randy Dunlap46711812008-03-19 17:00:41 -07003448/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3451 */
3452int shmem_zero_setup(struct vm_area_struct *vma)
3453{
3454 struct file *file;
3455 loff_t size = vma->vm_end - vma->vm_start;
3456
Hugh Dickins66fc1302015-06-14 09:48:09 -07003457 /*
3458 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3459 * between XFS directory reading and selinux: since this file is only
3460 * accessible to the user through its mapping, use S_PRIVATE flag to
3461 * bypass file security, in the same way as shmem_kernel_file_setup().
3462 */
3463 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 if (IS_ERR(file))
3465 return PTR_ERR(file);
3466
3467 if (vma->vm_file)
3468 fput(vma->vm_file);
3469 vma->vm_file = file;
3470 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 return 0;
3472}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003473
3474/**
3475 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3476 * @mapping: the page's address_space
3477 * @index: the page index
3478 * @gfp: the page allocator flags to use if allocating
3479 *
3480 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3481 * with any new page allocations done using the specified allocation flags.
3482 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3483 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3484 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3485 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003486 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3487 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003488 */
3489struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3490 pgoff_t index, gfp_t gfp)
3491{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003492#ifdef CONFIG_SHMEM
3493 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003494 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003495 int error;
3496
3497 BUG_ON(mapping->a_ops != &shmem_aops);
3498 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3499 if (error)
3500 page = ERR_PTR(error);
3501 else
3502 unlock_page(page);
3503 return page;
3504#else
3505 /*
3506 * The tiny !SHMEM case uses ramfs without swap
3507 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003508 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003509#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003510}
3511EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);