blob: 440e2a7e6c1c2ca706376eb72d719fb8b83af05a [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
379restart:
380 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
381 if (iter.index >= end)
382 break;
383
384 page = radix_tree_deref_slot(slot);
385
386 /*
387 * This should only be possible to happen at index 0, so we
388 * don't need to reset the counter, nor do we risk infinite
389 * restarts.
390 */
391 if (radix_tree_deref_retry(page))
392 goto restart;
393
394 if (radix_tree_exceptional_entry(page))
395 swapped++;
396
397 if (need_resched()) {
398 cond_resched_rcu();
399 start = iter.index + 1;
400 goto restart;
401 }
402 }
403
404 rcu_read_unlock();
405
406 return swapped << PAGE_SHIFT;
407}
408
409/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800410 * Determine (in bytes) how many of the shmem object's pages mapped by the
411 * given vma is swapped out.
412 *
413 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
414 * as long as the inode doesn't go away and racy results are not a problem.
415 */
416unsigned long shmem_swap_usage(struct vm_area_struct *vma)
417{
418 struct inode *inode = file_inode(vma->vm_file);
419 struct shmem_inode_info *info = SHMEM_I(inode);
420 struct address_space *mapping = inode->i_mapping;
421 unsigned long swapped;
422
423 /* Be careful as we don't hold info->lock */
424 swapped = READ_ONCE(info->swapped);
425
426 /*
427 * The easier cases are when the shmem object has nothing in swap, or
428 * the vma maps it whole. Then we can simply use the stats that we
429 * already track.
430 */
431 if (!swapped)
432 return 0;
433
434 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
435 return swapped << PAGE_SHIFT;
436
437 /* Here comes the more involved part */
438 return shmem_partial_swap_usage(mapping,
439 linear_page_index(vma, vma->vm_start),
440 linear_page_index(vma, vma->vm_end));
441}
442
443/*
Hugh Dickins24513262012-01-20 14:34:21 -0800444 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
445 */
446void shmem_unlock_mapping(struct address_space *mapping)
447{
448 struct pagevec pvec;
449 pgoff_t indices[PAGEVEC_SIZE];
450 pgoff_t index = 0;
451
452 pagevec_init(&pvec, 0);
453 /*
454 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
455 */
456 while (!mapping_unevictable(mapping)) {
457 /*
458 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
459 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
460 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700461 pvec.nr = find_get_entries(mapping, index,
462 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800463 if (!pvec.nr)
464 break;
465 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700466 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800467 check_move_unevictable_pages(pvec.pages, pvec.nr);
468 pagevec_release(&pvec);
469 cond_resched();
470 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700471}
472
473/*
474 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700475 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700476 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700477static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
478 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700480 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700482 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700483 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
484 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
485 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700486 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700487 pgoff_t indices[PAGEVEC_SIZE];
488 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700489 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700490 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700492 if (lend == -1)
493 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700494
495 pagevec_init(&pvec, 0);
496 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700497 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700498 pvec.nr = find_get_entries(mapping, index,
499 min(end - index, (pgoff_t)PAGEVEC_SIZE),
500 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700501 if (!pvec.nr)
502 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700503 for (i = 0; i < pagevec_count(&pvec); i++) {
504 struct page *page = pvec.pages[i];
505
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700506 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700507 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700508 break;
509
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700510 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700511 if (unfalloc)
512 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700513 nr_swaps_freed += !shmem_free_swap(mapping,
514 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700515 continue;
516 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700517
518 if (!trylock_page(page))
519 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700520 if (!unfalloc || !PageUptodate(page)) {
521 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800522 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700523 truncate_inode_page(mapping, page);
524 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700525 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700526 unlock_page(page);
527 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700528 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800529 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700530 cond_resched();
531 index++;
532 }
533
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700534 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700535 struct page *page = NULL;
536 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
537 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700538 unsigned int top = PAGE_CACHE_SIZE;
539 if (start > end) {
540 top = partial_end;
541 partial_end = 0;
542 }
543 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700544 set_page_dirty(page);
545 unlock_page(page);
546 page_cache_release(page);
547 }
548 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700549 if (partial_end) {
550 struct page *page = NULL;
551 shmem_getpage(inode, end, &page, SGP_READ, NULL);
552 if (page) {
553 zero_user_segment(page, 0, partial_end);
554 set_page_dirty(page);
555 unlock_page(page);
556 page_cache_release(page);
557 }
558 }
559 if (start >= end)
560 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700561
562 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700563 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700564 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700565
566 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700567 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700568 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700569 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700570 /* If all gone or hole-punch or unfalloc, we're done */
571 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700572 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700573 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700574 index = start;
575 continue;
576 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700577 for (i = 0; i < pagevec_count(&pvec); i++) {
578 struct page *page = pvec.pages[i];
579
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700580 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700581 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700582 break;
583
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700584 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700585 if (unfalloc)
586 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700587 if (shmem_free_swap(mapping, index, page)) {
588 /* Swap was replaced by page: retry */
589 index--;
590 break;
591 }
592 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700593 continue;
594 }
595
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700596 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700597 if (!unfalloc || !PageUptodate(page)) {
598 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800599 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700600 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700601 } else {
602 /* Page was replaced by swap: retry */
603 unlock_page(page);
604 index--;
605 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700606 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700607 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700608 unlock_page(page);
609 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700610 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800611 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700612 index++;
613 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700616 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 shmem_recalc_inode(inode);
618 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700619}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700621void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
622{
623 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700624 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700626EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Yu Zhao44a30222015-09-08 15:03:33 -0700628static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
629 struct kstat *stat)
630{
631 struct inode *inode = dentry->d_inode;
632 struct shmem_inode_info *info = SHMEM_I(inode);
633
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800634 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
635 spin_lock(&info->lock);
636 shmem_recalc_inode(inode);
637 spin_unlock(&info->lock);
638 }
Yu Zhao44a30222015-09-08 15:03:33 -0700639 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700640 return 0;
641}
642
Hugh Dickins94c1e622011-06-27 16:18:03 -0700643static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
David Howells75c3cfa2015-03-17 22:26:12 +0000645 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700646 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 int error;
648
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200649 error = inode_change_ok(inode, attr);
650 if (error)
651 return error;
652
Hugh Dickins94c1e622011-06-27 16:18:03 -0700653 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
654 loff_t oldsize = inode->i_size;
655 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000656
David Herrmann40e041a2014-08-08 14:25:27 -0700657 /* protected by i_mutex */
658 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
659 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
660 return -EPERM;
661
Hugh Dickins94c1e622011-06-27 16:18:03 -0700662 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700663 error = shmem_reacct_size(SHMEM_I(inode)->flags,
664 oldsize, newsize);
665 if (error)
666 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700667 i_size_write(inode, newsize);
668 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
669 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700670 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700671 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800672 if (oldsize > holebegin)
673 unmap_mapping_range(inode->i_mapping,
674 holebegin, 0, 1);
675 if (info->alloced)
676 shmem_truncate_range(inode,
677 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700678 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800679 if (oldsize > holebegin)
680 unmap_mapping_range(inode->i_mapping,
681 holebegin, 0, 1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700682 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 }
684
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200685 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200686 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -0800687 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return error;
689}
690
Al Viro1f895f72010-06-05 19:10:41 -0400691static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 struct shmem_inode_info *info = SHMEM_I(inode);
694
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000695 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 shmem_unacct_size(info->flags, inode->i_size);
697 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000698 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800700 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800702 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
Al Viro3ed47db2016-01-22 18:08:52 -0500704 }
Eric Parisb09e0fa2011-05-24 17:12:39 -0700705
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400706 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800707 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800708 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200709 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700712/*
713 * If swap found in inode, free it and move page from swapcache to filecache.
714 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700715static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700716 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700718 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700719 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700720 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700721 gfp_t gfp;
722 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700724 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700725 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700726 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -0700727 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800728
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800729 /*
730 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400731 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800732 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700733 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800734 */
735 if (shmem_swaplist.next != &info->swaplist)
736 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800737
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700738 gfp = mapping_gfp_mask(mapping);
739 if (shmem_should_replace_page(*pagep, gfp)) {
740 mutex_unlock(&shmem_swaplist_mutex);
741 error = shmem_replace_page(pagep, gfp, info, index);
742 mutex_lock(&shmem_swaplist_mutex);
743 /*
744 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700745 * allocation, but the inode might have been freed while we
746 * dropped it: although a racing shmem_evict_inode() cannot
747 * complete without emptying the radix_tree, our page lock
748 * on this swapcache page is not enough to prevent that -
749 * free_swap_and_cache() of our swap entry will only
750 * trylock_page(), removing swap from radix_tree whatever.
751 *
752 * We must not proceed to shmem_add_to_page_cache() if the
753 * inode has been freed, but of course we cannot rely on
754 * inode or mapping or info to check that. However, we can
755 * safely check if our swap entry is still in use (and here
756 * it can't have got reused for another page): if it's still
757 * in use, then the inode cannot have been freed yet, and we
758 * can safely proceed (if it's no longer in use, that tells
759 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700760 */
761 if (!page_swapcount(*pagep))
762 error = -ENOENT;
763 }
764
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800765 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700766 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
767 * but also to hold up shmem_evict_inode(): so inode cannot be freed
768 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800769 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700770 if (!error)
771 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700772 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700773 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700774 /*
775 * Truncation and eviction use free_swap_and_cache(), which
776 * only does trylock page: if we raced, best clean up here.
777 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700778 delete_from_swap_cache(*pagep);
779 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700780 if (!error) {
781 spin_lock(&info->lock);
782 info->swapped--;
783 spin_unlock(&info->lock);
784 swap_free(swap);
785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800787 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
790/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700791 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700793int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700795 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -0700797 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700798 int error = 0;
799
800 /*
801 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700802 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700803 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700804 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700805 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700806
807 /*
808 * Charge page using GFP_KERNEL while we can wait, before taking
809 * the shmem_swaplist_mutex which might hold up shmem_writepage().
