blob: 2f17cb5f00a43f87b7868a2a2192dd911f4501bd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins6922c0c2011-08-03 16:21:25 -07009 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -070011 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 *
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 *
Matt Mackall853ac432009-01-06 14:40:20 -080018 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * This file is released under the GPL.
22 */
23
Matt Mackall853ac432009-01-06 14:40:20 -080024#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
Andrew Morton250297e2013-04-29 15:06:12 -070028#include <linux/ramfs.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070029#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080030#include <linux/file.h>
31#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040032#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080033#include <linux/swap.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070034#include <linux/aio.h>
Matt Mackall853ac432009-01-06 14:40:20 -080035
36static struct vfsmount *shm_mnt;
37
38#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
43 */
44
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070045#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070046#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010047#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080048#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/string.h>
51#include <linux/slab.h>
52#include <linux/backing-dev.h>
53#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070056#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070057#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070058#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070059#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/security.h>
61#include <linux/swapops.h>
62#include <linux/mempolicy.h>
63#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000064#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070065#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070066#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080067#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040068#include <linux/magic.h>
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* Pretend that each entry is of this size in directory's i_size */
80#define BOGO_DIRENT_SIZE 20
81
Hugh Dickins69f07ec2011-08-03 16:21:26 -070082/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
83#define SHORT_SYMLINK_LEN 128
84
Hugh Dickins1aac1402012-05-29 15:06:42 -070085/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070086 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
87 * inode->i_private (with i_mutex making sure that it has only one user at
88 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070089 */
90struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -070091 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -070092 pgoff_t start; /* start of range currently being fallocated */
93 pgoff_t next; /* the next page offset to be fallocated */
94 pgoff_t nr_falloced; /* how many new pages have been fallocated */
95 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
96};
97
Hugh Dickins285b2c42011-08-03 16:21:20 -070098/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 SGP_READ, /* don't exceed i_size, don't allocate page */
101 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800102 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700103 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
104 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105};
106
Andrew Mortonb76db732008-02-08 04:21:49 -0800107#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800108static unsigned long shmem_default_max_blocks(void)
109{
110 return totalram_pages / 2;
111}
112
113static unsigned long shmem_default_max_inodes(void)
114{
115 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
116}
Andrew Mortonb76db732008-02-08 04:21:49 -0800117#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800118
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700119static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
120static int shmem_replace_page(struct page **pagep, gfp_t gfp,
121 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700122static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
123 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
124
125static inline int shmem_getpage(struct inode *inode, pgoff_t index,
126 struct page **pagep, enum sgp_type sgp, int *fault_type)
127{
128 return shmem_getpage_gfp(inode, index, pagep, sgp,
129 mapping_gfp_mask(inode->i_mapping), fault_type);
130}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
133{
134 return sb->s_fs_info;
135}
136
137/*
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
142 */
143static inline int shmem_acct_size(unsigned long flags, loff_t size)
144{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000145 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000146 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
149static inline void shmem_unacct_size(unsigned long flags, loff_t size)
150{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000151 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 vm_unacct_memory(VM_ACCT(size));
153}
154
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700155static inline int shmem_reacct_size(unsigned long flags,
156 loff_t oldsize, loff_t newsize)
157{
158 if (!(flags & VM_NORESERVE)) {
159 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
160 return security_vm_enough_memory_mm(current->mm,
161 VM_ACCT(newsize) - VM_ACCT(oldsize));
162 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
163 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
164 }
165 return 0;
166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/*
169 * ... whereas tmpfs objects are accounted incrementally as
170 * pages are allocated, in order to allow huge sparse files.
171 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
172 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
173 */
174static inline int shmem_acct_block(unsigned long flags)
175{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000176 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000177 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
180static inline void shmem_unacct_blocks(unsigned long flags, long pages)
181{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000182 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
184}
185
Hugh Dickins759b9772007-03-05 00:30:28 -0800186static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700187static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800188static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800189static const struct inode_operations shmem_inode_operations;
190static const struct inode_operations shmem_dir_inode_operations;
191static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400192static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800195static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800197static int shmem_reserve_inode(struct super_block *sb)
198{
199 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
200 if (sbinfo->max_inodes) {
201 spin_lock(&sbinfo->stat_lock);
202 if (!sbinfo->free_inodes) {
203 spin_unlock(&sbinfo->stat_lock);
204 return -ENOSPC;
205 }
206 sbinfo->free_inodes--;
207 spin_unlock(&sbinfo->stat_lock);
208 }
209 return 0;
210}
211
212static void shmem_free_inode(struct super_block *sb)
213{
214 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
215 if (sbinfo->max_inodes) {
216 spin_lock(&sbinfo->stat_lock);
217 sbinfo->free_inodes++;
218 spin_unlock(&sbinfo->stat_lock);
219 }
220}
221
Randy Dunlap46711812008-03-19 17:00:41 -0700222/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700223 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 * @inode: inode to recalc
225 *
226 * We have to calculate the free blocks since the mm can drop
227 * undirtied hole pages behind our back.
228 *
229 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
230 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
231 *
232 * It has to be called with the spinlock held.
233 */
234static void shmem_recalc_inode(struct inode *inode)
235{
236 struct shmem_inode_info *info = SHMEM_I(inode);
237 long freed;
238
239 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
240 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
242 if (sbinfo->max_blocks)
243 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700245 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
248}
249
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700250/*
251 * Replace item expected in radix tree by a new item, while holding tree lock.
252 */
253static int shmem_radix_tree_replace(struct address_space *mapping,
254 pgoff_t index, void *expected, void *replacement)
255{
256 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700257 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700258
259 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700260 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700261 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700262 if (!pslot)
263 return -ENOENT;
264 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700265 if (item != expected)
266 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700267 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700268 return 0;
269}
270
271/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700272 * Sometimes, before we decide whether to proceed or to fail, we must check
273 * that an entry was not already brought back from swap by a racing thread.
274 *
275 * Checking page is not enough: by the time a SwapCache page is locked, it
276 * might be reused, and again be SwapCache, using the same swap as before.
277 */
278static bool shmem_confirm_swap(struct address_space *mapping,
279 pgoff_t index, swp_entry_t swap)
280{
281 void *item;
282
283 rcu_read_lock();
284 item = radix_tree_lookup(&mapping->page_tree, index);
285 rcu_read_unlock();
286 return item == swp_to_radix_entry(swap);
287}
288
289/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700290 * Like add_to_page_cache_locked, but error if expected item has gone.
291 */
292static int shmem_add_to_page_cache(struct page *page,
293 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700294 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700295{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700296 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700297
Sasha Levin309381fea2014-01-23 15:52:54 -0800298 VM_BUG_ON_PAGE(!PageLocked(page), page);
299 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700300
Hugh Dickinsb065b432012-07-11 14:02:48 -0700301 page_cache_get(page);
302 page->mapping = mapping;
303 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700304
Hugh Dickinsb065b432012-07-11 14:02:48 -0700305 spin_lock_irq(&mapping->tree_lock);
306 if (!expected)
307 error = radix_tree_insert(&mapping->page_tree, index, page);
308 else
309 error = shmem_radix_tree_replace(mapping, index, expected,
310 page);
311 if (!error) {
312 mapping->nrpages++;
313 __inc_zone_page_state(page, NR_FILE_PAGES);
314 __inc_zone_page_state(page, NR_SHMEM);
315 spin_unlock_irq(&mapping->tree_lock);
316 } else {
317 page->mapping = NULL;
318 spin_unlock_irq(&mapping->tree_lock);
319 page_cache_release(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700320 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700321 return error;
322}
323
324/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700325 * Like delete_from_page_cache, but substitutes swap for page.
326 */
327static void shmem_delete_from_page_cache(struct page *page, void *radswap)
328{
329 struct address_space *mapping = page->mapping;
330 int error;
331
332 spin_lock_irq(&mapping->tree_lock);
333 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
334 page->mapping = NULL;
335 mapping->nrpages--;
336 __dec_zone_page_state(page, NR_FILE_PAGES);
337 __dec_zone_page_state(page, NR_SHMEM);
338 spin_unlock_irq(&mapping->tree_lock);
339 page_cache_release(page);
340 BUG_ON(error);
341}
342
343/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700344 * Remove swap entry from radix tree, free the swap and its page cache.
