blob: 9187eee4128b45769583dae2a73720e3545ee933 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins6922c0c2011-08-03 16:21:25 -07009 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -070011 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 *
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 *
Matt Mackall853ac432009-01-06 14:40:20 -080018 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * This file is released under the GPL.
22 */
23
Matt Mackall853ac432009-01-06 14:40:20 -080024#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
Andrew Morton250297e2013-04-29 15:06:12 -070028#include <linux/ramfs.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070029#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080030#include <linux/file.h>
31#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040032#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080033#include <linux/swap.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080034#include <linux/uio.h>
Matt Mackall853ac432009-01-06 14:40:20 -080035
36static struct vfsmount *shm_mnt;
37
38#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
43 */
44
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070045#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070046#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010047#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080048#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/string.h>
51#include <linux/slab.h>
52#include <linux/backing-dev.h>
53#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070056#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070057#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070058#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070059#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/security.h>
61#include <linux/swapops.h>
62#include <linux/mempolicy.h>
63#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000064#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070065#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070066#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080067#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040068#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070069#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070070#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070071#include <uapi/linux/memfd.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070072
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <asm/pgtable.h>
75
Mel Gormandd56b042015-11-06 16:28:43 -080076#include "internal.h"
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/* Pretend that each entry is of this size in directory's i_size */
82#define BOGO_DIRENT_SIZE 20
83
Hugh Dickins69f07ec2011-08-03 16:21:26 -070084/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
85#define SHORT_SYMLINK_LEN 128
86
Hugh Dickins1aac1402012-05-29 15:06:42 -070087/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070088 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
89 * inode->i_private (with i_mutex making sure that it has only one user at
90 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070091 */
92struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -070093 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -070094 pgoff_t start; /* start of range currently being fallocated */
95 pgoff_t next; /* the next page offset to be fallocated */
96 pgoff_t nr_falloced; /* how many new pages have been fallocated */
97 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
98};
99
Hugh Dickins285b2c42011-08-03 16:21:20 -0700100/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 SGP_READ, /* don't exceed i_size, don't allocate page */
103 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800104 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700105 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
106 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107};
108
Andrew Mortonb76db732008-02-08 04:21:49 -0800109#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800110static unsigned long shmem_default_max_blocks(void)
111{
112 return totalram_pages / 2;
113}
114
115static unsigned long shmem_default_max_inodes(void)
116{
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118}
Andrew Mortonb76db732008-02-08 04:21:49 -0800119#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800120
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700121static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
122static int shmem_replace_page(struct page **pagep, gfp_t gfp,
123 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700124static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
125 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
126
127static inline int shmem_getpage(struct inode *inode, pgoff_t index,
128 struct page **pagep, enum sgp_type sgp, int *fault_type)
129{
130 return shmem_getpage_gfp(inode, index, pagep, sgp,
131 mapping_gfp_mask(inode->i_mapping), fault_type);
132}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
135{
136 return sb->s_fs_info;
137}
138
139/*
140 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141 * for shared memory and for shared anonymous (/dev/zero) mappings
142 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143 * consistent with the pre-accounting of private mappings ...
144 */
145static inline int shmem_acct_size(unsigned long flags, loff_t size)
146{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000147 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000148 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151static inline void shmem_unacct_size(unsigned long flags, loff_t size)
152{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000153 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 vm_unacct_memory(VM_ACCT(size));
155}
156
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700157static inline int shmem_reacct_size(unsigned long flags,
158 loff_t oldsize, loff_t newsize)
159{
160 if (!(flags & VM_NORESERVE)) {
161 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
162 return security_vm_enough_memory_mm(current->mm,
163 VM_ACCT(newsize) - VM_ACCT(oldsize));
164 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
165 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
166 }
167 return 0;
168}
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/*
171 * ... whereas tmpfs objects are accounted incrementally as
172 * pages are allocated, in order to allow huge sparse files.