810 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700811 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800812 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
813 false);
Hugh Dickins778dd892011-05-11 15:13:37 -0700814 if (error)
815 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700816 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -0700817 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800819 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700820 list_for_each_safe(this, next, &shmem_swaplist) {
821 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700822 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -0700823 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700824 else
825 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800826 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -0700827 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -0700828 break;
Johannes Weiner00501b52014-08-08 14:19:20 -0700829 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800831 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700832
Johannes Weiner00501b52014-08-08 14:19:20 -0700833 if (error) {
834 if (error != -ENOMEM)
835 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800836 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700837 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800838 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -0700839out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800840 unlock_page(page);
841 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700842 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
845/*
846 * Move the page from the page cache to the swap cache.
847 */
848static int shmem_writepage(struct page *page, struct writeback_control *wbc)
849{
850 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700853 swp_entry_t swap;
854 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 mapping = page->mapping;
858 index = page->index;
859 inode = mapping->host;
860 info = SHMEM_I(inode);
861 if (info->flags & VM_LOCKED)
862 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800863 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 goto redirty;
865
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800866 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100867 * Our capabilities prevent regular writeback or sync from ever calling
868 * shmem_writepage; but a stacking filesystem might use ->writepage of
869 * its underlying filesystem, in which case tmpfs should write out to
870 * swap only in response to memory pressure, and not for the writeback
871 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800872 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700873 if (!wbc->for_reclaim) {
874 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
875 goto redirty;
876 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700877
878 /*
879 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
880 * value into swapfile.c, the only way we can correctly account for a
881 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700882 *
883 * That's okay for a page already fallocated earlier, but if we have
884 * not yet completed the fallocation, then (a) we want to keep track
885 * of this page in case we have to undo it, and (b) it may not be a
886 * good idea to continue anyway, once we're pushing into swap. So
887 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700888 */
889 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700890 if (inode->i_private) {
891 struct shmem_falloc *shmem_falloc;
892 spin_lock(&inode->i_lock);
893 shmem_falloc = inode->i_private;
894 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -0700895 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -0700896 index >= shmem_falloc->start &&
897 index < shmem_falloc->next)
898 shmem_falloc->nr_unswapped++;
899 else
900 shmem_falloc = NULL;
901 spin_unlock(&inode->i_lock);
902 if (shmem_falloc)
903 goto redirty;
904 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700905 clear_highpage(page);
906 flush_dcache_page(page);
907 SetPageUptodate(page);
908 }
909
Hugh Dickins48f170f2011-07-25 17:12:37 -0700910 swap = get_swap_page();
911 if (!swap.val)
912 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800913
Vladimir Davydov37e84352016-01-20 15:02:56 -0800914 if (mem_cgroup_try_charge_swap(page, swap))
915 goto free_swap;
916
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700917 /*
918 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700919 * if it's not already there. Do it now before the page is
920 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700921 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700922 * we've incremented swapped, because shmem_unuse_inode() will
923 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700924 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700925 mutex_lock(&shmem_swaplist_mutex);
926 if (list_empty(&info->swaplist))
927 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700928
Hugh Dickins48f170f2011-07-25 17:12:37 -0700929 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins267a4c72015-12-11 13:40:55 -0800930 spin_lock(&info->lock);
931 shmem_recalc_inode(inode);
932 info->swapped++;
933 spin_unlock(&info->lock);
934
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800935 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700936 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
937
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700938 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800939 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700940 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 return 0;
942 }
943
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700944 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -0800945free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700946 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947redirty:
948 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800949 if (wbc->for_reclaim)
950 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
951 unlock_page(page);
952 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
955#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800956#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700957static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800958{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700959 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800960
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700961 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700962 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800963
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800964 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800965
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700966 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800967}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700968
969static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
970{
971 struct mempolicy *mpol = NULL;
972 if (sbinfo->mpol) {
973 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
974 mpol = sbinfo->mpol;
975 mpol_get(mpol);
976 spin_unlock(&sbinfo->stat_lock);
977 }
978 return mpol;
979}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800980#endif /* CONFIG_TMPFS */
981
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700982static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
983 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800986 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800988 /* Create a pseudo vma that just contains the policy */
989 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700990 /* Bias interleave by inode number to distribute better across nodes */
991 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800992 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700993 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700994
Mel Gorman18a2f372012-12-05 14:01:41 -0800995 page = swapin_readahead(swap, gfp, &pvma, 0);
996
997 /* Drop reference taken by mpol_shared_policy_lookup() */
998 mpol_cond_put(pvma.vm_policy);
999
1000 return page;
1001}
1002
1003static struct page *shmem_alloc_page(gfp_t gfp,
1004 struct shmem_inode_info *info, pgoff_t index)
1005{
1006 struct vm_area_struct pvma;
1007 struct page *page;
1008
1009 /* Create a pseudo vma that just contains the policy */
1010 pvma.vm_start = 0;
1011 /* Bias interleave by inode number to distribute better across nodes */
1012 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
1013 pvma.vm_ops = NULL;
1014 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1015
1016 page = alloc_page_vma(gfp, &pvma, 0);
1017
1018 /* Drop reference taken by mpol_shared_policy_lookup() */
1019 mpol_cond_put(pvma.vm_policy);
1020
1021 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001023#else /* !CONFIG_NUMA */
1024#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001025static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001026{
1027}
1028#endif /* CONFIG_TMPFS */
1029
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001030static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1031 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001033 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034}
1035
Hugh Dickins02098fe2008-02-04 22:28:42 -08001036static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001037 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001039 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001041#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001043#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1044static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1045{
1046 return NULL;
1047}
1048#endif
1049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001051 * When a page is moved from swapcache to shmem filecache (either by the
1052 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1053 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1054 * ignorance of the mapping it belongs to. If that mapping has special
1055 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1056 * we may need to copy to a suitable page before moving to filecache.
1057 *
1058 * In a future release, this may well be extended to respect cpuset and
1059 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1060 * but for now it is a simple matter of zone.
1061 */
1062static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1063{
1064 return page_zonenum(page) > gfp_zone(gfp);
1065}
1066
1067static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1068 struct shmem_inode_info *info, pgoff_t index)
1069{
1070 struct page *oldpage, *newpage;
1071 struct address_space *swap_mapping;
1072 pgoff_t swap_index;
1073 int error;
1074
1075 oldpage = *pagep;
1076 swap_index = page_private(oldpage);
1077 swap_mapping = page_mapping(oldpage);
1078
1079 /*
1080 * We have arrived here because our zones are constrained, so don't
1081 * limit chance of success by further cpuset and node constraints.
1082 */
1083 gfp &= ~GFP_CONSTRAINT_MASK;
1084 newpage = shmem_alloc_page(gfp, info, index);
1085 if (!newpage)
1086 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001087
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001088 page_cache_get(newpage);
1089 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001090 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001091
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001092 __SetPageLocked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001093 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001094 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001095 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001096 SetPageSwapCache(newpage);
1097
1098 /*
1099 * Our caller will very soon move newpage out of swapcache, but it's
1100 * a nice clean interface for us to replace oldpage by newpage there.
1101 */
1102 spin_lock_irq(&swap_mapping->tree_lock);
1103 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1104 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001105 if (!error) {
1106 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1107 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1108 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001109 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001110
Hugh Dickins0142ef62012-06-07 14:21:09 -07001111 if (unlikely(error)) {
1112 /*
1113 * Is this possible? I think not, now that our callers check
1114 * both PageSwapCache and page_private after getting page lock;
1115 * but be defensive. Reverse old to newpage for clear and free.
1116 */
1117 oldpage = newpage;
1118 } else {
Hugh Dickins45637ba2015-11-05 18:49:40 -08001119 mem_cgroup_replace_page(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001120 lru_cache_add_anon(newpage);
1121 *pagep = newpage;
1122 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001123
1124 ClearPageSwapCache(oldpage);
1125 set_page_private(oldpage, 0);
1126
1127 unlock_page(oldpage);
1128 page_cache_release(oldpage);
1129 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001130 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001131}
1132
1133/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001134 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 *
1136 * If we allocate a new one we do not mark it dirty. That's up to the
1137 * vm. If we swap it in we mark it dirty since we also free the swap
1138 * entry since a page cannot live in both the swap and page cache
1139 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001140static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001141 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142{
1143 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001144 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 struct shmem_sb_info *sbinfo;
Johannes Weiner00501b52014-08-08 14:19:20 -07001146 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001147 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 swp_entry_t swap;
1149 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001150 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001151 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001153 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001156 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001157 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001158 if (radix_tree_exceptional_entry(page)) {
1159 swap = radix_to_swp_entry(page);
1160 page = NULL;
1161 }
1162
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001163 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001164 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1165 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001166 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001167 }
1168
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001169 if (page && sgp == SGP_WRITE)
1170 mark_page_accessed(page);
1171
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001172 /* fallocated page? */
1173 if (page && !PageUptodate(page)) {
1174 if (sgp != SGP_READ)
1175 goto clear;
1176 unlock_page(page);
1177 page_cache_release(page);
1178 page = NULL;
1179 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001180 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001181 *pagep = page;
1182 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
1185 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001186 * Fast cache lookup did not find it:
1187 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001189 info = SHMEM_I(inode);
1190 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 if (swap.val) {
1193 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001194 page = lookup_swap_cache(swap);
1195 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001196 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001197 if (fault_type)
1198 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001199 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001200 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001201 error = -ENOMEM;
1202 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205
1206 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001207 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001208 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001209 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001210 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001211 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001212 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001213 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001215 goto failed;
1216 }
1217 wait_on_page_writeback(page);
1218
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001219 if (shmem_should_replace_page(page, gfp)) {
1220 error = shmem_replace_page(&page, gfp, info, index);
1221 if (error)
1222 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
1224
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001225 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1226 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001227 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001228 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001229 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001230 /*
1231 * We already confirmed swap under page lock, and make
1232 * no memory allocation here, so usually no possibility
1233 * of error; but free_swap_and_cache() only trylocks a
1234 * page, so it is just possible that the entry has been
1235 * truncated or holepunched since swap was confirmed.