345 */
346static int shmem_free_swap(struct address_space *mapping,
347 pgoff_t index, void *radswap)
348{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700349 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700350
351 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700352 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700353 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700354 if (old != radswap)
355 return -ENOENT;
356 free_swap_and_cache(radix_to_swp_entry(radswap));
357 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700358}
359
360/*
Hugh Dickins24513262012-01-20 14:34:21 -0800361 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
362 */
363void shmem_unlock_mapping(struct address_space *mapping)
364{
365 struct pagevec pvec;
366 pgoff_t indices[PAGEVEC_SIZE];
367 pgoff_t index = 0;
368
369 pagevec_init(&pvec, 0);
370 /*
371 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
372 */
373 while (!mapping_unevictable(mapping)) {
374 /*
375 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
376 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
377 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700378 pvec.nr = find_get_entries(mapping, index,
379 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800380 if (!pvec.nr)
381 break;
382 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700383 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800384 check_move_unevictable_pages(pvec.pages, pvec.nr);
385 pagevec_release(&pvec);
386 cond_resched();
387 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700388}
389
390/*
391 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700392 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700393 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700394static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
395 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700397 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700399 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700400 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
401 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
402 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700403 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700404 pgoff_t indices[PAGEVEC_SIZE];
405 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700406 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700407 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700409 if (lend == -1)
410 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700411
412 pagevec_init(&pvec, 0);
413 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700414 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700415 pvec.nr = find_get_entries(mapping, index,
416 min(end - index, (pgoff_t)PAGEVEC_SIZE),
417 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700418 if (!pvec.nr)
419 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700420 for (i = 0; i < pagevec_count(&pvec); i++) {
421 struct page *page = pvec.pages[i];
422
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700423 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700424 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700425 break;
426
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700427 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700428 if (unfalloc)
429 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700430 nr_swaps_freed += !shmem_free_swap(mapping,
431 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700432 continue;
433 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700434
435 if (!trylock_page(page))
436 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700437 if (!unfalloc || !PageUptodate(page)) {
438 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800439 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700440 truncate_inode_page(mapping, page);
441 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700442 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700443 unlock_page(page);
444 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700445 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800446 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700447 cond_resched();
448 index++;
449 }
450
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700451 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700452 struct page *page = NULL;
453 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
454 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700455 unsigned int top = PAGE_CACHE_SIZE;
456 if (start > end) {
457 top = partial_end;
458 partial_end = 0;
459 }
460 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700461 set_page_dirty(page);
462 unlock_page(page);
463 page_cache_release(page);
464 }
465 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700466 if (partial_end) {
467 struct page *page = NULL;
468 shmem_getpage(inode, end, &page, SGP_READ, NULL);
469 if (page) {
470 zero_user_segment(page, 0, partial_end);
471 set_page_dirty(page);
472 unlock_page(page);
473 page_cache_release(page);
474 }
475 }
476 if (start >= end)
477 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700478
479 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700480 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700481 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700482
483 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700484 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700485 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700486 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700487 /* If all gone or hole-punch or unfalloc, we're done */
488 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700489 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700490 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700491 index = start;
492 continue;
493 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700494 for (i = 0; i < pagevec_count(&pvec); i++) {
495 struct page *page = pvec.pages[i];
496
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700497 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700498 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700499 break;
500
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700501 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700502 if (unfalloc)
503 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700504 if (shmem_free_swap(mapping, index, page)) {
505 /* Swap was replaced by page: retry */
506 index--;
507 break;
508 }
509 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700510 continue;
511 }
512
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700513 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700514 if (!unfalloc || !PageUptodate(page)) {
515 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800516 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700517 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700518 } else {
519 /* Page was replaced by swap: retry */
520 unlock_page(page);
521 index--;
522 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700523 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700524 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700525 unlock_page(page);
526 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700527 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800528 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700529 index++;
530 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700533 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 shmem_recalc_inode(inode);
535 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700536}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700538void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
539{
540 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700541 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700543EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Hugh Dickins94c1e622011-06-27 16:18:03 -0700545static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct inode *inode = dentry->d_inode;
David Herrmann40e041a2014-08-08 14:25:27 -0700548 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 int error;
550
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200551 error = inode_change_ok(inode, attr);
552 if (error)
553 return error;
554
Hugh Dickins94c1e622011-06-27 16:18:03 -0700555 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
556 loff_t oldsize = inode->i_size;
557 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000558
David Herrmann40e041a2014-08-08 14:25:27 -0700559 /* protected by i_mutex */
560 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
561 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
562 return -EPERM;
563
Hugh Dickins94c1e622011-06-27 16:18:03 -0700564 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700565 error = shmem_reacct_size(SHMEM_I(inode)->flags,
566 oldsize, newsize);
567 if (error)
568 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700569 i_size_write(inode, newsize);
570 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
571 }
572 if (newsize < oldsize) {
573 loff_t holebegin = round_up(newsize, PAGE_SIZE);
574 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
575 shmem_truncate_range(inode, newsize, (loff_t)-1);
576 /* unmap again to remove racily COWed private pages */
577 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200581 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200582 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -0800583 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return error;
585}
586
Al Viro1f895f72010-06-05 19:10:41 -0400587static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 struct shmem_inode_info *info = SHMEM_I(inode);
590
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000591 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 shmem_unacct_size(info->flags, inode->i_size);
593 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000594 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800596 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800598 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700600 } else
601 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700602
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400603 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800604 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800605 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200606 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607}
608
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700609/*
610 * If swap found in inode, free it and move page from swapcache to filecache.
611 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700612static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700613 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700615 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700616 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700617 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700618 gfp_t gfp;
619 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700621 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700622 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700623 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -0700624 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800625
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800626 /*
627 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400628 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800629 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700630 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800631 */
632 if (shmem_swaplist.next != &info->swaplist)
633 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800634
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700635 gfp = mapping_gfp_mask(mapping);
636 if (shmem_should_replace_page(*pagep, gfp)) {
637 mutex_unlock(&shmem_swaplist_mutex);
638 error = shmem_replace_page(pagep, gfp, info, index);
639 mutex_lock(&shmem_swaplist_mutex);
640 /*
641 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700642 * allocation, but the inode might have been freed while we
643 * dropped it: although a racing shmem_evict_inode() cannot
644 * complete without emptying the radix_tree, our page lock
645 * on this swapcache page is not enough to prevent that -
646 * free_swap_and_cache() of our swap entry will only
647 * trylock_page(), removing swap from radix_tree whatever.
648 *
649 * We must not proceed to shmem_add_to_page_cache() if the
650 * inode has been freed, but of course we cannot rely on
651 * inode or mapping or info to check that. However, we can
652 * safely check if our swap entry is still in use (and here
653 * it can't have got reused for another page): if it's still
654 * in use, then the inode cannot have been freed yet, and we
655 * can safely proceed (if it's no longer in use, that tells
656 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700657 */
658 if (!page_swapcount(*pagep))
659 error = -ENOENT;
660 }
661
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800662 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700663 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
664 * but also to hold up shmem_evict_inode(): so inode cannot be freed
665 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800666 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700667 if (!error)
668 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700669 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700670 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700671 /*
672 * Truncation and eviction use free_swap_and_cache(), which
673 * only does trylock page: if we raced, best clean up here.
674 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700675 delete_from_swap_cache(*pagep);
676 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700677 if (!error) {
678 spin_lock(&info->lock);
679 info->swapped--;
680 spin_unlock(&info->lock);
681 swap_free(swap);
682 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800684 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
687/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700688 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700690int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700692 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -0700694 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700695 int error = 0;
696
697 /*
698 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700699 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700700 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700701 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700702 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700703
704 /*
705 * Charge page using GFP_KERNEL while we can wait, before taking
706 * the shmem_swaplist_mutex which might hold up shmem_writepage().
707 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700708 */
Johannes Weiner00501b52014-08-08 14:19:20 -0700709 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg);
Hugh Dickins778dd892011-05-11 15:13:37 -0700710 if (error)
711 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700712 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -0700713 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800715 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700716 list_for_each_safe(this, next, &shmem_swaplist) {
717 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700718 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -0700719 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700720 else
721 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800722 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -0700723 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -0700724 break;
Johannes Weiner00501b52014-08-08 14:19:20 -0700725 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800727 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700728
Johannes Weiner00501b52014-08-08 14:19:20 -0700729 if (error) {
730 if (error != -ENOMEM)
731 error = 0;
732 mem_cgroup_cancel_charge(page, memcg);
733 } else
734 mem_cgroup_commit_charge(page, memcg, true);
Hugh Dickins778dd892011-05-11 15:13:37 -0700735out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800736 unlock_page(page);
737 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700738 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
741/*
742 * Move the page from the page cache to the swap cache.
743 */
744static int shmem_writepage(struct page *page, struct writeback_control *wbc)
745{
746 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700749 swp_entry_t swap;
750 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 mapping = page->mapping;
754 index = page->index;
755 inode = mapping->host;
756 info = SHMEM_I(inode);
757 if (info->flags & VM_LOCKED)
758 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800759 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 goto redirty;
761
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800762 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100763 * Our capabilities prevent regular writeback or sync from ever calling
764 * shmem_writepage; but a stacking filesystem might use ->writepage of
765 * its underlying filesystem, in which case tmpfs should write out to
766 * swap only in response to memory pressure, and not for the writeback
767 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800768 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700769 if (!wbc->for_reclaim) {
770 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
771 goto redirty;
772 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700773
774 /*
775 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
776 * value into swapfile.c, the only way we can correctly account for a
777 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700778 *
779 * That's okay for a page already fallocated earlier, but if we have
780 * not yet completed the fallocation, then (a) we want to keep track
781 * of this page in case we have to undo it, and (b) it may not be a
782 * good idea to continue anyway, once we're pushing into swap. So
783 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700784 */
785 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700786 if (inode->i_private) {
787 struct shmem_falloc *shmem_falloc;
788 spin_lock(&inode->i_lock);
789 shmem_falloc = inode->i_private;
790 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -0700791 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -0700792 index >= shmem_falloc->start &&
793 index < shmem_falloc->next)
794 shmem_falloc->nr_unswapped++;
795 else
796 shmem_falloc = NULL;
797 spin_unlock(&inode->i_lock);
798 if (shmem_falloc)
799 goto redirty;
800 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700801 clear_highpage(page);
802 flush_dcache_page(page);
803 SetPageUptodate(page);
804 }
805
Hugh Dickins48f170f2011-07-25 17:12:37 -0700806 swap = get_swap_page();
807 if (!swap.val)
808 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800809
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700810 /*
811 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700812 * if it's not already there. Do it now before the page is
813 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700814 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700815 * we've incremented swapped, because shmem_unuse_inode() will
816 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700817 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700818 mutex_lock(&shmem_swaplist_mutex);
819 if (list_empty(&info->swaplist))
820 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700821
Hugh Dickins48f170f2011-07-25 17:12:37 -0700822 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800823 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700824 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
825
826 spin_lock(&info->lock);
827 info->swapped++;
828 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700829 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700830
831 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800832 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700833 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 return 0;
835 }
836
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700837 mutex_unlock(&shmem_swaplist_mutex);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700838 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839redirty:
840 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800841 if (wbc->for_reclaim)
842 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
843 unlock_page(page);
844 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845}
846
847#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800848#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700849static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800850{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700851 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800852
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700853 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700854 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800855
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800856 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800857
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700858 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800859}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700860
861static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
862{
863 struct mempolicy *mpol = NULL;
864 if (sbinfo->mpol) {
865 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
866 mpol = sbinfo->mpol;
867 mpol_get(mpol);
868 spin_unlock(&sbinfo->stat_lock);
869 }
870 return mpol;
871}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800872#endif /* CONFIG_TMPFS */
873
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700874static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
875 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800878 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800880 /* Create a pseudo vma that just contains the policy */
881 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700882 /* Bias interleave by inode number to distribute better across nodes */
883 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800884 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700885 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700886
Mel Gorman18a2f372012-12-05 14:01:41 -0800887 page = swapin_readahead(swap, gfp, &pvma, 0);
888
889 /* Drop reference taken by mpol_shared_policy_lookup() */
890 mpol_cond_put(pvma.vm_policy);
891
892 return page;
893}
894
895static struct page *shmem_alloc_page(gfp_t gfp,
896 struct shmem_inode_info *info, pgoff_t index)
897{
898 struct vm_area_struct pvma;
899 struct page *page;
900
901 /* Create a pseudo vma that just contains the policy */
902 pvma.vm_start = 0;
903 /* Bias interleave by inode number to distribute better across nodes */
904 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
905 pvma.vm_ops = NULL;
906 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
907
908 page = alloc_page_vma(gfp, &pvma, 0);
909
910 /* Drop reference taken by mpol_shared_policy_lookup() */
911 mpol_cond_put(pvma.vm_policy);
912
913 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800915#else /* !CONFIG_NUMA */
916#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700917static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800918{
919}
920#endif /* CONFIG_TMPFS */
921
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700922static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
923 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700925 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
927
Hugh Dickins02098fe2008-02-04 22:28:42 -0800928static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700929 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000931 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800933#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700935#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
936static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
937{
938 return NULL;
939}
940#endif
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700943 * When a page is moved from swapcache to shmem filecache (either by the
944 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
945 * shmem_unuse_inode()), it may have been read in earlier from swap, in
946 * ignorance of the mapping it belongs to. If that mapping has special
947 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
948 * we may need to copy to a suitable page before moving to filecache.
949 *
950 * In a future release, this may well be extended to respect cpuset and
951 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
952 * but for now it is a simple matter of zone.