173 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
175 */
176static inline int shmem_acct_block(unsigned long flags)
177{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000178 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000179 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180}
181
182static inline void shmem_unacct_blocks(unsigned long flags, long pages)
183{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000184 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
186}
187
Hugh Dickins759b9772007-03-05 00:30:28 -0800188static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700189static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800190static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800191static const struct inode_operations shmem_inode_operations;
192static const struct inode_operations shmem_dir_inode_operations;
193static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400194static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800197static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800199static int shmem_reserve_inode(struct super_block *sb)
200{
201 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
202 if (sbinfo->max_inodes) {
203 spin_lock(&sbinfo->stat_lock);
204 if (!sbinfo->free_inodes) {
205 spin_unlock(&sbinfo->stat_lock);
206 return -ENOSPC;
207 }
208 sbinfo->free_inodes--;
209 spin_unlock(&sbinfo->stat_lock);
210 }
211 return 0;
212}
213
214static void shmem_free_inode(struct super_block *sb)
215{
216 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
217 if (sbinfo->max_inodes) {
218 spin_lock(&sbinfo->stat_lock);
219 sbinfo->free_inodes++;
220 spin_unlock(&sbinfo->stat_lock);
221 }
222}
223
Randy Dunlap46711812008-03-19 17:00:41 -0700224/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700225 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 * @inode: inode to recalc
227 *
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
230 *
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
233 *
234 * It has to be called with the spinlock held.
235 */
236static void shmem_recalc_inode(struct inode *inode)
237{
238 struct shmem_inode_info *info = SHMEM_I(inode);
239 long freed;
240
241 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
242 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700243 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
244 if (sbinfo->max_blocks)
245 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700247 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
250}
251
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700252/*
253 * Replace item expected in radix tree by a new item, while holding tree lock.
254 */
255static int shmem_radix_tree_replace(struct address_space *mapping,
256 pgoff_t index, void *expected, void *replacement)
257{
258 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700259 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700260
261 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700262 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700263 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700264 if (!pslot)
265 return -ENOENT;
266 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700267 if (item != expected)
268 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700269 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700270 return 0;
271}
272
273/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
276 *
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
279 */
280static bool shmem_confirm_swap(struct address_space *mapping,
281 pgoff_t index, swp_entry_t swap)
282{
283 void *item;
284
285 rcu_read_lock();
286 item = radix_tree_lookup(&mapping->page_tree, index);
287 rcu_read_unlock();
288 return item == swp_to_radix_entry(swap);
289}
290
291/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700292 * Like add_to_page_cache_locked, but error if expected item has gone.
293 */
294static int shmem_add_to_page_cache(struct page *page,
295 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700296 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700297{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700298 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700299
Sasha Levin309381fea2014-01-23 15:52:54 -0800300 VM_BUG_ON_PAGE(!PageLocked(page), page);
301 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700302
Hugh Dickinsb065b432012-07-11 14:02:48 -0700303 page_cache_get(page);
304 page->mapping = mapping;
305 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700306
Hugh Dickinsb065b432012-07-11 14:02:48 -0700307 spin_lock_irq(&mapping->tree_lock);
308 if (!expected)
309 error = radix_tree_insert(&mapping->page_tree, index, page);
310 else
311 error = shmem_radix_tree_replace(mapping, index, expected,
312 page);
313 if (!error) {
314 mapping->nrpages++;
315 __inc_zone_page_state(page, NR_FILE_PAGES);
316 __inc_zone_page_state(page, NR_SHMEM);
317 spin_unlock_irq(&mapping->tree_lock);
318 } else {
319 page->mapping = NULL;
320 spin_unlock_irq(&mapping->tree_lock);
321 page_cache_release(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700322 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700323 return error;
324}
325
326/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700327 * Like delete_from_page_cache, but substitutes swap for page.