1236 * shmem_undo_range() will have done some of the
1237 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001238 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001239 * Reset swap.val? No, leave it so "failed" goes back to
1240 * "repeat": reading a hole and writing should succeed.
1241 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001242 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001243 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001244 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001245 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001246 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001247 if (error)
1248 goto failed;
1249
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001250 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001251
Hugh Dickins54af60422011-08-03 16:21:24 -07001252 spin_lock(&info->lock);
1253 info->swapped--;
1254 shmem_recalc_inode(inode);
1255 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001256
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001257 if (sgp == SGP_WRITE)
1258 mark_page_accessed(page);
1259
Hugh Dickins27ab7002011-07-25 17:12:36 -07001260 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001261 set_page_dirty(page);
1262 swap_free(swap);
1263
Hugh Dickins54af60422011-08-03 16:21:24 -07001264 } else {
1265 if (shmem_acct_block(info->flags)) {
1266 error = -ENOSPC;
1267 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001269 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001270 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001271 sbinfo->max_blocks) >= 0) {
1272 error = -ENOSPC;
1273 goto unacct;
1274 }
Tim Chen7e496292010-08-09 17:19:05 -07001275 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
Hugh Dickins54af60422011-08-03 16:21:24 -07001278 page = shmem_alloc_page(gfp, info, index);
1279 if (!page) {
1280 error = -ENOMEM;
1281 goto decused;
1282 }
1283
Mel Gorman07a42782014-06-04 16:10:24 -07001284 __SetPageSwapBacked(page);
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001285 __SetPageLocked(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001286 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001287 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001288
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001289 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1290 false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001291 if (error)
1292 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001293 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001294 if (!error) {
1295 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001296 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001297 radix_tree_preload_end();
1298 }
1299 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001300 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001301 goto decused;
1302 }
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001303 mem_cgroup_commit_charge(page, memcg, false, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001304 lru_cache_add_anon(page);
1305
1306 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001308 inode->i_blocks += BLOCKS_PER_PAGE;
1309 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001311 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001312
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001313 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001314 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1315 */
1316 if (sgp == SGP_FALLOC)
1317 sgp = SGP_WRITE;
1318clear:
1319 /*
1320 * Let SGP_WRITE caller clear ends if write does not fill page;
1321 * but SGP_FALLOC on a page fallocated earlier must initialize
1322 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001323 */
1324 if (sgp != SGP_WRITE) {
1325 clear_highpage(page);
1326 flush_dcache_page(page);
1327 SetPageUptodate(page);
1328 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001329 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001330 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001331 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001332
Hugh Dickins54af60422011-08-03 16:21:24 -07001333 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001334 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001335 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001336 if (alloced) {
1337 ClearPageDirty(page);
1338 delete_from_page_cache(page);
1339 spin_lock(&info->lock);
1340 shmem_recalc_inode(inode);
1341 spin_unlock(&info->lock);
1342 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001343 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001344 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001345 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001346 *pagep = page;
1347 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001348
Nick Piggind0217ac2007-07-19 01:47:03 -07001349 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001350 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001352decused:
1353 if (sbinfo->max_blocks)
1354 percpu_counter_add(&sbinfo->used_blocks, -1);
1355unacct:
1356 shmem_unacct_blocks(info->flags, 1);
1357failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001358 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001359 error = -EEXIST;
1360unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001361 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001362 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001363 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001364 }
1365 if (error == -ENOSPC && !once++) {
1366 info = SHMEM_I(inode);
1367 spin_lock(&info->lock);
1368 shmem_recalc_inode(inode);
1369 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001370 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001371 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001372 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001373 goto repeat;
1374 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
1377static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1378{
Al Viro496ad9a2013-01-23 17:07:38 -05001379 struct inode *inode = file_inode(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001381 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001383 /*
1384 * Trinity finds that probing a hole which tmpfs is punching can
1385 * prevent the hole-punch from ever completing: which in turn
1386 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001387 * faulting pages into the hole while it's being punched. Although
1388 * shmem_undo_range() does remove the additions, it may be unable to
1389 * keep up, as each new page needs its own unmap_mapping_range() call,
1390 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1391 *
1392 * It does not matter if we sometimes reach this check just before the
1393 * hole-punch begins, so that one fault then races with the punch:
1394 * we just need to make racing faults a rare case.
1395 *
1396 * The implementation below would be much simpler if we just used a
1397 * standard mutex or completion: but we cannot take i_mutex in fault,
1398 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001399 */
1400 if (unlikely(inode->i_private)) {
1401 struct shmem_falloc *shmem_falloc;
1402
1403 spin_lock(&inode->i_lock);
1404 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001405 if (shmem_falloc &&
1406 shmem_falloc->waitq &&
1407 vmf->pgoff >= shmem_falloc->start &&
1408 vmf->pgoff < shmem_falloc->next) {
1409 wait_queue_head_t *shmem_falloc_waitq;
1410 DEFINE_WAIT(shmem_fault_wait);
1411
1412 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001413 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1414 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001415 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001416 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001417 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001418 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001419
1420 shmem_falloc_waitq = shmem_falloc->waitq;
1421 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1422 TASK_UNINTERRUPTIBLE);
1423 spin_unlock(&inode->i_lock);
1424 schedule();
1425
1426 /*
1427 * shmem_falloc_waitq points into the shmem_fallocate()
1428 * stack of the hole-punching task: shmem_falloc_waitq
1429 * is usually invalid by the time we reach here, but
1430 * finish_wait() does not dereference it in that case;
1431 * though i_lock needed lest racing with wake_up_all().
1432 */
1433 spin_lock(&inode->i_lock);
1434 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1435 spin_unlock(&inode->i_lock);
1436 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001437 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001438 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001439 }
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1442 if (error)
1443 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001444
Ying Han456f9982011-05-26 16:25:38 -07001445 if (ret & VM_FAULT_MAJOR) {
1446 count_vm_event(PGMAJFAULT);
1447 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1448 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001449 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450}
1451
1452#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001453static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454{
Al Viro496ad9a2013-01-23 17:07:38 -05001455 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001456 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001459static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1460 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
Al Viro496ad9a2013-01-23 17:07:38 -05001462 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001463 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001465 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1466 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468#endif
1469
1470int shmem_lock(struct file *file, int lock, struct user_struct *user)
1471{
Al Viro496ad9a2013-01-23 17:07:38 -05001472 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 struct shmem_inode_info *info = SHMEM_I(inode);
1474 int retval = -ENOMEM;
1475
1476 spin_lock(&info->lock);
1477 if (lock && !(info->flags & VM_LOCKED)) {
1478 if (!user_shm_lock(inode->i_size, user))
1479 goto out_nomem;
1480 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001481 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483 if (!lock && (info->flags & VM_LOCKED) && user) {
1484 user_shm_unlock(inode->i_size, user);
1485 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001486 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 }
1488 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490out_nomem:
1491 spin_unlock(&info->lock);
1492 return retval;
1493}
1494
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001495static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496{
1497 file_accessed(file);
1498 vma->vm_ops = &shmem_vm_ops;
1499 return 0;
1500}
1501
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001502static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001503 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504{
1505 struct inode *inode;
1506 struct shmem_inode_info *info;
1507 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1508
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001509 if (shmem_reserve_inode(sb))
1510 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 inode = new_inode(sb);
1513 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001514 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001515 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001518 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 info = SHMEM_I(inode);
1520 memset(info, 0, (char *)inode - (char *)info);
1521 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07001522 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001523 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001525 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001526 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 switch (mode & S_IFMT) {
1529 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001530 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 init_special_inode(inode, mode, dev);
1532 break;
1533 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001534 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 inode->i_op = &shmem_inode_operations;
1536 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001537 mpol_shared_policy_init(&info->policy,
1538 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 break;
1540 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001541 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 /* Some things misbehave if size == 0 on a directory */
1543 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1544 inode->i_op = &shmem_dir_inode_operations;
1545 inode->i_fop = &simple_dir_operations;
1546 break;
1547 case S_IFLNK:
1548 /*
1549 * Must not load anything in the rbtree,
1550 * mpol_free_shared_policy will not be called.