953 */
954static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
955{
956 return page_zonenum(page) > gfp_zone(gfp);
957}
958
959static int shmem_replace_page(struct page **pagep, gfp_t gfp,
960 struct shmem_inode_info *info, pgoff_t index)
961{
962 struct page *oldpage, *newpage;
963 struct address_space *swap_mapping;
964 pgoff_t swap_index;
965 int error;
966
967 oldpage = *pagep;
968 swap_index = page_private(oldpage);
969 swap_mapping = page_mapping(oldpage);
970
971 /*
972 * We have arrived here because our zones are constrained, so don't
973 * limit chance of success by further cpuset and node constraints.
974 */
975 gfp &= ~GFP_CONSTRAINT_MASK;
976 newpage = shmem_alloc_page(gfp, info, index);
977 if (!newpage)
978 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700979
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700980 page_cache_get(newpage);
981 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -0700982 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700983
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700984 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700985 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700986 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700987 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700988 SetPageSwapCache(newpage);
989
990 /*
991 * Our caller will very soon move newpage out of swapcache, but it's
992 * a nice clean interface for us to replace oldpage by newpage there.
993 */
994 spin_lock_irq(&swap_mapping->tree_lock);
995 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
996 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -0700997 if (!error) {
998 __inc_zone_page_state(newpage, NR_FILE_PAGES);
999 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1000 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001001 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001002
Hugh Dickins0142ef62012-06-07 14:21:09 -07001003 if (unlikely(error)) {
1004 /*
1005 * Is this possible? I think not, now that our callers check
1006 * both PageSwapCache and page_private after getting page lock;
1007 * but be defensive. Reverse old to newpage for clear and free.
1008 */
1009 oldpage = newpage;
1010 } else {
Michal Hockof5e03a42015-02-05 12:25:14 -08001011 mem_cgroup_migrate(oldpage, newpage, true);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001012 lru_cache_add_anon(newpage);
1013 *pagep = newpage;
1014 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001015
1016 ClearPageSwapCache(oldpage);
1017 set_page_private(oldpage, 0);
1018
1019 unlock_page(oldpage);
1020 page_cache_release(oldpage);
1021 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001022 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001023}
1024
1025/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001026 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 *
1028 * If we allocate a new one we do not mark it dirty. That's up to the
1029 * vm. If we swap it in we mark it dirty since we also free the swap
1030 * entry since a page cannot live in both the swap and page cache
1031 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001032static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001033 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
1035 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001036 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 struct shmem_sb_info *sbinfo;
Johannes Weiner00501b52014-08-08 14:19:20 -07001038 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001039 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 swp_entry_t swap;
1041 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001042 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001043 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001045 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001048 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001049 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001050 if (radix_tree_exceptional_entry(page)) {
1051 swap = radix_to_swp_entry(page);
1052 page = NULL;
1053 }
1054
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001055 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001056 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1057 error = -EINVAL;
1058 goto failed;
1059 }
1060
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001061 if (page && sgp == SGP_WRITE)
1062 mark_page_accessed(page);
1063
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001064 /* fallocated page? */
1065 if (page && !PageUptodate(page)) {
1066 if (sgp != SGP_READ)
1067 goto clear;
1068 unlock_page(page);
1069 page_cache_release(page);
1070 page = NULL;
1071 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001072 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001073 *pagep = page;
1074 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001075 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001078 * Fast cache lookup did not find it:
1079 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001081 info = SHMEM_I(inode);
1082 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001083
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 if (swap.val) {
1085 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001086 page = lookup_swap_cache(swap);
1087 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001088 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001089 if (fault_type)
1090 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001091 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001092 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001093 error = -ENOMEM;
1094 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097
1098 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001099 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001100 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001101 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001102 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001103 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001104 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001105 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001107 goto failed;
1108 }
1109 wait_on_page_writeback(page);
1110
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001111 if (shmem_should_replace_page(page, gfp)) {
1112 error = shmem_replace_page(&page, gfp, info, index);
1113 if (error)
1114 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 }
1116
Johannes Weiner00501b52014-08-08 14:19:20 -07001117 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001118 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001119 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001120 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001121 /*
1122 * We already confirmed swap under page lock, and make
1123 * no memory allocation here, so usually no possibility
1124 * of error; but free_swap_and_cache() only trylocks a
1125 * page, so it is just possible that the entry has been
1126 * truncated or holepunched since swap was confirmed.
1127 * shmem_undo_range() will have done some of the
1128 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001129 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001130 * Reset swap.val? No, leave it so "failed" goes back to
1131 * "repeat": reading a hole and writing should succeed.
1132 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001133 if (error) {
1134 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001135 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001136 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001137 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001138 if (error)
1139 goto failed;
1140
Johannes Weiner00501b52014-08-08 14:19:20 -07001141 mem_cgroup_commit_charge(page, memcg, true);
1142
Hugh Dickins54af60422011-08-03 16:21:24 -07001143 spin_lock(&info->lock);
1144 info->swapped--;
1145 shmem_recalc_inode(inode);
1146 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001147
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001148 if (sgp == SGP_WRITE)
1149 mark_page_accessed(page);
1150
Hugh Dickins27ab7002011-07-25 17:12:36 -07001151 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001152 set_page_dirty(page);
1153 swap_free(swap);
1154
Hugh Dickins54af60422011-08-03 16:21:24 -07001155 } else {
1156 if (shmem_acct_block(info->flags)) {
1157 error = -ENOSPC;
1158 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001160 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001161 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001162 sbinfo->max_blocks) >= 0) {
1163 error = -ENOSPC;
1164 goto unacct;
1165 }
Tim Chen7e496292010-08-09 17:19:05 -07001166 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Hugh Dickins54af60422011-08-03 16:21:24 -07001169 page = shmem_alloc_page(gfp, info, index);
1170 if (!page) {
1171 error = -ENOMEM;
1172 goto decused;
1173 }
1174
Mel Gorman07a42782014-06-04 16:10:24 -07001175 __SetPageSwapBacked(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001176 __set_page_locked(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001177 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001178 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001179
Johannes Weiner00501b52014-08-08 14:19:20 -07001180 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg);
Hugh Dickins54af60422011-08-03 16:21:24 -07001181 if (error)
1182 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001183 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001184 if (!error) {
1185 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001186 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001187 radix_tree_preload_end();
1188 }
1189 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -07001190 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001191 goto decused;
1192 }
Johannes Weiner00501b52014-08-08 14:19:20 -07001193 mem_cgroup_commit_charge(page, memcg, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001194 lru_cache_add_anon(page);
1195
1196 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001198 inode->i_blocks += BLOCKS_PER_PAGE;
1199 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001201 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001202
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001203 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001204 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1205 */
1206 if (sgp == SGP_FALLOC)
1207 sgp = SGP_WRITE;
1208clear:
1209 /*
1210 * Let SGP_WRITE caller clear ends if write does not fill page;
1211 * but SGP_FALLOC on a page fallocated earlier must initialize
1212 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001213 */
1214 if (sgp != SGP_WRITE) {
1215 clear_highpage(page);
1216 flush_dcache_page(page);
1217 SetPageUptodate(page);
1218 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001219 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001220 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001221 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001222
Hugh Dickins54af60422011-08-03 16:21:24 -07001223 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001224 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001225 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1226 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001227 if (alloced)
1228 goto trunc;
1229 else
1230 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001231 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001232 *pagep = page;
1233 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001234
Nick Piggind0217ac2007-07-19 01:47:03 -07001235 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001236 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001238trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001239 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001240 ClearPageDirty(page);
1241 delete_from_page_cache(page);
1242 spin_lock(&info->lock);
1243 info->alloced--;
1244 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001246decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001247 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001248 if (sbinfo->max_blocks)
1249 percpu_counter_add(&sbinfo->used_blocks, -1);
1250unacct:
1251 shmem_unacct_blocks(info->flags, 1);
1252failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001253 if (swap.val && error != -EINVAL &&
1254 !shmem_confirm_swap(mapping, index, swap))
1255 error = -EEXIST;
1256unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001257 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001258 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001259 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001260 }
1261 if (error == -ENOSPC && !once++) {
1262 info = SHMEM_I(inode);
1263 spin_lock(&info->lock);
1264 shmem_recalc_inode(inode);
1265 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001266 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001267 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001268 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001269 goto repeat;
1270 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271}
1272
1273static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1274{
Al Viro496ad9a2013-01-23 17:07:38 -05001275 struct inode *inode = file_inode(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001277 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001279 /*
1280 * Trinity finds that probing a hole which tmpfs is punching can
1281 * prevent the hole-punch from ever completing: which in turn
1282 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001283 * faulting pages into the hole while it's being punched. Although
1284 * shmem_undo_range() does remove the additions, it may be unable to
1285 * keep up, as each new page needs its own unmap_mapping_range() call,
1286 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1287 *
1288 * It does not matter if we sometimes reach this check just before the
1289 * hole-punch begins, so that one fault then races with the punch:
1290 * we just need to make racing faults a rare case.
1291 *
1292 * The implementation below would be much simpler if we just used a
1293 * standard mutex or completion: but we cannot take i_mutex in fault,
1294 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001295 */
1296 if (unlikely(inode->i_private)) {
1297 struct shmem_falloc *shmem_falloc;
1298
1299 spin_lock(&inode->i_lock);
1300 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001301 if (shmem_falloc &&
1302 shmem_falloc->waitq &&
1303 vmf->pgoff >= shmem_falloc->start &&
1304 vmf->pgoff < shmem_falloc->next) {
1305 wait_queue_head_t *shmem_falloc_waitq;
1306 DEFINE_WAIT(shmem_fault_wait);
1307
1308 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001309 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1310 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001311 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001312 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001313 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001314 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001315
1316 shmem_falloc_waitq = shmem_falloc->waitq;
1317 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1318 TASK_UNINTERRUPTIBLE);
1319 spin_unlock(&inode->i_lock);
1320 schedule();
1321
1322 /*
1323 * shmem_falloc_waitq points into the shmem_fallocate()
1324 * stack of the hole-punching task: shmem_falloc_waitq
1325 * is usually invalid by the time we reach here, but
1326 * finish_wait() does not dereference it in that case;
1327 * though i_lock needed lest racing with wake_up_all().