328 */
329static void shmem_delete_from_page_cache(struct page *page, void *radswap)
330{
331 struct address_space *mapping = page->mapping;
332 int error;
333
334 spin_lock_irq(&mapping->tree_lock);
335 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
336 page->mapping = NULL;
337 mapping->nrpages--;
338 __dec_zone_page_state(page, NR_FILE_PAGES);
339 __dec_zone_page_state(page, NR_SHMEM);
340 spin_unlock_irq(&mapping->tree_lock);
341 page_cache_release(page);
342 BUG_ON(error);
343}
344
345/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700346 * Remove swap entry from radix tree, free the swap and its page cache.
347 */
348static int shmem_free_swap(struct address_space *mapping,
349 pgoff_t index, void *radswap)
350{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700351 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700352
353 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700354 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700355 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700356 if (old != radswap)
357 return -ENOENT;
358 free_swap_and_cache(radix_to_swp_entry(radswap));
359 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700360}
361
362/*
Hugh Dickins24513262012-01-20 14:34:21 -0800363 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
364 */
365void shmem_unlock_mapping(struct address_space *mapping)
366{
367 struct pagevec pvec;
368 pgoff_t indices[PAGEVEC_SIZE];
369 pgoff_t index = 0;
370
371 pagevec_init(&pvec, 0);
372 /*
373 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
374 */
375 while (!mapping_unevictable(mapping)) {
376 /*
377 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
378 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
379 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700380 pvec.nr = find_get_entries(mapping, index,
381 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800382 if (!pvec.nr)
383 break;
384 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700385 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800386 check_move_unevictable_pages(pvec.pages, pvec.nr);
387 pagevec_release(&pvec);
388 cond_resched();
389 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700390}
391
392/*
393 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700394 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700395 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700396static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
397 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700399 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700401 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700402 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
403 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
404 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700405 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700406 pgoff_t indices[PAGEVEC_SIZE];
407 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700408 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700409 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700411 if (lend == -1)
412 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700413
414 pagevec_init(&pvec, 0);
415 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700416 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700417 pvec.nr = find_get_entries(mapping, index,
418 min(end - index, (pgoff_t)PAGEVEC_SIZE),
419 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700420 if (!pvec.nr)
421 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700422 for (i = 0; i < pagevec_count(&pvec); i++) {
423 struct page *page = pvec.pages[i];
424
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700425 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700426 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700427 break;
428
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700429 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700430 if (unfalloc)
431 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700432 nr_swaps_freed += !shmem_free_swap(mapping,
433 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700434 continue;
435 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700436
437 if (!trylock_page(page))
438 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700439 if (!unfalloc || !PageUptodate(page)) {
440 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800441 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700442 truncate_inode_page(mapping, page);
443 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700444 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700445 unlock_page(page);
446 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700447 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800448 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700449 cond_resched();
450 index++;
451 }
452
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700453 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700454 struct page *page = NULL;
455 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
456 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700457 unsigned int top = PAGE_CACHE_SIZE;
458 if (start > end) {
459 top = partial_end;
460 partial_end = 0;
461 }
462 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700463 set_page_dirty(page);
464 unlock_page(page);
465 page_cache_release(page);
466 }
467 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700468 if (partial_end) {
469 struct page *page = NULL;
470 shmem_getpage(inode, end, &page, SGP_READ, NULL);
471 if (page) {
472 zero_user_segment(page, 0, partial_end);
473 set_page_dirty(page);
474 unlock_page(page);
475 page_cache_release(page);
476 }
477 }
478 if (start >= end)
479 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700480
481 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700482 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700483 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700484
485 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700486 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700487 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700488 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700489 /* If all gone or hole-punch or unfalloc, we're done */
490 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700491 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700492 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700493 index = start;
494 continue;
495 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700496 for (i = 0; i < pagevec_count(&pvec); i++) {
497 struct page *page = pvec.pages[i];
498
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700499 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700500 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700501 break;
502
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700503 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700504 if (unfalloc)
505 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700506 if (shmem_free_swap(mapping, index, page)) {