1551 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001552 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 break;
1554 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001555 } else
1556 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 return inode;
1558}
1559
Johannes Weiner0cd61442014-04-03 14:47:46 -07001560bool shmem_mapping(struct address_space *mapping)
1561{
Sasha Levinf0774d82015-02-23 05:38:00 -05001562 if (!mapping->host)
1563 return false;
1564
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001565 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001566}
1567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001569static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001570static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001572#ifdef CONFIG_TMPFS_XATTR
1573static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1574#else
1575#define shmem_initxattrs NULL
1576#endif
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001579shmem_write_begin(struct file *file, struct address_space *mapping,
1580 loff_t pos, unsigned len, unsigned flags,
1581 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
Nick Piggin800d15a2007-10-16 01:25:03 -07001583 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07001584 struct shmem_inode_info *info = SHMEM_I(inode);
Nick Piggin800d15a2007-10-16 01:25:03 -07001585 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07001586
1587 /* i_mutex is held by caller */
1588 if (unlikely(info->seals)) {
1589 if (info->seals & F_SEAL_WRITE)
1590 return -EPERM;
1591 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1592 return -EPERM;
1593 }
1594
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001595 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
Nick Piggin800d15a2007-10-16 01:25:03 -07001596}
1597
1598static int
1599shmem_write_end(struct file *file, struct address_space *mapping,
1600 loff_t pos, unsigned len, unsigned copied,
1601 struct page *page, void *fsdata)
1602{
1603 struct inode *inode = mapping->host;
1604
Hugh Dickinsd3602442008-02-04 22:28:44 -08001605 if (pos + copied > inode->i_size)
1606 i_size_write(inode, pos + copied);
1607
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001608 if (!PageUptodate(page)) {
1609 if (copied < PAGE_CACHE_SIZE) {
1610 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1611 zero_user_segments(page, 0, from,
1612 from + copied, PAGE_CACHE_SIZE);
1613 }
1614 SetPageUptodate(page);
1615 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001616 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001617 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001618 page_cache_release(page);
1619
Nick Piggin800d15a2007-10-16 01:25:03 -07001620 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
1622
Al Viro2ba5bbe2014-04-02 20:00:02 -04001623static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624{
Al Viro6e58e792014-02-03 17:07:03 -05001625 struct file *file = iocb->ki_filp;
1626 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001628 pgoff_t index;
1629 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001630 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02001631 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05001632 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05001633 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001634
1635 /*
1636 * Might this read be for a stacking filesystem? Then when reading
1637 * holes of a sparse file, we actually need to allocate those pages,
1638 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1639 */
Al Viro777eda22014-12-17 04:46:46 -05001640 if (!iter_is_iovec(to))
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001641 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
1643 index = *ppos >> PAGE_CACHE_SHIFT;
1644 offset = *ppos & ~PAGE_CACHE_MASK;
1645
1646 for (;;) {
1647 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001648 pgoff_t end_index;
1649 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 loff_t i_size = i_size_read(inode);
1651
1652 end_index = i_size >> PAGE_CACHE_SHIFT;
1653 if (index > end_index)
1654 break;
1655 if (index == end_index) {
1656 nr = i_size & ~PAGE_CACHE_MASK;
1657 if (nr <= offset)
1658 break;
1659 }
1660
Al Viro6e58e792014-02-03 17:07:03 -05001661 error = shmem_getpage(inode, index, &page, sgp, NULL);
1662 if (error) {
1663 if (error == -EINVAL)
1664 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 break;
1666 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001667 if (page)
1668 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
1670 /*
1671 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001672 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 */
1674 nr = PAGE_CACHE_SIZE;
1675 i_size = i_size_read(inode);
1676 end_index = i_size >> PAGE_CACHE_SHIFT;
1677 if (index == end_index) {
1678 nr = i_size & ~PAGE_CACHE_MASK;
1679 if (nr <= offset) {
1680 if (page)
1681 page_cache_release(page);
1682 break;
1683 }
1684 }
1685 nr -= offset;
1686
1687 if (page) {
1688 /*
1689 * If users can be writing to this page using arbitrary
1690 * virtual addresses, take care about potential aliasing
1691 * before reading the page on the kernel side.
1692 */
1693 if (mapping_writably_mapped(mapping))
1694 flush_dcache_page(page);
1695 /*
1696 * Mark the page accessed if we read the beginning.
1697 */
1698 if (!offset)
1699 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001700 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001702 page_cache_get(page);
1703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 /*
1706 * Ok, we have the page, and it's up-to-date, so
1707 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04001709 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05001710 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 offset += ret;
1712 index += offset >> PAGE_CACHE_SHIFT;
1713 offset &= ~PAGE_CACHE_MASK;
1714
1715 page_cache_release(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04001716 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 break;
Al Viro6e58e792014-02-03 17:07:03 -05001718 if (ret < nr) {
1719 error = -EFAULT;
1720 break;
1721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 cond_resched();
1723 }
1724
1725 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05001726 file_accessed(file);
1727 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728}
1729
Hugh Dickins708e3502011-07-25 17:12:32 -07001730static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1731 struct pipe_inode_info *pipe, size_t len,
1732 unsigned int flags)
1733{
1734 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001735 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001736 unsigned int loff, nr_pages, req_pages;
1737 struct page *pages[PIPE_DEF_BUFFERS];
1738 struct partial_page partial[PIPE_DEF_BUFFERS];
1739 struct page *page;
1740 pgoff_t index, end_index;
1741 loff_t isize, left;
1742 int error, page_nr;
1743 struct splice_pipe_desc spd = {
1744 .pages = pages,
1745 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001746 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001747 .flags = flags,
1748 .ops = &page_cache_pipe_buf_ops,
1749 .spd_release = spd_release_page,
1750 };
1751
Hugh Dickins71f0e072011-07-25 17:12:33 -07001752 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001753 if (unlikely(*ppos >= isize))
1754 return 0;
1755
1756 left = isize - *ppos;
1757 if (unlikely(left < len))
1758 len = left;
1759
1760 if (splice_grow_spd(pipe, &spd))
1761 return -ENOMEM;
1762
1763 index = *ppos >> PAGE_CACHE_SHIFT;
1764 loff = *ppos & ~PAGE_CACHE_MASK;
1765 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04001766 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07001767
Hugh Dickins708e3502011-07-25 17:12:32 -07001768 spd.nr_pages = find_get_pages_contig(mapping, index,
1769 nr_pages, spd.pages);
1770 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001771 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001772
Hugh Dickins708e3502011-07-25 17:12:32 -07001773 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001774 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1775 if (error)
1776 break;
1777 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001778 spd.pages[spd.nr_pages++] = page;
1779 index++;
1780 }
1781
Hugh Dickins708e3502011-07-25 17:12:32 -07001782 index = *ppos >> PAGE_CACHE_SHIFT;
1783 nr_pages = spd.nr_pages;
1784 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001785
Hugh Dickins708e3502011-07-25 17:12:32 -07001786 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1787 unsigned int this_len;
1788
1789 if (!len)
1790 break;
1791
Hugh Dickins708e3502011-07-25 17:12:32 -07001792 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1793 page = spd.pages[page_nr];
1794
Hugh Dickins71f0e072011-07-25 17:12:33 -07001795 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001796 error = shmem_getpage(inode, index, &page,
1797 SGP_CACHE, NULL);
1798 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001799 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001800 unlock_page(page);
1801 page_cache_release(spd.pages[page_nr]);
1802 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001803 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001804
1805 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001806 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1807 if (unlikely(!isize || index > end_index))
1808 break;
1809
Hugh Dickins708e3502011-07-25 17:12:32 -07001810 if (end_index == index) {
1811 unsigned int plen;
1812
Hugh Dickins708e3502011-07-25 17:12:32 -07001813 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1814 if (plen <= loff)
1815 break;
1816
Hugh Dickins708e3502011-07-25 17:12:32 -07001817 this_len = min(this_len, plen - loff);
1818 len = this_len;
1819 }
1820
1821 spd.partial[page_nr].offset = loff;
1822 spd.partial[page_nr].len = this_len;
1823 len -= this_len;
1824 loff = 0;
1825 spd.nr_pages++;
1826 index++;
1827 }
1828
Hugh Dickins708e3502011-07-25 17:12:32 -07001829 while (page_nr < nr_pages)
1830 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001831
1832 if (spd.nr_pages)
1833 error = splice_to_pipe(pipe, &spd);
1834
Eric Dumazet047fe362012-06-12 15:24:40 +02001835 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001836
1837 if (error > 0) {
1838 *ppos += error;
1839 file_accessed(in);
1840 }
1841 return error;
1842}
1843
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001844/*
1845 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1846 */
1847static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001848 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001849{
1850 struct page *page;
1851 struct pagevec pvec;
1852 pgoff_t indices[PAGEVEC_SIZE];
1853 bool done = false;
1854 int i;
1855
1856 pagevec_init(&pvec, 0);
1857 pvec.nr = 1; /* start small: we may be there already */
1858 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001859 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001860 pvec.nr, pvec.pages, indices);
1861 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001862 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001863 index = end;
1864 break;
1865 }
1866 for (i = 0; i < pvec.nr; i++, index++) {
1867 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001868 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001869 done = true;
1870 break;
1871 }
1872 index = indices[i];
1873 }
1874 page = pvec.pages[i];
1875 if (page && !radix_tree_exceptional_entry(page)) {
1876 if (!PageUptodate(page))
1877 page = NULL;
1878 }
1879 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001880 (page && whence == SEEK_DATA) ||
1881 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001882 done = true;
1883 break;
1884 }
1885 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001886 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001887 pagevec_release(&pvec);
1888 pvec.nr = PAGEVEC_SIZE;
1889 cond_resched();
1890 }
1891 return index;
1892}
1893
Andrew Morton965c8e52012-12-17 15:59:39 -08001894static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001895{
1896 struct address_space *mapping = file->f_mapping;
1897 struct inode *inode = mapping->host;
1898 pgoff_t start, end;
1899 loff_t new_offset;
1900
Andrew Morton965c8e52012-12-17 15:59:39 -08001901 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1902 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001903 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05001904 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001905 /* We're holding i_mutex so we can access i_size directly */
1906
1907 if (offset < 0)
1908 offset = -EINVAL;
1909 else if (offset >= inode->i_size)
1910 offset = -ENXIO;
1911 else {
1912 start = offset >> PAGE_CACHE_SHIFT;
1913 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001914 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001915 new_offset <<= PAGE_CACHE_SHIFT;
1916 if (new_offset > offset) {
1917 if (new_offset < inode->i_size)
1918 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001919 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001920 offset = -ENXIO;
1921 else
1922 offset = inode->i_size;
1923 }
1924 }
1925
Hugh Dickins387aae62013-08-04 11:30:25 -07001926 if (offset >= 0)
1927 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05001928 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001929 return offset;
1930}
1931
David Herrmann05f65b52014-08-08 14:25:36 -07001932/*
1933 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1934 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1935 */
1936#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1937#define LAST_SCAN 4 /* about 150ms max */
1938
1939static void shmem_tag_pins(struct address_space *mapping)
1940{
1941 struct radix_tree_iter iter;
1942 void **slot;
1943 pgoff_t start;
1944 struct page *page;
1945
1946 lru_add_drain();
1947 start = 0;
1948 rcu_read_lock();
1949
1950restart:
1951 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1952 page = radix_tree_deref_slot(slot);
1953 if (!page || radix_tree_exception(page)) {
1954 if (radix_tree_deref_retry(page))
1955 goto restart;
1956 } else if (page_count(page) - page_mapcount(page) > 1) {
1957 spin_lock_irq(&mapping->tree_lock);
1958 radix_tree_tag_set(&mapping->page_tree, iter.index,
1959 SHMEM_TAG_PINNED);
1960 spin_unlock_irq(&mapping->tree_lock);
1961 }
1962
1963 if (need_resched()) {
1964 cond_resched_rcu();
1965 start = iter.index + 1;
1966 goto restart;
1967 }
1968 }
1969 rcu_read_unlock();
1970}
1971
1972/*
1973 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1974 * via get_user_pages(), drivers might have some pending I/O without any active
1975 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1976 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1977 * them to be dropped.