1328 */
1329 spin_lock(&inode->i_lock);
1330 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1331 spin_unlock(&inode->i_lock);
1332 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001333 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001334 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001335 }
1336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1338 if (error)
1339 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001340
Ying Han456f9982011-05-26 16:25:38 -07001341 if (ret & VM_FAULT_MAJOR) {
1342 count_vm_event(PGMAJFAULT);
1343 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1344 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001345 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346}
1347
1348#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001349static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350{
Al Viro496ad9a2013-01-23 17:07:38 -05001351 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001352 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001355static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1356 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
Al Viro496ad9a2013-01-23 17:07:38 -05001358 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001359 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001361 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1362 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363}
1364#endif
1365
1366int shmem_lock(struct file *file, int lock, struct user_struct *user)
1367{
Al Viro496ad9a2013-01-23 17:07:38 -05001368 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 struct shmem_inode_info *info = SHMEM_I(inode);
1370 int retval = -ENOMEM;
1371
1372 spin_lock(&info->lock);
1373 if (lock && !(info->flags & VM_LOCKED)) {
1374 if (!user_shm_lock(inode->i_size, user))
1375 goto out_nomem;
1376 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001377 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
1379 if (!lock && (info->flags & VM_LOCKED) && user) {
1380 user_shm_unlock(inode->i_size, user);
1381 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001382 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
1384 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386out_nomem:
1387 spin_unlock(&info->lock);
1388 return retval;
1389}
1390
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001391static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 file_accessed(file);
1394 vma->vm_ops = &shmem_vm_ops;
1395 return 0;
1396}
1397
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001398static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001399 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 struct inode *inode;
1402 struct shmem_inode_info *info;
1403 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1404
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001405 if (shmem_reserve_inode(sb))
1406 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
1408 inode = new_inode(sb);
1409 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001410 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001411 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001414 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 info = SHMEM_I(inode);
1416 memset(info, 0, (char *)inode - (char *)info);
1417 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07001418 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001419 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001421 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001422 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 switch (mode & S_IFMT) {
1425 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001426 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 init_special_inode(inode, mode, dev);
1428 break;
1429 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001430 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 inode->i_op = &shmem_inode_operations;
1432 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001433 mpol_shared_policy_init(&info->policy,
1434 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 break;
1436 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001437 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 /* Some things misbehave if size == 0 on a directory */
1439 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1440 inode->i_op = &shmem_dir_inode_operations;
1441 inode->i_fop = &simple_dir_operations;
1442 break;
1443 case S_IFLNK:
1444 /*
1445 * Must not load anything in the rbtree,
1446 * mpol_free_shared_policy will not be called.
1447 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001448 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 break;
1450 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001451 } else
1452 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 return inode;
1454}
1455
Johannes Weiner0cd61442014-04-03 14:47:46 -07001456bool shmem_mapping(struct address_space *mapping)
1457{
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001458 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001459}
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001462static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001463static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001465#ifdef CONFIG_TMPFS_XATTR
1466static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1467#else
1468#define shmem_initxattrs NULL
1469#endif
1470
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001472shmem_write_begin(struct file *file, struct address_space *mapping,
1473 loff_t pos, unsigned len, unsigned flags,
1474 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475{
Nick Piggin800d15a2007-10-16 01:25:03 -07001476 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07001477 struct shmem_inode_info *info = SHMEM_I(inode);
Nick Piggin800d15a2007-10-16 01:25:03 -07001478 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07001479
1480 /* i_mutex is held by caller */
1481 if (unlikely(info->seals)) {
1482 if (info->seals & F_SEAL_WRITE)
1483 return -EPERM;
1484 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1485 return -EPERM;
1486 }
1487
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001488 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
Nick Piggin800d15a2007-10-16 01:25:03 -07001489}
1490
1491static int
1492shmem_write_end(struct file *file, struct address_space *mapping,
1493 loff_t pos, unsigned len, unsigned copied,
1494 struct page *page, void *fsdata)
1495{
1496 struct inode *inode = mapping->host;
1497
Hugh Dickinsd3602442008-02-04 22:28:44 -08001498 if (pos + copied > inode->i_size)
1499 i_size_write(inode, pos + copied);
1500
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001501 if (!PageUptodate(page)) {
1502 if (copied < PAGE_CACHE_SIZE) {
1503 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1504 zero_user_segments(page, 0, from,
1505 from + copied, PAGE_CACHE_SIZE);
1506 }
1507 SetPageUptodate(page);
1508 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001509 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001510 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001511 page_cache_release(page);
1512
Nick Piggin800d15a2007-10-16 01:25:03 -07001513 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514}
1515
Al Viro2ba5bbe2014-04-02 20:00:02 -04001516static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
Al Viro6e58e792014-02-03 17:07:03 -05001518 struct file *file = iocb->ki_filp;
1519 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001521 pgoff_t index;
1522 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001523 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02001524 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05001525 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05001526 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001527
1528 /*
1529 * Might this read be for a stacking filesystem? Then when reading
1530 * holes of a sparse file, we actually need to allocate those pages,
1531 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1532 */
Al Viro777eda22014-12-17 04:46:46 -05001533 if (!iter_is_iovec(to))
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001534 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536 index = *ppos >> PAGE_CACHE_SHIFT;
1537 offset = *ppos & ~PAGE_CACHE_MASK;
1538
1539 for (;;) {
1540 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001541 pgoff_t end_index;
1542 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 loff_t i_size = i_size_read(inode);
1544
1545 end_index = i_size >> PAGE_CACHE_SHIFT;
1546 if (index > end_index)
1547 break;
1548 if (index == end_index) {
1549 nr = i_size & ~PAGE_CACHE_MASK;
1550 if (nr <= offset)
1551 break;
1552 }
1553
Al Viro6e58e792014-02-03 17:07:03 -05001554 error = shmem_getpage(inode, index, &page, sgp, NULL);
1555 if (error) {
1556 if (error == -EINVAL)
1557 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 break;
1559 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001560 if (page)
1561 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
1563 /*
1564 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001565 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 */
1567 nr = PAGE_CACHE_SIZE;
1568 i_size = i_size_read(inode);
1569 end_index = i_size >> PAGE_CACHE_SHIFT;
1570 if (index == end_index) {
1571 nr = i_size & ~PAGE_CACHE_MASK;
1572 if (nr <= offset) {
1573 if (page)
1574 page_cache_release(page);
1575 break;
1576 }
1577 }
1578 nr -= offset;
1579
1580 if (page) {
1581 /*
1582 * If users can be writing to this page using arbitrary
1583 * virtual addresses, take care about potential aliasing
1584 * before reading the page on the kernel side.
1585 */
1586 if (mapping_writably_mapped(mapping))
1587 flush_dcache_page(page);
1588 /*
1589 * Mark the page accessed if we read the beginning.
1590 */
1591 if (!offset)
1592 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001593 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001595 page_cache_get(page);
1596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
1598 /*
1599 * Ok, we have the page, and it's up-to-date, so
1600 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04001602 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05001603 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 offset += ret;
1605 index += offset >> PAGE_CACHE_SHIFT;
1606 offset &= ~PAGE_CACHE_MASK;
1607
1608 page_cache_release(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04001609 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 break;
Al Viro6e58e792014-02-03 17:07:03 -05001611 if (ret < nr) {
1612 error = -EFAULT;
1613 break;
1614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 cond_resched();
1616 }
1617
1618 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05001619 file_accessed(file);
1620 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
1622
Hugh Dickins708e3502011-07-25 17:12:32 -07001623static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1624 struct pipe_inode_info *pipe, size_t len,
1625 unsigned int flags)
1626{
1627 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001628 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001629 unsigned int loff, nr_pages, req_pages;
1630 struct page *pages[PIPE_DEF_BUFFERS];
1631 struct partial_page partial[PIPE_DEF_BUFFERS];
1632 struct page *page;
1633 pgoff_t index, end_index;
1634 loff_t isize, left;
1635 int error, page_nr;
1636 struct splice_pipe_desc spd = {
1637 .pages = pages,
1638 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001639 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001640 .flags = flags,
1641 .ops = &page_cache_pipe_buf_ops,
1642 .spd_release = spd_release_page,
1643 };
1644
Hugh Dickins71f0e072011-07-25 17:12:33 -07001645 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001646 if (unlikely(*ppos >= isize))
1647 return 0;
1648
1649 left = isize - *ppos;
1650 if (unlikely(left < len))
1651 len = left;
1652
1653 if (splice_grow_spd(pipe, &spd))
1654 return -ENOMEM;
1655
1656 index = *ppos >> PAGE_CACHE_SHIFT;
1657 loff = *ppos & ~PAGE_CACHE_MASK;
1658 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04001659 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07001660
Hugh Dickins708e3502011-07-25 17:12:32 -07001661 spd.nr_pages = find_get_pages_contig(mapping, index,
1662 nr_pages, spd.pages);
1663 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001664 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001665
Hugh Dickins708e3502011-07-25 17:12:32 -07001666 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001667 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1668 if (error)
1669 break;
1670 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001671 spd.pages[spd.nr_pages++] = page;
1672 index++;
1673 }
1674
Hugh Dickins708e3502011-07-25 17:12:32 -07001675 index = *ppos >> PAGE_CACHE_SHIFT;
1676 nr_pages = spd.nr_pages;
1677 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001678
Hugh Dickins708e3502011-07-25 17:12:32 -07001679 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1680 unsigned int this_len;
1681
1682 if (!len)
1683 break;
1684
Hugh Dickins708e3502011-07-25 17:12:32 -07001685 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1686 page = spd.pages[page_nr];
1687
Hugh Dickins71f0e072011-07-25 17:12:33 -07001688 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001689 error = shmem_getpage(inode, index, &page,
1690 SGP_CACHE, NULL);
1691 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001692 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001693 unlock_page(page);
1694 page_cache_release(spd.pages[page_nr]);
1695 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001696 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001697