507 /* Swap was replaced by page: retry */
508 index--;
509 break;
510 }
511 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700512 continue;
513 }
514
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700515 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700516 if (!unfalloc || !PageUptodate(page)) {
517 if (page->mapping == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800518 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700519 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700520 } else {
521 /* Page was replaced by swap: retry */
522 unlock_page(page);
523 index--;
524 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700525 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700526 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700527 unlock_page(page);
528 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700529 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800530 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700531 index++;
532 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700535 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 shmem_recalc_inode(inode);
537 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700538}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700540void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
541{
542 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700543 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700545EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Yu Zhao44a30222015-09-08 15:03:33 -0700547static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
548 struct kstat *stat)
549{
550 struct inode *inode = dentry->d_inode;
551 struct shmem_inode_info *info = SHMEM_I(inode);
552
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800553 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
554 spin_lock(&info->lock);
555 shmem_recalc_inode(inode);
556 spin_unlock(&info->lock);
557 }
Yu Zhao44a30222015-09-08 15:03:33 -0700558 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700559 return 0;
560}
561
Hugh Dickins94c1e622011-06-27 16:18:03 -0700562static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
David Howells75c3cfa2015-03-17 22:26:12 +0000564 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700565 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 int error;
567
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200568 error = inode_change_ok(inode, attr);
569 if (error)
570 return error;
571
Hugh Dickins94c1e622011-06-27 16:18:03 -0700572 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
573 loff_t oldsize = inode->i_size;
574 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000575
David Herrmann40e041a2014-08-08 14:25:27 -0700576 /* protected by i_mutex */
577 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
578 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
579 return -EPERM;
580
Hugh Dickins94c1e622011-06-27 16:18:03 -0700581 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700582 error = shmem_reacct_size(SHMEM_I(inode)->flags,
583 oldsize, newsize);
584 if (error)
585 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700586 i_size_write(inode, newsize);
587 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
588 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700589 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700590 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800591 if (oldsize > holebegin)
592 unmap_mapping_range(inode->i_mapping,
593 holebegin, 0, 1);
594 if (info->alloced)
595 shmem_truncate_range(inode,
596 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700597 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800598 if (oldsize > holebegin)
599 unmap_mapping_range(inode->i_mapping,
600 holebegin, 0, 1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 }
603
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200604 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200605 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -0800606 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 return error;
608}
609
Al Viro1f895f72010-06-05 19:10:41 -0400610static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 struct shmem_inode_info *info = SHMEM_I(inode);
613
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000614 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 shmem_unacct_size(info->flags, inode->i_size);
616 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000617 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800619 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800621 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700623 } else
624 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700625
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400626 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800627 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800628 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200629 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630}
631
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700632/*
633 * If swap found in inode, free it and move page from swapcache to filecache.
634 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700635static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700636 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700638 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700639 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700640 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700641 gfp_t gfp;
642 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700644 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700645 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700646 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -0700647 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800648
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800649 /*
650 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400651 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800652 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700653 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800654 */
655 if (shmem_swaplist.next != &info->swaplist)
656 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800657
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700658 gfp = mapping_gfp_mask(mapping);
659 if (shmem_should_replace_page(*pagep, gfp)) {
660 mutex_unlock(&shmem_swaplist_mutex);
661 error = shmem_replace_page(pagep, gfp, info, index);
662 mutex_lock(&shmem_swaplist_mutex);
663 /*
664 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700665 * allocation, but the inode might have been freed while we
666 * dropped it: although a racing shmem_evict_inode() cannot
667 * complete without emptying the radix_tree, our page lock
668 * on this swapcache page is not enough to prevent that -
669 * free_swap_and_cache() of our swap entry will only
670 * trylock_page(), removing swap from radix_tree whatever.