1978 * The caller must guarantee that no new user will acquire writable references
1979 * to those pages to avoid races.
1980 */
David Herrmann40e041a2014-08-08 14:25:27 -07001981static int shmem_wait_for_pins(struct address_space *mapping)
1982{
David Herrmann05f65b52014-08-08 14:25:36 -07001983 struct radix_tree_iter iter;
1984 void **slot;
1985 pgoff_t start;
1986 struct page *page;
1987 int error, scan;
1988
1989 shmem_tag_pins(mapping);
1990
1991 error = 0;
1992 for (scan = 0; scan <= LAST_SCAN; scan++) {
1993 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1994 break;
1995
1996 if (!scan)
1997 lru_add_drain_all();
1998 else if (schedule_timeout_killable((HZ << scan) / 200))
1999 scan = LAST_SCAN;
2000
2001 start = 0;
2002 rcu_read_lock();
2003restart:
2004 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2005 start, SHMEM_TAG_PINNED) {
2006
2007 page = radix_tree_deref_slot(slot);
2008 if (radix_tree_exception(page)) {
2009 if (radix_tree_deref_retry(page))
2010 goto restart;
2011
2012 page = NULL;
2013 }
2014
2015 if (page &&
2016 page_count(page) - page_mapcount(page) != 1) {
2017 if (scan < LAST_SCAN)
2018 goto continue_resched;
2019
2020 /*
2021 * On the last scan, we clean up all those tags
2022 * we inserted; but make a note that we still
2023 * found pages pinned.
2024 */
2025 error = -EBUSY;
2026 }
2027
2028 spin_lock_irq(&mapping->tree_lock);
2029 radix_tree_tag_clear(&mapping->page_tree,
2030 iter.index, SHMEM_TAG_PINNED);
2031 spin_unlock_irq(&mapping->tree_lock);
2032continue_resched:
2033 if (need_resched()) {
2034 cond_resched_rcu();
2035 start = iter.index + 1;
2036 goto restart;
2037 }
2038 }
2039 rcu_read_unlock();
2040 }
2041
2042 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002043}
2044
2045#define F_ALL_SEALS (F_SEAL_SEAL | \
2046 F_SEAL_SHRINK | \
2047 F_SEAL_GROW | \
2048 F_SEAL_WRITE)
2049
2050int shmem_add_seals(struct file *file, unsigned int seals)
2051{
2052 struct inode *inode = file_inode(file);
2053 struct shmem_inode_info *info = SHMEM_I(inode);
2054 int error;
2055
2056 /*
2057 * SEALING
2058 * Sealing allows multiple parties to share a shmem-file but restrict
2059 * access to a specific subset of file operations. Seals can only be
2060 * added, but never removed. This way, mutually untrusted parties can
2061 * share common memory regions with a well-defined policy. A malicious
2062 * peer can thus never perform unwanted operations on a shared object.
2063 *
2064 * Seals are only supported on special shmem-files and always affect
2065 * the whole underlying inode. Once a seal is set, it may prevent some
2066 * kinds of access to the file. Currently, the following seals are
2067 * defined:
2068 * SEAL_SEAL: Prevent further seals from being set on this file
2069 * SEAL_SHRINK: Prevent the file from shrinking
2070 * SEAL_GROW: Prevent the file from growing
2071 * SEAL_WRITE: Prevent write access to the file
2072 *
2073 * As we don't require any trust relationship between two parties, we
2074 * must prevent seals from being removed. Therefore, sealing a file
2075 * only adds a given set of seals to the file, it never touches
2076 * existing seals. Furthermore, the "setting seals"-operation can be
2077 * sealed itself, which basically prevents any further seal from being
2078 * added.
2079 *
2080 * Semantics of sealing are only defined on volatile files. Only
2081 * anonymous shmem files support sealing. More importantly, seals are
2082 * never written to disk. Therefore, there's no plan to support it on
2083 * other file types.
2084 */
2085
2086 if (file->f_op != &shmem_file_operations)
2087 return -EINVAL;
2088 if (!(file->f_mode & FMODE_WRITE))
2089 return -EPERM;
2090 if (seals & ~(unsigned int)F_ALL_SEALS)
2091 return -EINVAL;
2092
Al Viro59551022016-01-22 15:40:57 -05002093 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002094
2095 if (info->seals & F_SEAL_SEAL) {
2096 error = -EPERM;
2097 goto unlock;
2098 }
2099
2100 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2101 error = mapping_deny_writable(file->f_mapping);
2102 if (error)
2103 goto unlock;
2104
2105 error = shmem_wait_for_pins(file->f_mapping);
2106 if (error) {
2107 mapping_allow_writable(file->f_mapping);
2108 goto unlock;
2109 }
2110 }
2111
2112 info->seals |= seals;
2113 error = 0;
2114
2115unlock:
Al Viro59551022016-01-22 15:40:57 -05002116 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002117 return error;
2118}
2119EXPORT_SYMBOL_GPL(shmem_add_seals);
2120
2121int shmem_get_seals(struct file *file)
2122{
2123 if (file->f_op != &shmem_file_operations)
2124 return -EINVAL;
2125
2126 return SHMEM_I(file_inode(file))->seals;
2127}
2128EXPORT_SYMBOL_GPL(shmem_get_seals);
2129
2130long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2131{
2132 long error;
2133
2134 switch (cmd) {
2135 case F_ADD_SEALS:
2136 /* disallow upper 32bit */
2137 if (arg > UINT_MAX)
2138 return -EINVAL;
2139
2140 error = shmem_add_seals(file, arg);
2141 break;
2142 case F_GET_SEALS:
2143 error = shmem_get_seals(file);
2144 break;
2145 default:
2146 error = -EINVAL;
2147 break;
2148 }
2149
2150 return error;
2151}
2152
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002153static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2154 loff_t len)
2155{
Al Viro496ad9a2013-01-23 17:07:38 -05002156 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002157 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002158 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002159 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002160 pgoff_t start, index, end;
2161 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002162
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002163 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2164 return -EOPNOTSUPP;
2165
Al Viro59551022016-01-22 15:40:57 -05002166 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002167
2168 if (mode & FALLOC_FL_PUNCH_HOLE) {
2169 struct address_space *mapping = file->f_mapping;
2170 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2171 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002172 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002173
David Herrmann40e041a2014-08-08 14:25:27 -07002174 /* protected by i_mutex */
2175 if (info->seals & F_SEAL_WRITE) {
2176 error = -EPERM;
2177 goto out;
2178 }
2179
Hugh Dickins8e205f72014-07-23 14:00:10 -07002180 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002181 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2182 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2183 spin_lock(&inode->i_lock);
2184 inode->i_private = &shmem_falloc;
2185 spin_unlock(&inode->i_lock);
2186
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002187 if ((u64)unmap_end > (u64)unmap_start)
2188 unmap_mapping_range(mapping, unmap_start,
2189 1 + unmap_end - unmap_start, 0);
2190 shmem_truncate_range(inode, offset, offset + len - 1);
2191 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002192
2193 spin_lock(&inode->i_lock);
2194 inode->i_private = NULL;
2195 wake_up_all(&shmem_falloc_waitq);
2196 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002197 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002198 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002199 }
2200
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002201 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2202 error = inode_newsize_ok(inode, offset + len);
2203 if (error)
2204 goto out;
2205
David Herrmann40e041a2014-08-08 14:25:27 -07002206 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2207 error = -EPERM;
2208 goto out;
2209 }
2210
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002211 start = offset >> PAGE_CACHE_SHIFT;
2212 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2213 /* Try to avoid a swapstorm if len is impossible to satisfy */
2214 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2215 error = -ENOSPC;
2216 goto out;
2217 }
2218
Hugh Dickins8e205f72014-07-23 14:00:10 -07002219 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002220 shmem_falloc.start = start;
2221 shmem_falloc.next = start;
2222 shmem_falloc.nr_falloced = 0;
2223 shmem_falloc.nr_unswapped = 0;
2224 spin_lock(&inode->i_lock);
2225 inode->i_private = &shmem_falloc;
2226 spin_unlock(&inode->i_lock);
2227
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002228 for (index = start; index < end; index++) {
2229 struct page *page;
2230
2231 /*
2232 * Good, the fallocate(2) manpage permits EINTR: we may have
2233 * been interrupted because we are using up too much memory.
2234 */
2235 if (signal_pending(current))
2236 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002237 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2238 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002239 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002240 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002241 NULL);
2242 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002243 /* Remove the !PageUptodate pages we added */
2244 shmem_undo_range(inode,
2245 (loff_t)start << PAGE_CACHE_SHIFT,
2246 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002247 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002248 }
2249
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002250 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002251 * Inform shmem_writepage() how far we have reached.
2252 * No need for lock or barrier: we have the page lock.
2253 */
2254 shmem_falloc.next++;
2255 if (!PageUptodate(page))
2256 shmem_falloc.nr_falloced++;
2257
2258 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002259 * If !PageUptodate, leave it that way so that freeable pages
2260 * can be recognized if we need to rollback on error later.
2261 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002262 * than free the pages we are allocating (and SGP_CACHE pages
2263 * might still be clean: we now need to mark those dirty too).