1698 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001699 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1700 if (unlikely(!isize || index > end_index))
1701 break;
1702
Hugh Dickins708e3502011-07-25 17:12:32 -07001703 if (end_index == index) {
1704 unsigned int plen;
1705
Hugh Dickins708e3502011-07-25 17:12:32 -07001706 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1707 if (plen <= loff)
1708 break;
1709
Hugh Dickins708e3502011-07-25 17:12:32 -07001710 this_len = min(this_len, plen - loff);
1711 len = this_len;
1712 }
1713
1714 spd.partial[page_nr].offset = loff;
1715 spd.partial[page_nr].len = this_len;
1716 len -= this_len;
1717 loff = 0;
1718 spd.nr_pages++;
1719 index++;
1720 }
1721
Hugh Dickins708e3502011-07-25 17:12:32 -07001722 while (page_nr < nr_pages)
1723 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001724
1725 if (spd.nr_pages)
1726 error = splice_to_pipe(pipe, &spd);
1727
Eric Dumazet047fe362012-06-12 15:24:40 +02001728 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001729
1730 if (error > 0) {
1731 *ppos += error;
1732 file_accessed(in);
1733 }
1734 return error;
1735}
1736
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001737/*
1738 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1739 */
1740static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001741 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001742{
1743 struct page *page;
1744 struct pagevec pvec;
1745 pgoff_t indices[PAGEVEC_SIZE];
1746 bool done = false;
1747 int i;
1748
1749 pagevec_init(&pvec, 0);
1750 pvec.nr = 1; /* start small: we may be there already */
1751 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001752 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001753 pvec.nr, pvec.pages, indices);
1754 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001755 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001756 index = end;
1757 break;
1758 }
1759 for (i = 0; i < pvec.nr; i++, index++) {
1760 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001761 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001762 done = true;
1763 break;
1764 }
1765 index = indices[i];
1766 }
1767 page = pvec.pages[i];
1768 if (page && !radix_tree_exceptional_entry(page)) {
1769 if (!PageUptodate(page))
1770 page = NULL;
1771 }
1772 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001773 (page && whence == SEEK_DATA) ||
1774 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001775 done = true;
1776 break;
1777 }
1778 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001779 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001780 pagevec_release(&pvec);
1781 pvec.nr = PAGEVEC_SIZE;
1782 cond_resched();
1783 }
1784 return index;
1785}
1786
Andrew Morton965c8e52012-12-17 15:59:39 -08001787static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001788{
1789 struct address_space *mapping = file->f_mapping;
1790 struct inode *inode = mapping->host;
1791 pgoff_t start, end;
1792 loff_t new_offset;
1793
Andrew Morton965c8e52012-12-17 15:59:39 -08001794 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1795 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001796 MAX_LFS_FILESIZE, i_size_read(inode));
1797 mutex_lock(&inode->i_mutex);
1798 /* We're holding i_mutex so we can access i_size directly */
1799
1800 if (offset < 0)
1801 offset = -EINVAL;
1802 else if (offset >= inode->i_size)
1803 offset = -ENXIO;
1804 else {
1805 start = offset >> PAGE_CACHE_SHIFT;
1806 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001807 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001808 new_offset <<= PAGE_CACHE_SHIFT;
1809 if (new_offset > offset) {
1810 if (new_offset < inode->i_size)
1811 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001812 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001813 offset = -ENXIO;
1814 else
1815 offset = inode->i_size;
1816 }
1817 }
1818
Hugh Dickins387aae62013-08-04 11:30:25 -07001819 if (offset >= 0)
1820 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001821 mutex_unlock(&inode->i_mutex);
1822 return offset;
1823}
1824
David Herrmann05f65b52014-08-08 14:25:36 -07001825/*
1826 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1827 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1828 */
1829#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1830#define LAST_SCAN 4 /* about 150ms max */
1831
1832static void shmem_tag_pins(struct address_space *mapping)
1833{
1834 struct radix_tree_iter iter;
1835 void **slot;
1836 pgoff_t start;
1837 struct page *page;
1838
1839 lru_add_drain();
1840 start = 0;
1841 rcu_read_lock();
1842
1843restart:
1844 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1845 page = radix_tree_deref_slot(slot);
1846 if (!page || radix_tree_exception(page)) {
1847 if (radix_tree_deref_retry(page))
1848 goto restart;
1849 } else if (page_count(page) - page_mapcount(page) > 1) {
1850 spin_lock_irq(&mapping->tree_lock);
1851 radix_tree_tag_set(&mapping->page_tree, iter.index,
1852 SHMEM_TAG_PINNED);
1853 spin_unlock_irq(&mapping->tree_lock);
1854 }
1855
1856 if (need_resched()) {
1857 cond_resched_rcu();
1858 start = iter.index + 1;
1859 goto restart;
1860 }
1861 }
1862 rcu_read_unlock();
1863}
1864
1865/*
1866 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1867 * via get_user_pages(), drivers might have some pending I/O without any active
1868 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1869 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1870 * them to be dropped.
1871 * The caller must guarantee that no new user will acquire writable references
1872 * to those pages to avoid races.
1873 */
David Herrmann40e041a2014-08-08 14:25:27 -07001874static int shmem_wait_for_pins(struct address_space *mapping)
1875{
David Herrmann05f65b52014-08-08 14:25:36 -07001876 struct radix_tree_iter iter;
1877 void **slot;
1878 pgoff_t start;
1879 struct page *page;
1880 int error, scan;
1881
1882 shmem_tag_pins(mapping);
1883
1884 error = 0;
1885 for (scan = 0; scan <= LAST_SCAN; scan++) {
1886 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1887 break;
1888
1889 if (!scan)
1890 lru_add_drain_all();
1891 else if (schedule_timeout_killable((HZ << scan) / 200))
1892 scan = LAST_SCAN;
1893
1894 start = 0;
1895 rcu_read_lock();
1896restart:
1897 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
1898 start, SHMEM_TAG_PINNED) {
1899
1900 page = radix_tree_deref_slot(slot);
1901 if (radix_tree_exception(page)) {
1902 if (radix_tree_deref_retry(page))
1903 goto restart;
1904
1905 page = NULL;
1906 }
1907
1908 if (page &&
1909 page_count(page) - page_mapcount(page) != 1) {
1910 if (scan < LAST_SCAN)
1911 goto continue_resched;
1912
1913 /*
1914 * On the last scan, we clean up all those tags
1915 * we inserted; but make a note that we still
1916 * found pages pinned.
1917 */
1918 error = -EBUSY;
1919 }
1920
1921 spin_lock_irq(&mapping->tree_lock);
1922 radix_tree_tag_clear(&mapping->page_tree,
1923 iter.index, SHMEM_TAG_PINNED);
1924 spin_unlock_irq(&mapping->tree_lock);
1925continue_resched:
1926 if (need_resched()) {
1927 cond_resched_rcu();
1928 start = iter.index + 1;
1929 goto restart;
1930 }
1931 }
1932 rcu_read_unlock();
1933 }
1934
1935 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07001936}
1937
1938#define F_ALL_SEALS (F_SEAL_SEAL | \
1939 F_SEAL_SHRINK | \
1940 F_SEAL_GROW | \
1941 F_SEAL_WRITE)
1942
1943int shmem_add_seals(struct file *file, unsigned int seals)
1944{
1945 struct inode *inode = file_inode(file);
1946 struct shmem_inode_info *info = SHMEM_I(inode);
1947 int error;
1948
1949 /*
1950 * SEALING
1951 * Sealing allows multiple parties to share a shmem-file but restrict
1952 * access to a specific subset of file operations. Seals can only be
1953 * added, but never removed. This way, mutually untrusted parties can
1954 * share common memory regions with a well-defined policy. A malicious
1955 * peer can thus never perform unwanted operations on a shared object.
1956 *
1957 * Seals are only supported on special shmem-files and always affect
1958 * the whole underlying inode. Once a seal is set, it may prevent some
1959 * kinds of access to the file. Currently, the following seals are
1960 * defined:
1961 * SEAL_SEAL: Prevent further seals from being set on this file
1962 * SEAL_SHRINK: Prevent the file from shrinking
1963 * SEAL_GROW: Prevent the file from growing
1964 * SEAL_WRITE: Prevent write access to the file
1965 *
1966 * As we don't require any trust relationship between two parties, we
1967 * must prevent seals from being removed. Therefore, sealing a file
1968 * only adds a given set of seals to the file, it never touches
1969 * existing seals. Furthermore, the "setting seals"-operation can be
1970 * sealed itself, which basically prevents any further seal from being
1971 * added.
1972 *
1973 * Semantics of sealing are only defined on volatile files. Only
1974 * anonymous shmem files support sealing. More importantly, seals are
1975 * never written to disk. Therefore, there's no plan to support it on
1976 * other file types.
1977 */
1978
1979 if (file->f_op != &shmem_file_operations)
1980 return -EINVAL;
1981 if (!(file->f_mode & FMODE_WRITE))
1982 return -EPERM;
1983 if (seals & ~(unsigned int)F_ALL_SEALS)
1984 return -EINVAL;
1985
1986 mutex_lock(&inode->i_mutex);
1987
1988 if (info->seals & F_SEAL_SEAL) {
1989 error = -EPERM;
1990 goto unlock;
1991 }
1992
1993 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
1994 error = mapping_deny_writable(file->f_mapping);
1995 if (error)
1996 goto unlock;
1997
1998 error = shmem_wait_for_pins(file->f_mapping);
1999 if (error) {
2000 mapping_allow_writable(file->f_mapping);
2001 goto unlock;
2002 }
2003 }
2004
2005 info->seals |= seals;
2006 error = 0;
2007
2008unlock:
2009 mutex_unlock(&inode->i_mutex);
2010 return error;
2011}
2012EXPORT_SYMBOL_GPL(shmem_add_seals);
2013
2014int shmem_get_seals(struct file *file)
2015{
2016 if (file->f_op != &shmem_file_operations)
2017 return -EINVAL;
2018
2019 return SHMEM_I(file_inode(file))->seals;
2020}
2021EXPORT_SYMBOL_GPL(shmem_get_seals);
2022
2023long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2024{
2025 long error;
2026
2027 switch (cmd) {
2028 case F_ADD_SEALS:
2029 /* disallow upper 32bit */
2030 if (arg > UINT_MAX)
2031 return -EINVAL;
2032
2033 error = shmem_add_seals(file, arg);
2034 break;
2035 case F_GET_SEALS:
2036 error = shmem_get_seals(file);
2037 break;
2038 default:
2039 error = -EINVAL;
2040 break;
2041 }
2042
2043 return error;
2044}
2045
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002046static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2047 loff_t len)
2048{
Al Viro496ad9a2013-01-23 17:07:38 -05002049 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002050 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002051 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002052 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002053 pgoff_t start, index, end;
2054 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002055
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002056 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2057 return -EOPNOTSUPP;
2058
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002059 mutex_lock(&inode->i_mutex);
2060
2061 if (mode & FALLOC_FL_PUNCH_HOLE) {
2062 struct address_space *mapping = file->f_mapping;
2063 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2064 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002065 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002066
David Herrmann40e041a2014-08-08 14:25:27 -07002067 /* protected by i_mutex */
2068 if (info->seals & F_SEAL_WRITE) {
2069 error = -EPERM;
2070 goto out;
2071 }
2072
Hugh Dickins8e205f72014-07-23 14:00:10 -07002073 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002074 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2075 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2076 spin_lock(&inode->i_lock);
2077 inode->i_private = &shmem_falloc;
2078 spin_unlock(&inode->i_lock);
2079
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002080 if ((u64)unmap_end > (u64)unmap_start)
2081 unmap_mapping_range(mapping, unmap_start,
2082 1 + unmap_end - unmap_start, 0);
2083 shmem_truncate_range(inode, offset, offset + len - 1);
2084 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002085
2086 spin_lock(&inode->i_lock);
2087 inode->i_private = NULL;
2088 wake_up_all(&shmem_falloc_waitq);
2089 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002090 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002091 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002092 }
2093
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002094 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2095 error = inode_newsize_ok(inode, offset + len);
2096 if (error)
2097 goto out;
2098
David Herrmann40e041a2014-08-08 14:25:27 -07002099 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2100 error = -EPERM;
2101 goto out;
2102 }
2103
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002104 start = offset >> PAGE_CACHE_SHIFT;
2105 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2106 /* Try to avoid a swapstorm if len is impossible to satisfy */
2107 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2108 error = -ENOSPC;
2109 goto out;
2110 }
2111
Hugh Dickins8e205f72014-07-23 14:00:10 -07002112 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002113 shmem_falloc.start = start;
2114 shmem_falloc.next = start;
2115 shmem_falloc.nr_falloced = 0;
2116 shmem_falloc.nr_unswapped = 0;
2117 spin_lock(&inode->i_lock);
2118 inode->i_private = &shmem_falloc;
2119 spin_unlock(&inode->i_lock);
2120
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002121 for (index = start; index < end; index++) {
2122 struct page *page;
2123
2124 /*
2125 * Good, the fallocate(2) manpage permits EINTR: we may have
2126 * been interrupted because we are using up too much memory.