671 *
672 * We must not proceed to shmem_add_to_page_cache() if the
673 * inode has been freed, but of course we cannot rely on
674 * inode or mapping or info to check that. However, we can
675 * safely check if our swap entry is still in use (and here
676 * it can't have got reused for another page): if it's still
677 * in use, then the inode cannot have been freed yet, and we
678 * can safely proceed (if it's no longer in use, that tells
679 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700680 */
681 if (!page_swapcount(*pagep))
682 error = -ENOENT;
683 }
684
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800685 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700686 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
687 * but also to hold up shmem_evict_inode(): so inode cannot be freed
688 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800689 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700690 if (!error)
691 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700692 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700693 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700694 /*
695 * Truncation and eviction use free_swap_and_cache(), which
696 * only does trylock page: if we raced, best clean up here.
697 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700698 delete_from_swap_cache(*pagep);
699 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700700 if (!error) {
701 spin_lock(&info->lock);
702 info->swapped--;
703 spin_unlock(&info->lock);
704 swap_free(swap);
705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800707 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
710/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700711 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700713int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700715 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -0700717 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700718 int error = 0;
719
720 /*
721 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700722 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700723 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700724 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700725 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700726
727 /*
728 * Charge page using GFP_KERNEL while we can wait, before taking
729 * the shmem_swaplist_mutex which might hold up shmem_writepage().
730 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700731 */
Johannes Weiner00501b52014-08-08 14:19:20 -0700732 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg);
Hugh Dickins778dd892011-05-11 15:13:37 -0700733 if (error)
734 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700735 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -0700736 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800738 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700739 list_for_each_safe(this, next, &shmem_swaplist) {
740 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700741 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -0700742 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700743 else
744 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800745 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -0700746 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -0700747 break;
Johannes Weiner00501b52014-08-08 14:19:20 -0700748 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800750 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700751
Johannes Weiner00501b52014-08-08 14:19:20 -0700752 if (error) {
753 if (error != -ENOMEM)
754 error = 0;
755 mem_cgroup_cancel_charge(page, memcg);
756 } else
757 mem_cgroup_commit_charge(page, memcg, true);
Hugh Dickins778dd892011-05-11 15:13:37 -0700758out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800759 unlock_page(page);
760 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700761 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
764/*
765 * Move the page from the page cache to the swap cache.
766 */
767static int shmem_writepage(struct page *page, struct writeback_control *wbc)
768{
769 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700772 swp_entry_t swap;
773 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 mapping = page->mapping;
777 index = page->index;
778 inode = mapping->host;
779 info = SHMEM_I(inode);
780 if (info->flags & VM_LOCKED)
781 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800782 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 goto redirty;
784
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800785 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +0100786 * Our capabilities prevent regular writeback or sync from ever calling
787 * shmem_writepage; but a stacking filesystem might use ->writepage of
788 * its underlying filesystem, in which case tmpfs should write out to
789 * swap only in response to memory pressure, and not for the writeback
790 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800791 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700792 if (!wbc->for_reclaim) {
793 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
794 goto redirty;
795 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700796
797 /*
798 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
799 * value into swapfile.c, the only way we can correctly account for a
800 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700801 *
802 * That's okay for a page already fallocated earlier, but if we have
803 * not yet completed the fallocation, then (a) we want to keep track
804 * of this page in case we have to undo it, and (b) it may not be a
805 * good idea to continue anyway, once we're pushing into swap. So
806 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700807 */
808 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700809 if (inode->i_private) {
810 struct shmem_falloc *shmem_falloc;
811 spin_lock(&inode->i_lock);
812 shmem_falloc = inode->i_private;
813 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -0700814 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -0700815 index >= shmem_falloc->start &&
816 index < shmem_falloc->next)
817 shmem_falloc->nr_unswapped++;
818 else
819 shmem_falloc = NULL;
820 spin_unlock(&inode->i_lock);
821 if (shmem_falloc)
822 goto redirty;
823 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700824 clear_highpage(page);
825 flush_dcache_page(page);
826 SetPageUptodate(page);
827 }
828
Hugh Dickins48f170f2011-07-25 17:12:37 -0700829 swap = get_swap_page();
830 if (!swap.val)