2264 */
2265 set_page_dirty(page);
2266 unlock_page(page);
2267 page_cache_release(page);
2268 cond_resched();
2269 }
2270
2271 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2272 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002273 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002274undone:
2275 spin_lock(&inode->i_lock);
2276 inode->i_private = NULL;
2277 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002278out:
Al Viro59551022016-01-22 15:40:57 -05002279 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002280 return error;
2281}
2282
David Howells726c3342006-06-23 02:02:58 -07002283static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284{
David Howells726c3342006-06-23 02:02:58 -07002285 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287 buf->f_type = TMPFS_MAGIC;
2288 buf->f_bsize = PAGE_CACHE_SIZE;
2289 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002290 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002292 buf->f_bavail =
2293 buf->f_bfree = sbinfo->max_blocks -
2294 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002295 }
2296 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 buf->f_files = sbinfo->max_inodes;
2298 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 }
2300 /* else leave those fields 0 like simple_statfs */
2301 return 0;
2302}
2303
2304/*
2305 * File creation. Allocate an inode, and we're done..
2306 */
2307static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002308shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002310 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 int error = -ENOSPC;
2312
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002313 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002315 error = simple_acl_create(dir, inode);
2316 if (error)
2317 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002318 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002319 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002320 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002321 if (error && error != -EOPNOTSUPP)
2322 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002323
Al Viro718deb62009-12-16 19:35:36 -05002324 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 dir->i_size += BOGO_DIRENT_SIZE;
2326 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2327 d_instantiate(dentry, inode);
2328 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 }
2330 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002331out_iput:
2332 iput(inode);
2333 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334}
2335
Al Viro60545d02013-06-07 01:20:27 -04002336static int
2337shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2338{
2339 struct inode *inode;
2340 int error = -ENOSPC;
2341
2342 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2343 if (inode) {
2344 error = security_inode_init_security(inode, dir,
2345 NULL,
2346 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002347 if (error && error != -EOPNOTSUPP)
2348 goto out_iput;
2349 error = simple_acl_create(dir, inode);
2350 if (error)
2351 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002352 d_tmpfile(dentry, inode);
2353 }
2354 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002355out_iput:
2356 iput(inode);
2357 return error;
Al Viro60545d02013-06-07 01:20:27 -04002358}
2359
Al Viro18bb1db2011-07-26 01:41:39 -04002360static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361{
2362 int error;
2363
2364 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2365 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002366 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 return 0;
2368}
2369
Al Viro4acdaf22011-07-26 01:42:34 -04002370static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002371 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372{
2373 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2374}
2375
2376/*
2377 * Link a file..
2378 */
2379static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2380{
David Howells75c3cfa2015-03-17 22:26:12 +00002381 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002382 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
2384 /*
2385 * No ordinary (disk based) filesystem counts links as inodes;
2386 * but each new link needs a new dentry, pinning lowmem, and
2387 * tmpfs dentries cannot be pruned until they are unlinked.
2388 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002389 ret = shmem_reserve_inode(inode->i_sb);
2390 if (ret)
2391 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
2393 dir->i_size += BOGO_DIRENT_SIZE;
2394 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002395 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002396 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 dget(dentry); /* Extra pinning count for the created dentry */
2398 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002399out:
2400 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401}
2402
2403static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2404{
David Howells75c3cfa2015-03-17 22:26:12 +00002405 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002407 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2408 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
2410 dir->i_size -= BOGO_DIRENT_SIZE;
2411 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002412 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 dput(dentry); /* Undo the count from "create" - this does all the work */
2414 return 0;
2415}
2416
2417static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2418{
2419 if (!simple_empty(dentry))
2420 return -ENOTEMPTY;
2421
David Howells75c3cfa2015-03-17 22:26:12 +00002422 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002423 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return shmem_unlink(dir, dentry);
2425}
2426
Miklos Szeredi37456772014-07-23 15:15:34 +02002427static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2428{
David Howellse36cb0b2015-01-29 12:02:35 +00002429 bool old_is_dir = d_is_dir(old_dentry);
2430 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002431
2432 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2433 if (old_is_dir) {
2434 drop_nlink(old_dir);
2435 inc_nlink(new_dir);
2436 } else {
2437 drop_nlink(new_dir);
2438 inc_nlink(old_dir);
2439 }
2440 }
2441 old_dir->i_ctime = old_dir->i_mtime =
2442 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002443 d_inode(old_dentry)->i_ctime =
2444 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
Miklos Szeredi37456772014-07-23 15:15:34 +02002445
2446 return 0;
2447}
2448
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002449static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2450{
2451 struct dentry *whiteout;
2452 int error;
2453
2454 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2455 if (!whiteout)
2456 return -ENOMEM;
2457
2458 error = shmem_mknod(old_dir, whiteout,
2459 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2460 dput(whiteout);
2461 if (error)
2462 return error;
2463
2464 /*
2465 * Cheat and hash the whiteout while the old dentry is still in
2466 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2467 *
2468 * d_lookup() will consistently find one of them at this point,
2469 * not sure which one, but that isn't even important.
2470 */
2471 d_rehash(whiteout);
2472 return 0;
2473}
2474
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475/*
2476 * The VFS layer already does all the dentry stuff for rename,
2477 * we just have to decrement the usage count for the target if
2478 * it exists so that the VFS layer correctly free's it when it
2479 * gets overwritten.
2480 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002481static 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 -07002482{
David Howells75c3cfa2015-03-17 22:26:12 +00002483 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 int they_are_dirs = S_ISDIR(inode->i_mode);
2485
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002486 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002487 return -EINVAL;
2488
Miklos Szeredi37456772014-07-23 15:15:34 +02002489 if (flags & RENAME_EXCHANGE)
2490 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2491
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 if (!simple_empty(new_dentry))
2493 return -ENOTEMPTY;
2494
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002495 if (flags & RENAME_WHITEOUT) {
2496 int error;
2497
2498 error = shmem_whiteout(old_dir, old_dentry);
2499 if (error)
2500 return error;
2501 }
2502
David Howells75c3cfa2015-03-17 22:26:12 +00002503 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002505 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002506 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002507 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002510 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002511 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 }
2513
2514 old_dir->i_size -= BOGO_DIRENT_SIZE;
2515 new_dir->i_size += BOGO_DIRENT_SIZE;
2516 old_dir->i_ctime = old_dir->i_mtime =
2517 new_dir->i_ctime = new_dir->i_mtime =
2518 inode->i_ctime = CURRENT_TIME;
2519 return 0;
2520}
2521
2522static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2523{
2524 int error;
2525 int len;
2526 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002527 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 struct shmem_inode_info *info;
2529
2530 len = strlen(symname) + 1;
2531 if (len > PAGE_CACHE_SIZE)
2532 return -ENAMETOOLONG;
2533
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002534 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 if (!inode)
2536 return -ENOSPC;
2537
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002538 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002539 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002540 if (error) {
2541 if (error != -EOPNOTSUPP) {
2542 iput(inode);
2543 return error;
2544 }
2545 error = 0;
2546 }
2547
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 info = SHMEM_I(inode);
2549 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002550 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05002551 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
2552 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002553 iput(inode);
2554 return -ENOMEM;
2555 }
2556 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05002558 inode_nohighmem(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2560 if (error) {
2561 iput(inode);
2562 return error;
2563 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002564 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05002566 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002567 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002569 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 page_cache_release(page);
2571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 dir->i_size += BOGO_DIRENT_SIZE;
2573 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2574 d_instantiate(dentry, inode);
2575 dget(dentry);
2576 return 0;
2577}
2578
Al Virofceef392015-12-29 15:58:39 -05002579static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580{
Al Virofceef392015-12-29 15:58:39 -05002581 mark_page_accessed(arg);
2582 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583}
2584
Al Viro6b255392015-11-17 10:20:54 -05002585static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05002586 struct inode *inode,
2587 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05002590 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05002591 if (!dentry) {
2592 page = find_get_page(inode->i_mapping, 0);
2593 if (!page)
2594 return ERR_PTR(-ECHILD);
2595 if (!PageUptodate(page)) {
2596 put_page(page);
2597 return ERR_PTR(-ECHILD);
2598 }
2599 } else {
2600 error = shmem_getpage(inode, 0, &page, SGP_READ, NULL);
2601 if (error)
2602 return ERR_PTR(error);
2603 unlock_page(page);
2604 }
Al Virofceef392015-12-29 15:58:39 -05002605 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05002606 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607}
2608
Eric Parisb09e0fa2011-05-24 17:12:39 -07002609#ifdef CONFIG_TMPFS_XATTR
2610/*
2611 * Superblocks without xattr inode operations may get some security.* xattr
2612 * support from the LSM "for free". As soon as we have any other xattrs
2613 * like ACLs, we also need to implement the security.* handlers at
2614 * filesystem level, though.
2615 */
2616
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002617/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002618 * Callback for security_inode_init_security() for acquiring xattrs.