2127 */
2128 if (signal_pending(current))
2129 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002130 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2131 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002132 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002133 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002134 NULL);
2135 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002136 /* Remove the !PageUptodate pages we added */
2137 shmem_undo_range(inode,
2138 (loff_t)start << PAGE_CACHE_SHIFT,
2139 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002140 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002141 }
2142
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002143 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002144 * Inform shmem_writepage() how far we have reached.
2145 * No need for lock or barrier: we have the page lock.
2146 */
2147 shmem_falloc.next++;
2148 if (!PageUptodate(page))
2149 shmem_falloc.nr_falloced++;
2150
2151 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002152 * If !PageUptodate, leave it that way so that freeable pages
2153 * can be recognized if we need to rollback on error later.
2154 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002155 * than free the pages we are allocating (and SGP_CACHE pages
2156 * might still be clean: we now need to mark those dirty too).
2157 */
2158 set_page_dirty(page);
2159 unlock_page(page);
2160 page_cache_release(page);
2161 cond_resched();
2162 }
2163
2164 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2165 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002166 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002167undone:
2168 spin_lock(&inode->i_lock);
2169 inode->i_private = NULL;
2170 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002171out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002172 mutex_unlock(&inode->i_mutex);
2173 return error;
2174}
2175
David Howells726c3342006-06-23 02:02:58 -07002176static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
David Howells726c3342006-06-23 02:02:58 -07002178 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
2180 buf->f_type = TMPFS_MAGIC;
2181 buf->f_bsize = PAGE_CACHE_SIZE;
2182 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002183 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002185 buf->f_bavail =
2186 buf->f_bfree = sbinfo->max_blocks -
2187 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002188 }
2189 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 buf->f_files = sbinfo->max_inodes;
2191 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 }
2193 /* else leave those fields 0 like simple_statfs */
2194 return 0;
2195}
2196
2197/*
2198 * File creation. Allocate an inode, and we're done..
2199 */
2200static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002201shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002203 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 int error = -ENOSPC;
2205
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002206 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002208 error = simple_acl_create(dir, inode);
2209 if (error)
2210 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002211 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002212 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002213 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002214 if (error && error != -EOPNOTSUPP)
2215 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002216
Al Viro718deb62009-12-16 19:35:36 -05002217 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 dir->i_size += BOGO_DIRENT_SIZE;
2219 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2220 d_instantiate(dentry, inode);
2221 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 }
2223 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002224out_iput:
2225 iput(inode);
2226 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227}
2228
Al Viro60545d02013-06-07 01:20:27 -04002229static int
2230shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2231{
2232 struct inode *inode;
2233 int error = -ENOSPC;
2234
2235 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2236 if (inode) {
2237 error = security_inode_init_security(inode, dir,
2238 NULL,
2239 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002240 if (error && error != -EOPNOTSUPP)
2241 goto out_iput;
2242 error = simple_acl_create(dir, inode);
2243 if (error)
2244 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002245 d_tmpfile(dentry, inode);
2246 }
2247 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002248out_iput:
2249 iput(inode);
2250 return error;
Al Viro60545d02013-06-07 01:20:27 -04002251}
2252
Al Viro18bb1db2011-07-26 01:41:39 -04002253static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254{
2255 int error;
2256
2257 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2258 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002259 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 return 0;
2261}
2262
Al Viro4acdaf22011-07-26 01:42:34 -04002263static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002264 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265{
2266 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2267}
2268
2269/*
2270 * Link a file..
2271 */
2272static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2273{
2274 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002275 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
2277 /*
2278 * No ordinary (disk based) filesystem counts links as inodes;
2279 * but each new link needs a new dentry, pinning lowmem, and
2280 * tmpfs dentries cannot be pruned until they are unlinked.
2281 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002282 ret = shmem_reserve_inode(inode->i_sb);
2283 if (ret)
2284 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
2286 dir->i_size += BOGO_DIRENT_SIZE;
2287 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002288 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002289 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 dget(dentry); /* Extra pinning count for the created dentry */
2291 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002292out:
2293 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294}
2295
2296static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2297{
2298 struct inode *inode = dentry->d_inode;
2299
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002300 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2301 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
2303 dir->i_size -= BOGO_DIRENT_SIZE;
2304 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002305 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 dput(dentry); /* Undo the count from "create" - this does all the work */
2307 return 0;
2308}
2309
2310static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2311{
2312 if (!simple_empty(dentry))
2313 return -ENOTEMPTY;
2314
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002315 drop_nlink(dentry->d_inode);
2316 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 return shmem_unlink(dir, dentry);
2318}
2319
Miklos Szeredi37456772014-07-23 15:15:34 +02002320static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2321{
David Howellse36cb0b2015-01-29 12:02:35 +00002322 bool old_is_dir = d_is_dir(old_dentry);
2323 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002324
2325 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2326 if (old_is_dir) {
2327 drop_nlink(old_dir);
2328 inc_nlink(new_dir);
2329 } else {
2330 drop_nlink(new_dir);
2331 inc_nlink(old_dir);
2332 }
2333 }
2334 old_dir->i_ctime = old_dir->i_mtime =
2335 new_dir->i_ctime = new_dir->i_mtime =
2336 old_dentry->d_inode->i_ctime =
2337 new_dentry->d_inode->i_ctime = CURRENT_TIME;
2338
2339 return 0;
2340}
2341
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002342static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2343{
2344 struct dentry *whiteout;
2345 int error;
2346
2347 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2348 if (!whiteout)
2349 return -ENOMEM;
2350
2351 error = shmem_mknod(old_dir, whiteout,
2352 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2353 dput(whiteout);
2354 if (error)
2355 return error;
2356
2357 /*
2358 * Cheat and hash the whiteout while the old dentry is still in
2359 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2360 *
2361 * d_lookup() will consistently find one of them at this point,
2362 * not sure which one, but that isn't even important.
2363 */
2364 d_rehash(whiteout);
2365 return 0;
2366}
2367
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368/*
2369 * The VFS layer already does all the dentry stuff for rename,
2370 * we just have to decrement the usage count for the target if
2371 * it exists so that the VFS layer correctly free's it when it
2372 * gets overwritten.
2373 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002374static 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 -07002375{
2376 struct inode *inode = old_dentry->d_inode;
2377 int they_are_dirs = S_ISDIR(inode->i_mode);
2378
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002379 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002380 return -EINVAL;
2381
Miklos Szeredi37456772014-07-23 15:15:34 +02002382 if (flags & RENAME_EXCHANGE)
2383 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (!simple_empty(new_dentry))
2386 return -ENOTEMPTY;
2387
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002388 if (flags & RENAME_WHITEOUT) {
2389 int error;
2390
2391 error = shmem_whiteout(old_dir, old_dentry);
2392 if (error)
2393 return error;
2394 }
2395
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 if (new_dentry->d_inode) {
2397 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002398 if (they_are_dirs) {
2399 drop_nlink(new_dentry->d_inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002400 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002401 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002403 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002404 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 }
2406
2407 old_dir->i_size -= BOGO_DIRENT_SIZE;
2408 new_dir->i_size += BOGO_DIRENT_SIZE;
2409 old_dir->i_ctime = old_dir->i_mtime =
2410 new_dir->i_ctime = new_dir->i_mtime =
2411 inode->i_ctime = CURRENT_TIME;
2412 return 0;
2413}
2414
2415static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2416{
2417 int error;
2418 int len;
2419 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002420 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 char *kaddr;
2422 struct shmem_inode_info *info;
2423
2424 len = strlen(symname) + 1;
2425 if (len > PAGE_CACHE_SIZE)
2426 return -ENAMETOOLONG;
2427
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002428 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 if (!inode)
2430 return -ENOSPC;
2431
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002432 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002433 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002434 if (error) {
2435 if (error != -EOPNOTSUPP) {
2436 iput(inode);
2437 return error;
2438 }
2439 error = 0;
2440 }
2441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 info = SHMEM_I(inode);
2443 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002444 if (len <= SHORT_SYMLINK_LEN) {
2445 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2446 if (!info->symlink) {
2447 iput(inode);
2448 return -ENOMEM;
2449 }
2450 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 } else {
2452 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2453 if (error) {
2454 iput(inode);
2455 return error;
2456 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002457 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002459 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002461 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002462 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002464 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 page_cache_release(page);
2466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 dir->i_size += BOGO_DIRENT_SIZE;
2468 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2469 d_instantiate(dentry, inode);
2470 dget(dentry);
2471 return 0;
2472}
2473
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002474static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002476 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002477 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478}
2479
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002480static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481{
2482 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002483 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2484 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002485 if (page)
2486 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002487 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488}
2489
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002490static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491{
2492 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002493 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 kunmap(page);
2495 mark_page_accessed(page);
2496 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 }
2498}
2499
Eric Parisb09e0fa2011-05-24 17:12:39 -07002500#ifdef CONFIG_TMPFS_XATTR
2501/*
2502 * Superblocks without xattr inode operations may get some security.* xattr
2503 * support from the LSM "for free". As soon as we have any other xattrs
2504 * like ACLs, we also need to implement the security.* handlers at
2505 * filesystem level, though.
2506 */
2507
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002508/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002509 * Callback for security_inode_init_security() for acquiring xattrs.
2510 */
2511static int shmem_initxattrs(struct inode *inode,
2512 const struct xattr *xattr_array,
2513 void *fs_info)
2514{
2515 struct shmem_inode_info *info = SHMEM_I(inode);
2516 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002517 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002518 size_t len;
2519
2520 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002521 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002522 if (!new_xattr)
2523 return -ENOMEM;
2524
2525 len = strlen(xattr->name) + 1;
2526 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2527 GFP_KERNEL);
2528 if (!new_xattr->name) {
2529 kfree(new_xattr);
2530 return -ENOMEM;
2531 }
2532
2533 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2534 XATTR_SECURITY_PREFIX_LEN);
2535 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2536 xattr->name, len);
2537
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002538 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002539 }
2540
2541 return 0;
2542}
2543
Eric Parisb09e0fa2011-05-24 17:12:39 -07002544static const struct xattr_handler *shmem_xattr_handlers[] = {
2545#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002546 &posix_acl_access_xattr_handler,
2547 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002548#endif
2549 NULL
2550};
2551
2552static int shmem_xattr_validate(const char *name)
2553{
2554 struct { const char *prefix; size_t len; } arr[] = {
2555 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2556 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2557 };
2558 int i;
2559
2560 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2561 size_t preflen = arr[i].len;
2562 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2563 if (!name[preflen])
2564 return -EINVAL;
2565 return 0;
2566 }
2567 }
2568 return -EOPNOTSUPP;
2569}
2570
2571static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2572 void *buffer, size_t size)
2573{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002574 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002575 int err;
2576
2577 /*
2578 * If this is a request for a synthetic attribute in the system.*
2579 * namespace use the generic infrastructure to resolve a handler
2580 * for it via sb->s_xattr.