831 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800832
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700833 /*
834 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700835 * if it's not already there. Do it now before the page is
836 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700837 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700838 * we've incremented swapped, because shmem_unuse_inode() will
839 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700840 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700841 mutex_lock(&shmem_swaplist_mutex);
842 if (list_empty(&info->swaplist))
843 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700844
Hugh Dickins48f170f2011-07-25 17:12:37 -0700845 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800846 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700847 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
848
849 spin_lock(&info->lock);
850 info->swapped++;
851 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700852 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700853
854 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800855 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700856 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return 0;
858 }
859
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700860 mutex_unlock(&shmem_swaplist_mutex);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700861 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862redirty:
863 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800864 if (wbc->for_reclaim)
865 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
866 unlock_page(page);
867 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868}
869
870#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800871#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700872static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800873{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700874 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800875
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700876 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700877 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800878
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800879 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800880
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700881 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800882}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700883
884static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
885{
886 struct mempolicy *mpol = NULL;
887 if (sbinfo->mpol) {
888 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
889 mpol = sbinfo->mpol;
890 mpol_get(mpol);
891 spin_unlock(&sbinfo->stat_lock);
892 }
893 return mpol;
894}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800895#endif /* CONFIG_TMPFS */
896
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700897static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
898 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800901 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800903 /* Create a pseudo vma that just contains the policy */
904 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700905 /* Bias interleave by inode number to distribute better across nodes */
906 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800907 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700908 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700909
Mel Gorman18a2f372012-12-05 14:01:41 -0800910 page = swapin_readahead(swap, gfp, &pvma, 0);
911
912 /* Drop reference taken by mpol_shared_policy_lookup() */
913 mpol_cond_put(pvma.vm_policy);
914
915 return page;
916}
917
918static struct page *shmem_alloc_page(gfp_t gfp,
919 struct shmem_inode_info *info, pgoff_t index)
920{
921 struct vm_area_struct pvma;
922 struct page *page;
923
924 /* Create a pseudo vma that just contains the policy */
925 pvma.vm_start = 0;
926 /* Bias interleave by inode number to distribute better across nodes */
927 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
928 pvma.vm_ops = NULL;
929 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
930
931 page = alloc_page_vma(gfp, &pvma, 0);
932
933 /* Drop reference taken by mpol_shared_policy_lookup() */
934 mpol_cond_put(pvma.vm_policy);
935
936 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800938#else /* !CONFIG_NUMA */
939#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700940static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800941{
942}
943#endif /* CONFIG_TMPFS */
944
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700945static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
946 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700948 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
Hugh Dickins02098fe2008-02-04 22:28:42 -0800951static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700952 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000954 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800956#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700958#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
959static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
960{
961 return NULL;
962}
963#endif
964
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700966 * When a page is moved from swapcache to shmem filecache (either by the
967 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
968 * shmem_unuse_inode()), it may have been read in earlier from swap, in
969 * ignorance of the mapping it belongs to. If that mapping has special
970 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
971 * we may need to copy to a suitable page before moving to filecache.
972 *
973 * In a future release, this may well be extended to respect cpuset and
974 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
975 * but for now it is a simple matter of zone.
976 */
977static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
978{
979 return page_zonenum(page) > gfp_zone(gfp);
980}
981
982static int shmem_replace_page(struct page **pagep, gfp_t gfp,
983 struct shmem_inode_info *info, pgoff_t index)
984{
985 struct page *oldpage, *newpage;
986 struct address_space *swap_mapping;
987 pgoff_t swap_index;
988 int error;
989
990 oldpage = *pagep;
991 swap_index = page_private(oldpage);
992 swap_mapping = page_mapping(oldpage);
993
994 /*
995 * We have arrived here because our zones are constrained, so don't
996 * limit chance of success by further cpuset and node constraints.