2619 */
2620static int shmem_initxattrs(struct inode *inode,
2621 const struct xattr *xattr_array,
2622 void *fs_info)
2623{
2624 struct shmem_inode_info *info = SHMEM_I(inode);
2625 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002626 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002627 size_t len;
2628
2629 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002630 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002631 if (!new_xattr)
2632 return -ENOMEM;
2633
2634 len = strlen(xattr->name) + 1;
2635 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2636 GFP_KERNEL);
2637 if (!new_xattr->name) {
2638 kfree(new_xattr);
2639 return -ENOMEM;
2640 }
2641
2642 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2643 XATTR_SECURITY_PREFIX_LEN);
2644 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2645 xattr->name, len);
2646
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002647 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002648 }
2649
2650 return 0;
2651}
2652
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002653static int shmem_xattr_handler_get(const struct xattr_handler *handler,
2654 struct dentry *dentry, const char *name,
2655 void *buffer, size_t size)
2656{
2657 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2658
2659 name = xattr_full_name(handler, name);
2660 return simple_xattr_get(&info->xattrs, name, buffer, size);
2661}
2662
2663static int shmem_xattr_handler_set(const struct xattr_handler *handler,
2664 struct dentry *dentry, const char *name,
2665 const void *value, size_t size, int flags)
2666{
2667 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2668
2669 name = xattr_full_name(handler, name);
2670 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2671}
2672
2673static const struct xattr_handler shmem_security_xattr_handler = {
2674 .prefix = XATTR_SECURITY_PREFIX,
2675 .get = shmem_xattr_handler_get,
2676 .set = shmem_xattr_handler_set,
2677};
2678
2679static const struct xattr_handler shmem_trusted_xattr_handler = {
2680 .prefix = XATTR_TRUSTED_PREFIX,
2681 .get = shmem_xattr_handler_get,
2682 .set = shmem_xattr_handler_set,
2683};
2684
Eric Parisb09e0fa2011-05-24 17:12:39 -07002685static const struct xattr_handler *shmem_xattr_handlers[] = {
2686#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002687 &posix_acl_access_xattr_handler,
2688 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002689#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002690 &shmem_security_xattr_handler,
2691 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002692 NULL
2693};
2694
Eric Parisb09e0fa2011-05-24 17:12:39 -07002695static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2696{
David Howells75c3cfa2015-03-17 22:26:12 +00002697 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01002698 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002699}
2700#endif /* CONFIG_TMPFS_XATTR */
2701
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002702static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05002704 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002705#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002706 .setxattr = generic_setxattr,
2707 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002708 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002709 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002710#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711};
2712
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002713static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05002715 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002716#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002717 .setxattr = generic_setxattr,
2718 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002719 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01002720 .removexattr = generic_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002721#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002722};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002723
David M. Grimes91828a42006-10-17 00:09:45 -07002724static struct dentry *shmem_get_parent(struct dentry *child)
2725{
2726 return ERR_PTR(-ESTALE);
2727}
2728
2729static int shmem_match(struct inode *ino, void *vfh)
2730{
2731 __u32 *fh = vfh;
2732 __u64 inum = fh[2];
2733 inum = (inum << 32) | fh[1];
2734 return ino->i_ino == inum && fh[0] == ino->i_generation;
2735}
2736
Christoph Hellwig480b1162007-10-21 16:42:13 -07002737static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2738 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002739{
David M. Grimes91828a42006-10-17 00:09:45 -07002740 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002741 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002742 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002743
Christoph Hellwig480b1162007-10-21 16:42:13 -07002744 if (fh_len < 3)
2745 return NULL;
2746
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002747 inum = fid->raw[2];
2748 inum = (inum << 32) | fid->raw[1];
2749
Christoph Hellwig480b1162007-10-21 16:42:13 -07002750 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2751 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002752 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002753 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002754 iput(inode);
2755 }
2756
Christoph Hellwig480b1162007-10-21 16:42:13 -07002757 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002758}
2759
Al Virob0b03822012-04-02 14:34:06 -04002760static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2761 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002762{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302763 if (*len < 3) {
2764 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002765 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302766 }
David M. Grimes91828a42006-10-17 00:09:45 -07002767
Al Viro1d3382cb2010-10-23 15:19:20 -04002768 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002769 /* Unfortunately insert_inode_hash is not idempotent,
2770 * so as we hash inodes here rather than at creation
2771 * time, we need a lock to ensure we only try
2772 * to do it once
2773 */
2774 static DEFINE_SPINLOCK(lock);
2775 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002776 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002777 __insert_inode_hash(inode,
2778 inode->i_ino + inode->i_generation);
2779 spin_unlock(&lock);
2780 }
2781
2782 fh[0] = inode->i_generation;
2783 fh[1] = inode->i_ino;
2784 fh[2] = ((__u64)inode->i_ino) >> 32;
2785
2786 *len = 3;
2787 return 1;
2788}
2789
Christoph Hellwig39655162007-10-21 16:42:17 -07002790static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002791 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002792 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002793 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002794};
2795
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002796static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2797 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798{
2799 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002800 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002801 uid_t uid;
2802 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002804 while (options != NULL) {
2805 this_char = options;
2806 for (;;) {
2807 /*
2808 * NUL-terminate this option: unfortunately,
2809 * mount options form a comma-separated list,
2810 * but mpol's nodelist may also contain commas.
2811 */
2812 options = strchr(options, ',');
2813 if (options == NULL)
2814 break;
2815 options++;
2816 if (!isdigit(*options)) {
2817 options[-1] = '\0';
2818 break;
2819 }
2820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 if (!*this_char)
2822 continue;
2823 if ((value = strchr(this_char,'=')) != NULL) {
2824 *value++ = 0;
2825 } else {
2826 printk(KERN_ERR
2827 "tmpfs: No value for mount option '%s'\n",
2828 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002829 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 }
2831
2832 if (!strcmp(this_char,"size")) {
2833 unsigned long long size;
2834 size = memparse(value,&rest);
2835 if (*rest == '%') {
2836 size <<= PAGE_SHIFT;
2837 size *= totalram_pages;
2838 do_div(size, 100);
2839 rest++;
2840 }
2841 if (*rest)
2842 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002843 sbinfo->max_blocks =
2844 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002846 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 if (*rest)
2848 goto bad_val;
2849 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002850 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 if (*rest)
2852 goto bad_val;
2853 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002854 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002856 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 if (*rest)
2858 goto bad_val;
2859 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002860 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002862 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 if (*rest)
2864 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002865 sbinfo->uid = make_kuid(current_user_ns(), uid);
2866 if (!uid_valid(sbinfo->uid))
2867 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002869 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002871 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 if (*rest)
2873 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002874 sbinfo->gid = make_kgid(current_user_ns(), gid);
2875 if (!gid_valid(sbinfo->gid))
2876 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002877 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002878 mpol_put(mpol);
2879 mpol = NULL;
2880 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002881 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 } else {
2883 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2884 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002885 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 }
2887 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002888 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 return 0;
2890
2891bad_val:
2892 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2893 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002894error:
2895 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 return 1;
2897
2898}
2899
2900static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2901{
2902 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002903 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002904 unsigned long inodes;
2905 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
Greg Thelen5f001102013-02-22 16:36:01 -08002907 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002908 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002909 return error;
2910
2911 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002912 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002913 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002914 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002915 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002916 goto out;
2917 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002918 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002919 * but we must separately disallow unlimited->limited, because
2920 * in that case we have no record of how much is already in use.
2921 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002922 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002923 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002924 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002925 goto out;
2926
2927 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002928 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002929 sbinfo->max_inodes = config.max_inodes;
2930 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002931
Greg Thelen5f001102013-02-22 16:36:01 -08002932 /*
2933 * Preserve previous mempolicy unless mpol remount option was specified.
2934 */
2935 if (config.mpol) {
2936 mpol_put(sbinfo->mpol);
2937 sbinfo->mpol = config.mpol; /* transfers initial ref */
2938 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002939out:
2940 spin_unlock(&sbinfo->stat_lock);
2941 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002943
Al Viro34c80b12011-12-08 21:32:45 -05002944static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002945{
Al Viro34c80b12011-12-08 21:32:45 -05002946 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002947
2948 if (sbinfo->max_blocks != shmem_default_max_blocks())
2949 seq_printf(seq, ",size=%luk",
2950 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2951 if (sbinfo->max_inodes != shmem_default_max_inodes())
2952 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2953 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002954 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002955 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2956 seq_printf(seq, ",uid=%u",
2957 from_kuid_munged(&init_user_ns, sbinfo->uid));
2958 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2959 seq_printf(seq, ",gid=%u",
2960 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002961 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002962 return 0;
2963}
David Herrmann9183df22014-08-08 14:25:29 -07002964
2965#define MFD_NAME_PREFIX "memfd:"
2966#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2967#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2968
2969#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2970
2971SYSCALL_DEFINE2(memfd_create,
2972 const char __user *, uname,
2973 unsigned int, flags)
2974{
2975 struct shmem_inode_info *info;
2976 struct file *file;
2977 int fd, error;
2978 char *name;
2979 long len;
2980
2981 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2982 return -EINVAL;
2983
2984 /* length includes terminating zero */
2985 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2986 if (len <= 0)
2987 return -EFAULT;
2988 if (len > MFD_NAME_MAX_LEN + 1)
2989 return -EINVAL;
2990
2991 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2992 if (!name)
2993 return -ENOMEM;
2994
2995 strcpy(name, MFD_NAME_PREFIX);
2996 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2997 error = -EFAULT;
2998 goto err_name;
2999 }
3000
3001 /* terminating-zero may have changed after strnlen_user() returned */
3002 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3003 error = -EFAULT;
3004 goto err_name;
3005 }
3006
3007 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3008 if (fd < 0) {
3009 error = fd;
3010 goto err_name;
3011 }
3012
3013 file = shmem_file_setup(name, 0, VM_NORESERVE);
3014 if (IS_ERR(file)) {
3015 error = PTR_ERR(file);
3016 goto err_fd;
3017 }
3018 info = SHMEM_I(file_inode(file));
3019 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3020 file->f_flags |= O_RDWR | O_LARGEFILE;
3021 if (flags & MFD_ALLOW_SEALING)
3022 info->seals &= ~F_SEAL_SEAL;
3023
3024 fd_install(fd, file);
3025 kfree(name);
3026 return fd;
3027
3028err_fd:
3029 put_unused_fd(fd);
3030err_name:
3031 kfree(name);
3032 return error;
3033}
3034
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003035#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
3037static void shmem_put_super(struct super_block *sb)
3038{
Hugh Dickins602586a2010-08-17 15:23:56 -07003039 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3040
3041 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003042 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003043 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 sb->s_fs_info = NULL;
3045}
3046
Kay Sievers2b2af542009-04-30 15:23:42 +02003047int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048{
3049 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003050 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003051 int err = -ENOMEM;
3052
3053 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003054 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003055 L1_CACHE_BYTES), GFP_KERNEL);
3056 if (!sbinfo)
3057 return -ENOMEM;
3058
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003059 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003060 sbinfo->uid = current_fsuid();
3061 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003062 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003064#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 /*
3066 * Per default we only allow half of the physical ram per
3067 * tmpfs instance, limiting inodes to one per page of lowmem;
3068 * but the internal instance is left unlimited.