2581 */
2582 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2583 return generic_getxattr(dentry, name, buffer, size);
2584
2585 err = shmem_xattr_validate(name);
2586 if (err)
2587 return err;
2588
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002589 return simple_xattr_get(&info->xattrs, name, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002590}
2591
2592static int shmem_setxattr(struct dentry *dentry, const char *name,
2593 const void *value, size_t size, int flags)
2594{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002595 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002596 int err;
2597
2598 /*
2599 * If this is a request for a synthetic attribute in the system.*
2600 * namespace use the generic infrastructure to resolve a handler
2601 * for it via sb->s_xattr.
2602 */
2603 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2604 return generic_setxattr(dentry, name, value, size, flags);
2605
2606 err = shmem_xattr_validate(name);
2607 if (err)
2608 return err;
2609
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002610 return simple_xattr_set(&info->xattrs, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002611}
2612
2613static int shmem_removexattr(struct dentry *dentry, const char *name)
2614{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002615 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002616 int err;
2617
2618 /*
2619 * If this is a request for a synthetic attribute in the system.*
2620 * namespace use the generic infrastructure to resolve a handler
2621 * for it via sb->s_xattr.
2622 */
2623 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2624 return generic_removexattr(dentry, name);
2625
2626 err = shmem_xattr_validate(name);
2627 if (err)
2628 return err;
2629
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002630 return simple_xattr_remove(&info->xattrs, name);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002631}
2632
2633static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2634{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002635 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2636 return simple_xattr_list(&info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002637}
2638#endif /* CONFIG_TMPFS_XATTR */
2639
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002640static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002642 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002643#ifdef CONFIG_TMPFS_XATTR
2644 .setxattr = shmem_setxattr,
2645 .getxattr = shmem_getxattr,
2646 .listxattr = shmem_listxattr,
2647 .removexattr = shmem_removexattr,
2648#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649};
2650
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002651static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 .readlink = generic_readlink,
2653 .follow_link = shmem_follow_link,
2654 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002655#ifdef CONFIG_TMPFS_XATTR
2656 .setxattr = shmem_setxattr,
2657 .getxattr = shmem_getxattr,
2658 .listxattr = shmem_listxattr,
2659 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002660#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002661};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002662
David M. Grimes91828a42006-10-17 00:09:45 -07002663static struct dentry *shmem_get_parent(struct dentry *child)
2664{
2665 return ERR_PTR(-ESTALE);
2666}
2667
2668static int shmem_match(struct inode *ino, void *vfh)
2669{
2670 __u32 *fh = vfh;
2671 __u64 inum = fh[2];
2672 inum = (inum << 32) | fh[1];
2673 return ino->i_ino == inum && fh[0] == ino->i_generation;
2674}
2675
Christoph Hellwig480b1162007-10-21 16:42:13 -07002676static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2677 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002678{
David M. Grimes91828a42006-10-17 00:09:45 -07002679 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002680 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002681 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002682
Christoph Hellwig480b1162007-10-21 16:42:13 -07002683 if (fh_len < 3)
2684 return NULL;
2685
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002686 inum = fid->raw[2];
2687 inum = (inum << 32) | fid->raw[1];
2688
Christoph Hellwig480b1162007-10-21 16:42:13 -07002689 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2690 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002691 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002692 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002693 iput(inode);
2694 }
2695
Christoph Hellwig480b1162007-10-21 16:42:13 -07002696 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002697}
2698
Al Virob0b03822012-04-02 14:34:06 -04002699static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2700 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002701{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302702 if (*len < 3) {
2703 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002704 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302705 }
David M. Grimes91828a42006-10-17 00:09:45 -07002706
Al Viro1d3382cb2010-10-23 15:19:20 -04002707 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002708 /* Unfortunately insert_inode_hash is not idempotent,
2709 * so as we hash inodes here rather than at creation
2710 * time, we need a lock to ensure we only try
2711 * to do it once
2712 */
2713 static DEFINE_SPINLOCK(lock);
2714 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002715 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002716 __insert_inode_hash(inode,
2717 inode->i_ino + inode->i_generation);
2718 spin_unlock(&lock);
2719 }
2720
2721 fh[0] = inode->i_generation;
2722 fh[1] = inode->i_ino;
2723 fh[2] = ((__u64)inode->i_ino) >> 32;
2724
2725 *len = 3;
2726 return 1;
2727}
2728
Christoph Hellwig39655162007-10-21 16:42:17 -07002729static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002730 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002731 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002732 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002733};
2734
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002735static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2736 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
2738 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002739 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002740 uid_t uid;
2741 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002743 while (options != NULL) {
2744 this_char = options;
2745 for (;;) {
2746 /*
2747 * NUL-terminate this option: unfortunately,
2748 * mount options form a comma-separated list,
2749 * but mpol's nodelist may also contain commas.
2750 */
2751 options = strchr(options, ',');
2752 if (options == NULL)
2753 break;
2754 options++;
2755 if (!isdigit(*options)) {
2756 options[-1] = '\0';
2757 break;
2758 }
2759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 if (!*this_char)
2761 continue;
2762 if ((value = strchr(this_char,'=')) != NULL) {
2763 *value++ = 0;
2764 } else {
2765 printk(KERN_ERR
2766 "tmpfs: No value for mount option '%s'\n",
2767 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002768 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 }
2770
2771 if (!strcmp(this_char,"size")) {
2772 unsigned long long size;
2773 size = memparse(value,&rest);
2774 if (*rest == '%') {
2775 size <<= PAGE_SHIFT;
2776 size *= totalram_pages;
2777 do_div(size, 100);
2778 rest++;
2779 }
2780 if (*rest)
2781 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002782 sbinfo->max_blocks =
2783 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002785 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if (*rest)
2787 goto bad_val;
2788 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002789 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if (*rest)
2791 goto bad_val;
2792 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002793 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002795 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 if (*rest)
2797 goto bad_val;
2798 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002799 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002801 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 if (*rest)
2803 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002804 sbinfo->uid = make_kuid(current_user_ns(), uid);
2805 if (!uid_valid(sbinfo->uid))
2806 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002808 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002810 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 if (*rest)
2812 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002813 sbinfo->gid = make_kgid(current_user_ns(), gid);
2814 if (!gid_valid(sbinfo->gid))
2815 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002816 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002817 mpol_put(mpol);
2818 mpol = NULL;
2819 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002820 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 } else {
2822 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2823 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002824 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 }
2826 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002827 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 return 0;
2829
2830bad_val:
2831 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2832 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002833error:
2834 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 return 1;
2836
2837}
2838
2839static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2840{
2841 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002842 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002843 unsigned long inodes;
2844 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845
Greg Thelen5f001102013-02-22 16:36:01 -08002846 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002847 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002848 return error;
2849
2850 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002851 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002852 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002853 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002854 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002855 goto out;
2856 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002857 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002858 * but we must separately disallow unlimited->limited, because
2859 * in that case we have no record of how much is already in use.
2860 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002861 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002862 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002863 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002864 goto out;
2865
2866 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002867 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002868 sbinfo->max_inodes = config.max_inodes;
2869 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002870
Greg Thelen5f001102013-02-22 16:36:01 -08002871 /*
2872 * Preserve previous mempolicy unless mpol remount option was specified.
2873 */
2874 if (config.mpol) {
2875 mpol_put(sbinfo->mpol);
2876 sbinfo->mpol = config.mpol; /* transfers initial ref */
2877 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002878out:
2879 spin_unlock(&sbinfo->stat_lock);
2880 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002882
Al Viro34c80b12011-12-08 21:32:45 -05002883static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002884{
Al Viro34c80b12011-12-08 21:32:45 -05002885 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002886
2887 if (sbinfo->max_blocks != shmem_default_max_blocks())
2888 seq_printf(seq, ",size=%luk",
2889 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2890 if (sbinfo->max_inodes != shmem_default_max_inodes())
2891 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2892 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002893 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002894 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2895 seq_printf(seq, ",uid=%u",
2896 from_kuid_munged(&init_user_ns, sbinfo->uid));
2897 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2898 seq_printf(seq, ",gid=%u",
2899 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002900 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002901 return 0;
2902}
David Herrmann9183df22014-08-08 14:25:29 -07002903
2904#define MFD_NAME_PREFIX "memfd:"
2905#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2906#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2907
2908#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2909
2910SYSCALL_DEFINE2(memfd_create,
2911 const char __user *, uname,
2912 unsigned int, flags)
2913{
2914 struct shmem_inode_info *info;
2915 struct file *file;
2916 int fd, error;
2917 char *name;
2918 long len;
2919
2920 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2921 return -EINVAL;
2922
2923 /* length includes terminating zero */
2924 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2925 if (len <= 0)
2926 return -EFAULT;
2927 if (len > MFD_NAME_MAX_LEN + 1)
2928 return -EINVAL;
2929
2930 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2931 if (!name)
2932 return -ENOMEM;
2933
2934 strcpy(name, MFD_NAME_PREFIX);
2935 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2936 error = -EFAULT;
2937 goto err_name;
2938 }
2939
2940 /* terminating-zero may have changed after strnlen_user() returned */
2941 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
2942 error = -EFAULT;
2943 goto err_name;
2944 }
2945
2946 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
2947 if (fd < 0) {
2948 error = fd;
2949 goto err_name;
2950 }
2951
2952 file = shmem_file_setup(name, 0, VM_NORESERVE);
2953 if (IS_ERR(file)) {
2954 error = PTR_ERR(file);
2955 goto err_fd;
2956 }
2957 info = SHMEM_I(file_inode(file));
2958 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
2959 file->f_flags |= O_RDWR | O_LARGEFILE;
2960 if (flags & MFD_ALLOW_SEALING)
2961 info->seals &= ~F_SEAL_SEAL;
2962
2963 fd_install(fd, file);
2964 kfree(name);
2965 return fd;
2966
2967err_fd:
2968 put_unused_fd(fd);
2969err_name:
2970 kfree(name);
2971 return error;
2972}
2973
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002974#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
2976static void shmem_put_super(struct super_block *sb)
2977{
Hugh Dickins602586a2010-08-17 15:23:56 -07002978 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2979
2980 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002981 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07002982 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 sb->s_fs_info = NULL;
2984}
2985
Kay Sievers2b2af542009-04-30 15:23:42 +02002986int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987{
2988 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002989 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002990 int err = -ENOMEM;
2991
2992 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002993 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002994 L1_CACHE_BYTES), GFP_KERNEL);
2995 if (!sbinfo)
2996 return -ENOMEM;
2997
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002998 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002999 sbinfo->uid = current_fsuid();
3000 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003001 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003003#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 /*
3005 * Per default we only allow half of the physical ram per
3006 * tmpfs instance, limiting inodes to one per page of lowmem;
3007 * but the internal instance is left unlimited.