997 */
998 gfp &= ~GFP_CONSTRAINT_MASK;
999 newpage = shmem_alloc_page(gfp, info, index);
1000 if (!newpage)
1001 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001002
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001003 page_cache_get(newpage);
1004 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001005 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001006
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001007 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001008 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001009 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001010 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001011 SetPageSwapCache(newpage);
1012
1013 /*
1014 * Our caller will very soon move newpage out of swapcache, but it's
1015 * a nice clean interface for us to replace oldpage by newpage there.
1016 */
1017 spin_lock_irq(&swap_mapping->tree_lock);
1018 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1019 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001020 if (!error) {
1021 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1022 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1023 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001024 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001025
Hugh Dickins0142ef62012-06-07 14:21:09 -07001026 if (unlikely(error)) {
1027 /*
1028 * Is this possible? I think not, now that our callers check
1029 * both PageSwapCache and page_private after getting page lock;
1030 * but be defensive. Reverse old to newpage for clear and free.
1031 */
1032 oldpage = newpage;
1033 } else {
Hugh Dickins45637ba2015-11-05 18:49:40 -08001034 mem_cgroup_replace_page(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001035 lru_cache_add_anon(newpage);
1036 *pagep = newpage;
1037 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001038
1039 ClearPageSwapCache(oldpage);
1040 set_page_private(oldpage, 0);
1041
1042 unlock_page(oldpage);
1043 page_cache_release(oldpage);
1044 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001045 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001046}
1047
1048/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001049 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 *
1051 * If we allocate a new one we do not mark it dirty. That's up to the
1052 * vm. If we swap it in we mark it dirty since we also free the swap
1053 * entry since a page cannot live in both the swap and page cache
1054 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001055static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001056 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
1058 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001059 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 struct shmem_sb_info *sbinfo;
Johannes Weiner00501b52014-08-08 14:19:20 -07001061 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001062 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 swp_entry_t swap;
1064 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001065 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001066 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001068 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001071 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001072 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001073 if (radix_tree_exceptional_entry(page)) {
1074 swap = radix_to_swp_entry(page);
1075 page = NULL;
1076 }
1077
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001078 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001079 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1080 error = -EINVAL;
1081 goto failed;
1082 }
1083
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001084 if (page && sgp == SGP_WRITE)
1085 mark_page_accessed(page);
1086
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001087 /* fallocated page? */
1088 if (page && !PageUptodate(page)) {
1089 if (sgp != SGP_READ)
1090 goto clear;
1091 unlock_page(page);
1092 page_cache_release(page);
1093 page = NULL;
1094 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001095 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001096 *pagep = page;
1097 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001101 * Fast cache lookup did not find it:
1102 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001104 info = SHMEM_I(inode);
1105 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 if (swap.val) {
1108 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001109 page = lookup_swap_cache(swap);
1110 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001111 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001112 if (fault_type)
1113 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001114 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001115 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001116 error = -ENOMEM;
1117 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120
1121 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001122 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001123 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001124 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001125 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001126 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001127 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001128 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001130 goto failed;
1131 }
1132 wait_on_page_writeback(page);
1133
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001134 if (shmem_should_replace_page(page, gfp)) {
1135 error = shmem_replace_page(&page, gfp, info, index);
1136 if (error)
1137 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 }
1139
Johannes Weiner00501b52014-08-08 14:19:20 -07001140 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001141 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001142 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001143 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001144 /*
1145 * We already confirmed swap under page lock, and make
1146 * no memory allocation here, so usually no possibility
1147 * of error; but free_swap_and_cache() only trylocks a
1148 * page, so it is just possible that the entry has been
1149 * truncated or holepunched since swap was confirmed.