3069 */
Al Viroca4e0512013-08-31 12:57:10 -04003070 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003071 sbinfo->max_blocks = shmem_default_max_blocks();
3072 sbinfo->max_inodes = shmem_default_max_inodes();
3073 if (shmem_parse_options(data, sbinfo, false)) {
3074 err = -EINVAL;
3075 goto failed;
3076 }
Al Viroca4e0512013-08-31 12:57:10 -04003077 } else {
3078 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 }
David M. Grimes91828a42006-10-17 00:09:45 -07003080 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003081 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082#else
3083 sb->s_flags |= MS_NOUSER;
3084#endif
3085
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003086 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003087 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003088 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003089 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003090
Hugh Dickins285b2c42011-08-03 16:21:20 -07003091 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 sb->s_blocksize = PAGE_CACHE_SIZE;
3093 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
3094 sb->s_magic = TMPFS_MAGIC;
3095 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003096 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003097#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003098 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003099#endif
3100#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003101 sb->s_flags |= MS_POSIXACL;
3102#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003103
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003104 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 if (!inode)
3106 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003107 inode->i_uid = sbinfo->uid;
3108 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003109 sb->s_root = d_make_root(inode);
3110 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003111 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 return 0;
3113
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114failed:
3115 shmem_put_super(sb);
3116 return err;
3117}
3118
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003119static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120
3121static struct inode *shmem_alloc_inode(struct super_block *sb)
3122{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003123 struct shmem_inode_info *info;
3124 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3125 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003127 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128}
3129
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003130static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003131{
3132 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro3ed47db2016-01-22 18:08:52 -05003133 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003134 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3135}
3136
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137static void shmem_destroy_inode(struct inode *inode)
3138{
Al Viro09208d12011-07-26 03:15:03 -04003139 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003141 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142}
3143
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003144static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003146 struct shmem_inode_info *info = foo;
3147 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148}
3149
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003150static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151{
3152 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3153 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003154 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 return 0;
3156}
3157
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003158static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003160 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161}
3162
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003163static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003165 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003167 .write_begin = shmem_write_begin,
3168 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003170#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003171 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003172#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003173 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174};
3175
Helge Deller15ad7cd2006-12-06 20:40:36 -08003176static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 .mmap = shmem_mmap,
3178#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003179 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003180 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003181 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003182 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003183 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003184 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003185 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186#endif
3187};
3188
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003189static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003190 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003191 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003192#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003193 .setxattr = generic_setxattr,
3194 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003195 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003196 .removexattr = generic_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003197 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003198#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199};
3200
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003201static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202#ifdef CONFIG_TMPFS
3203 .create = shmem_create,
3204 .lookup = simple_lookup,
3205 .link = shmem_link,
3206 .unlink = shmem_unlink,
3207 .symlink = shmem_symlink,
3208 .mkdir = shmem_mkdir,
3209 .rmdir = shmem_rmdir,
3210 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003211 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003212 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003214#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003215 .setxattr = generic_setxattr,
3216 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003217 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003218 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003219#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003220#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003221 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003222 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003223#endif
3224};
3225
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003226static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003227#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003228 .setxattr = generic_setxattr,
3229 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003230 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003231 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003232#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003233#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003234 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003235 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003236#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237};
3238
Hugh Dickins759b9772007-03-05 00:30:28 -08003239static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 .alloc_inode = shmem_alloc_inode,
3241 .destroy_inode = shmem_destroy_inode,
3242#ifdef CONFIG_TMPFS
3243 .statfs = shmem_statfs,
3244 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003245 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246#endif
Al Viro1f895f72010-06-05 19:10:41 -04003247 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 .drop_inode = generic_delete_inode,
3249 .put_super = shmem_put_super,
3250};
3251
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003252static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003253 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003254 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255#ifdef CONFIG_NUMA
3256 .set_policy = shmem_set_policy,
3257 .get_policy = shmem_get_policy,
3258#endif
3259};
3260
Al Viro3c26ff62010-07-25 11:46:36 +04003261static struct dentry *shmem_mount(struct file_system_type *fs_type,
3262 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263{
Al Viro3c26ff62010-07-25 11:46:36 +04003264 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265}
3266
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003267static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 .owner = THIS_MODULE,
3269 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003270 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003272 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003275int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276{
3277 int error;
3278
Rob Landley16203a72013-09-11 14:26:12 -07003279 /* If rootfs called this, don't re-init */
3280 if (shmem_inode_cachep)
3281 return 0;
3282
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003283 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284 if (error)
3285 goto out3;
3286
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003287 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 if (error) {
3289 printk(KERN_ERR "Could not register tmpfs\n");
3290 goto out2;
3291 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003292
Al Viroca4e0512013-08-31 12:57:10 -04003293 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 if (IS_ERR(shm_mnt)) {
3295 error = PTR_ERR(shm_mnt);
3296 printk(KERN_ERR "Could not kern_mount tmpfs\n");
3297 goto out1;
3298 }
3299 return 0;
3300
3301out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003302 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003304 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305out3:
3306 shm_mnt = ERR_PTR(error);
3307 return error;
3308}
Matt Mackall853ac432009-01-06 14:40:20 -08003309
3310#else /* !CONFIG_SHMEM */
3311
3312/*
3313 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3314 *
3315 * This is intended for small system where the benefits of the full
3316 * shmem code (swap-backed and resource-limited) are outweighed by
3317 * their complexity. On systems without swap this code should be
3318 * effectively equivalent, but much lighter weight.
3319 */
3320
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003321static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003322 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003323 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003324 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003325 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003326};
3327
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003328int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003329{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003330 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003331
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003332 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003333 BUG_ON(IS_ERR(shm_mnt));
3334
3335 return 0;
3336}
3337
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003338int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003339{
3340 return 0;
3341}
3342
Hugh Dickins3f96b792009-09-21 17:03:37 -07003343int shmem_lock(struct file *file, int lock, struct user_struct *user)
3344{
3345 return 0;
3346}
3347
Hugh Dickins24513262012-01-20 14:34:21 -08003348void shmem_unlock_mapping(struct address_space *mapping)
3349{
3350}
3351
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003352void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003353{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003354 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003355}
3356EXPORT_SYMBOL_GPL(shmem_truncate_range);
3357
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003358#define shmem_vm_ops generic_file_vm_ops
3359#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003360#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003361#define shmem_acct_size(flags, size) 0
3362#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003363
3364#endif /* CONFIG_SHMEM */
3365
3366/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367
Al Viro34515382013-02-14 22:38:02 -05003368static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003369 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003370};
3371
Eric Parisc7277092013-12-02 11:24:19 +00003372static struct file *__shmem_file_setup(const char *name, loff_t size,
3373 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374{
Al Viro6b4d0b22013-02-14 21:37:26 -05003375 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003377 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003378 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 struct qstr this;
3380
3381 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003382 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
Hugh Dickins285b2c42011-08-03 16:21:20 -07003384 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 return ERR_PTR(-EINVAL);
3386
3387 if (shmem_acct_size(flags, size))
3388 return ERR_PTR(-ENOMEM);
3389
Al Viro6b4d0b22013-02-14 21:37:26 -05003390 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 this.name = name;
3392 this.len = strlen(name);
3393 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003394 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003395 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003396 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003397 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003399 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400
Al Viro6b4d0b22013-02-14 21:37:26 -05003401 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05003402 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003404 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405
Eric Parisc7277092013-12-02 11:24:19 +00003406 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04003407 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003409 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05003410 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3411 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003412 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003413
Al Viro6b4d0b22013-02-14 21:37:26 -05003414 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04003415 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05003416 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003417 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003418
Al Viro6b4d0b22013-02-14 21:37:26 -05003419 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421put_memory:
3422 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003423put_path:
3424 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05003425 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426}
Eric Parisc7277092013-12-02 11:24:19 +00003427
3428/**
3429 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3430 * kernel internal. There will be NO LSM permission checks against the
3431 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07003432 * higher layer. The users are the big_key and shm implementations. LSM
3433 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00003434 * @name: name for dentry (to be seen in /proc/<pid>/maps
3435 * @size: size to be set for the file
3436 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3437 */
3438struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3439{
3440 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3441}
3442
3443/**
3444 * shmem_file_setup - get an unlinked file living in tmpfs
3445 * @name: name for dentry (to be seen in /proc/<pid>/maps
3446 * @size: size to be set for the file
3447 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3448 */
3449struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3450{
3451 return __shmem_file_setup(name, size, flags, 0);
3452}
Keith Packard395e0dd2008-06-20 00:08:06 -07003453EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
Randy Dunlap46711812008-03-19 17:00:41 -07003455/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3458 */
3459int shmem_zero_setup(struct vm_area_struct *vma)
3460{
3461 struct file *file;
3462 loff_t size = vma->vm_end - vma->vm_start;
3463
Hugh Dickins66fc1302015-06-14 09:48:09 -07003464 /*
3465 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3466 * between XFS directory reading and selinux: since this file is only
3467 * accessible to the user through its mapping, use S_PRIVATE flag to
3468 * bypass file security, in the same way as shmem_kernel_file_setup().
3469 */
3470 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 if (IS_ERR(file))
3472 return PTR_ERR(file);
3473
3474 if (vma->vm_file)
3475 fput(vma->vm_file);
3476 vma->vm_file = file;
3477 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 return 0;
3479}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003480
3481/**
3482 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3483 * @mapping: the page's address_space
3484 * @index: the page index
3485 * @gfp: the page allocator flags to use if allocating
3486 *
3487 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3488 * with any new page allocations done using the specified allocation flags.
3489 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3490 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3491 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3492 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003493 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3494 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003495 */
3496struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3497 pgoff_t index, gfp_t gfp)
3498{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003499#ifdef CONFIG_SHMEM
3500 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003501 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003502 int error;
3503
3504 BUG_ON(mapping->a_ops != &shmem_aops);
3505 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3506 if (error)
3507 page = ERR_PTR(error);
3508 else
3509 unlock_page(page);
3510 return page;
3511#else
3512 /*
3513 * The tiny !SHMEM case uses ramfs without swap
3514 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003515 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003516#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003517}
3518EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);