3008 */
Al Viroca4e0512013-08-31 12:57:10 -04003009 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003010 sbinfo->max_blocks = shmem_default_max_blocks();
3011 sbinfo->max_inodes = shmem_default_max_inodes();
3012 if (shmem_parse_options(data, sbinfo, false)) {
3013 err = -EINVAL;
3014 goto failed;
3015 }
Al Viroca4e0512013-08-31 12:57:10 -04003016 } else {
3017 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 }
David M. Grimes91828a42006-10-17 00:09:45 -07003019 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003020 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021#else
3022 sb->s_flags |= MS_NOUSER;
3023#endif
3024
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003025 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003026 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003027 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003028 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003029
Hugh Dickins285b2c42011-08-03 16:21:20 -07003030 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 sb->s_blocksize = PAGE_CACHE_SIZE;
3032 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
3033 sb->s_magic = TMPFS_MAGIC;
3034 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003035 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003036#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003037 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003038#endif
3039#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003040 sb->s_flags |= MS_POSIXACL;
3041#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003042
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003043 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 if (!inode)
3045 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003046 inode->i_uid = sbinfo->uid;
3047 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003048 sb->s_root = d_make_root(inode);
3049 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003050 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 return 0;
3052
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053failed:
3054 shmem_put_super(sb);
3055 return err;
3056}
3057
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003058static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
3060static struct inode *shmem_alloc_inode(struct super_block *sb)
3061{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003062 struct shmem_inode_info *info;
3063 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3064 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003066 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067}
3068
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003069static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003070{
3071 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003072 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3073}
3074
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075static void shmem_destroy_inode(struct inode *inode)
3076{
Al Viro09208d12011-07-26 03:15:03 -04003077 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003079 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080}
3081
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003082static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003084 struct shmem_inode_info *info = foo;
3085 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086}
3087
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003088static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089{
3090 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3091 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003092 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 return 0;
3094}
3095
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003096static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003098 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099}
3100
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003101static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003103 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003105 .write_begin = shmem_write_begin,
3106 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003108#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003109 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003110#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003111 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112};
3113
Helge Deller15ad7cd2006-12-06 20:40:36 -08003114static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 .mmap = shmem_mmap,
3116#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003117 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003118 .read = new_sync_read,
Al Viro81742022014-04-03 03:17:43 -04003119 .write = new_sync_write,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003120 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003121 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003122 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003123 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003124 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003125 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126#endif
3127};
3128
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003129static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07003130 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003131#ifdef CONFIG_TMPFS_XATTR
3132 .setxattr = shmem_setxattr,
3133 .getxattr = shmem_getxattr,
3134 .listxattr = shmem_listxattr,
3135 .removexattr = shmem_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003136 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003137#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138};
3139
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003140static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141#ifdef CONFIG_TMPFS
3142 .create = shmem_create,
3143 .lookup = simple_lookup,
3144 .link = shmem_link,
3145 .unlink = shmem_unlink,
3146 .symlink = shmem_symlink,
3147 .mkdir = shmem_mkdir,
3148 .rmdir = shmem_rmdir,
3149 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003150 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003151 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003153#ifdef CONFIG_TMPFS_XATTR
3154 .setxattr = shmem_setxattr,
3155 .getxattr = shmem_getxattr,
3156 .listxattr = shmem_listxattr,
3157 .removexattr = shmem_removexattr,
3158#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003159#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003160 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003161 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003162#endif
3163};
3164
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003165static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003166#ifdef CONFIG_TMPFS_XATTR
3167 .setxattr = shmem_setxattr,
3168 .getxattr = shmem_getxattr,
3169 .listxattr = shmem_listxattr,
3170 .removexattr = shmem_removexattr,
3171#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003172#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003173 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003174 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003175#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176};
3177
Hugh Dickins759b9772007-03-05 00:30:28 -08003178static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 .alloc_inode = shmem_alloc_inode,
3180 .destroy_inode = shmem_destroy_inode,
3181#ifdef CONFIG_TMPFS
3182 .statfs = shmem_statfs,
3183 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003184 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185#endif
Al Viro1f895f72010-06-05 19:10:41 -04003186 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 .drop_inode = generic_delete_inode,
3188 .put_super = shmem_put_super,
3189};
3190
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003191static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003192 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003193 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194#ifdef CONFIG_NUMA
3195 .set_policy = shmem_set_policy,
3196 .get_policy = shmem_get_policy,
3197#endif
3198};
3199
Al Viro3c26ff62010-07-25 11:46:36 +04003200static struct dentry *shmem_mount(struct file_system_type *fs_type,
3201 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202{
Al Viro3c26ff62010-07-25 11:46:36 +04003203 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204}
3205
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003206static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 .owner = THIS_MODULE,
3208 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003209 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003211 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003214int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215{
3216 int error;
3217
Rob Landley16203a72013-09-11 14:26:12 -07003218 /* If rootfs called this, don't re-init */
3219 if (shmem_inode_cachep)
3220 return 0;
3221
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003222 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 if (error)
3224 goto out3;
3225
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003226 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 if (error) {
3228 printk(KERN_ERR "Could not register tmpfs\n");
3229 goto out2;
3230 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003231
Al Viroca4e0512013-08-31 12:57:10 -04003232 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 if (IS_ERR(shm_mnt)) {
3234 error = PTR_ERR(shm_mnt);
3235 printk(KERN_ERR "Could not kern_mount tmpfs\n");
3236 goto out1;
3237 }
3238 return 0;
3239
3240out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003241 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003243 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244out3:
3245 shm_mnt = ERR_PTR(error);
3246 return error;
3247}
Matt Mackall853ac432009-01-06 14:40:20 -08003248
3249#else /* !CONFIG_SHMEM */
3250
3251/*
3252 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3253 *
3254 * This is intended for small system where the benefits of the full
3255 * shmem code (swap-backed and resource-limited) are outweighed by
3256 * their complexity. On systems without swap this code should be
3257 * effectively equivalent, but much lighter weight.
3258 */
3259
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003260static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003261 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003262 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003263 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003264 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003265};
3266
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003267int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003268{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003269 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003270
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003271 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003272 BUG_ON(IS_ERR(shm_mnt));
3273
3274 return 0;
3275}
3276
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003277int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003278{
3279 return 0;
3280}
3281
Hugh Dickins3f96b792009-09-21 17:03:37 -07003282int shmem_lock(struct file *file, int lock, struct user_struct *user)
3283{
3284 return 0;
3285}
3286
Hugh Dickins24513262012-01-20 14:34:21 -08003287void shmem_unlock_mapping(struct address_space *mapping)
3288{
3289}
3290
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003291void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003292{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003293 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003294}
3295EXPORT_SYMBOL_GPL(shmem_truncate_range);
3296
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003297#define shmem_vm_ops generic_file_vm_ops
3298#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003299#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003300#define shmem_acct_size(flags, size) 0
3301#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003302
3303#endif /* CONFIG_SHMEM */
3304
3305/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306
Al Viro34515382013-02-14 22:38:02 -05003307static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003308 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003309};
3310
Eric Parisc7277092013-12-02 11:24:19 +00003311static struct file *__shmem_file_setup(const char *name, loff_t size,
3312 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313{
Al Viro6b4d0b22013-02-14 21:37:26 -05003314 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003316 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003317 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 struct qstr this;
3319
3320 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003321 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322
Hugh Dickins285b2c42011-08-03 16:21:20 -07003323 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 return ERR_PTR(-EINVAL);
3325
3326 if (shmem_acct_size(flags, size))
3327 return ERR_PTR(-ENOMEM);
3328
Al Viro6b4d0b22013-02-14 21:37:26 -05003329 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 this.name = name;
3331 this.len = strlen(name);
3332 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003333 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003334 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003335 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003336 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003338 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
Al Viro6b4d0b22013-02-14 21:37:26 -05003340 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05003341 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003343 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344
Eric Parisc7277092013-12-02 11:24:19 +00003345 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04003346 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003348 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05003349 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3350 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003351 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003352
Al Viro6b4d0b22013-02-14 21:37:26 -05003353 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04003354 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05003355 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003356 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003357
Al Viro6b4d0b22013-02-14 21:37:26 -05003358 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360put_memory:
3361 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003362put_path:
3363 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05003364 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365}
Eric Parisc7277092013-12-02 11:24:19 +00003366
3367/**
3368 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3369 * kernel internal. There will be NO LSM permission checks against the
3370 * underlying inode. So users of this interface must do LSM checks at a
3371 * higher layer. The one user is the big_key implementation. LSM checks
3372 * are provided at the key level rather than the inode level.
3373 * @name: name for dentry (to be seen in /proc/<pid>/maps
3374 * @size: size to be set for the file
3375 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3376 */
3377struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3378{
3379 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3380}
3381
3382/**
3383 * shmem_file_setup - get an unlinked file living in tmpfs
3384 * @name: name for dentry (to be seen in /proc/<pid>/maps
3385 * @size: size to be set for the file
3386 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3387 */
3388struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3389{
3390 return __shmem_file_setup(name, size, flags, 0);
3391}
Keith Packard395e0dd2008-06-20 00:08:06 -07003392EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393
Randy Dunlap46711812008-03-19 17:00:41 -07003394/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3397 */
3398int shmem_zero_setup(struct vm_area_struct *vma)
3399{
3400 struct file *file;
3401 loff_t size = vma->vm_end - vma->vm_start;
3402
3403 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3404 if (IS_ERR(file))
3405 return PTR_ERR(file);
3406
3407 if (vma->vm_file)
3408 fput(vma->vm_file);
3409 vma->vm_file = file;
3410 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 return 0;
3412}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003413
3414/**
3415 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3416 * @mapping: the page's address_space
3417 * @index: the page index
3418 * @gfp: the page allocator flags to use if allocating
3419 *
3420 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3421 * with any new page allocations done using the specified allocation flags.
3422 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3423 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3424 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3425 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003426 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3427 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003428 */
3429struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3430 pgoff_t index, gfp_t gfp)
3431{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003432#ifdef CONFIG_SHMEM
3433 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003434 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003435 int error;
3436
3437 BUG_ON(mapping->a_ops != &shmem_aops);
3438 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3439 if (error)
3440 page = ERR_PTR(error);
3441 else
3442 unlock_page(page);
3443 return page;
3444#else
3445 /*
3446 * The tiny !SHMEM case uses ramfs without swap
3447 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003448 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003449#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003450}
3451EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);