1150 * shmem_undo_range() will have done some of the
1151 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001152 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001153 * Reset swap.val? No, leave it so "failed" goes back to
1154 * "repeat": reading a hole and writing should succeed.
1155 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001156 if (error) {
1157 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001158 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001159 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001160 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001161 if (error)
1162 goto failed;
1163
Johannes Weiner00501b52014-08-08 14:19:20 -07001164 mem_cgroup_commit_charge(page, memcg, true);
1165
Hugh Dickins54af60422011-08-03 16:21:24 -07001166 spin_lock(&info->lock);
1167 info->swapped--;
1168 shmem_recalc_inode(inode);
1169 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001170
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001171 if (sgp == SGP_WRITE)
1172 mark_page_accessed(page);
1173
Hugh Dickins27ab7002011-07-25 17:12:36 -07001174 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001175 set_page_dirty(page);
1176 swap_free(swap);
1177
Hugh Dickins54af60422011-08-03 16:21:24 -07001178 } else {
1179 if (shmem_acct_block(info->flags)) {
1180 error = -ENOSPC;
1181 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001183 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001184 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001185 sbinfo->max_blocks) >= 0) {
1186 error = -ENOSPC;
1187 goto unacct;
1188 }
Tim Chen7e496292010-08-09 17:19:05 -07001189 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Hugh Dickins54af60422011-08-03 16:21:24 -07001192 page = shmem_alloc_page(gfp, info, index);
1193 if (!page) {
1194 error = -ENOMEM;
1195 goto decused;
1196 }
1197
Mel Gorman07a42782014-06-04 16:10:24 -07001198 __SetPageSwapBacked(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001199 __set_page_locked(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001200 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001201 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001202
Johannes Weiner00501b52014-08-08 14:19:20 -07001203 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg);
Hugh Dickins54af60422011-08-03 16:21:24 -07001204 if (error)
1205 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001206 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001207 if (!error) {
1208 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001209 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001210 radix_tree_preload_end();
1211 }
1212 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -07001213 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001214 goto decused;
1215 }
Johannes Weiner00501b52014-08-08 14:19:20 -07001216 mem_cgroup_commit_charge(page, memcg, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001217 lru_cache_add_anon(page);
1218
1219 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001221 inode->i_blocks += BLOCKS_PER_PAGE;
1222 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001224 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001225
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001226 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001227 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1228 */
1229 if (sgp == SGP_FALLOC)
1230 sgp = SGP_WRITE;
1231clear:
1232 /*
1233 * Let SGP_WRITE caller clear ends if write does not fill page;
1234 * but SGP_FALLOC on a page fallocated earlier must initialize
1235 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001236 */
1237 if (sgp != SGP_WRITE) {
1238 clear_highpage(page);
1239 flush_dcache_page(page);
1240 SetPageUptodate(page);
1241 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001242 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001243 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001244 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001245
Hugh Dickins54af60422011-08-03 16:21:24 -07001246 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001247 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001248 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1249 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001250 if (alloced)
1251 goto trunc;
1252 else
1253 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001254 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001255 *pagep = page;
1256 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001257
Nick Piggind0217ac2007-07-19 01:47:03 -07001258 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001259 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001261trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001262 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001263 ClearPageDirty(page);
1264 delete_from_page_cache(page);
1265 spin_lock(&info->lock);
1266 info->alloced--;
1267 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001269decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001270 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001271 if (sbinfo->max_blocks)
1272 percpu_counter_add(&sbinfo->used_blocks, -1);
1273unacct:
1274 shmem_unacct_blocks(info->flags, 1);
1275failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001276 if (swap.val && error != -EINVAL &&
1277 !shmem_confirm_swap(mapping, index, swap))
1278 error = -EEXIST;
1279unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001280 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001281 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001282 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001283 }
1284 if (error == -ENOSPC && !once++) {
1285 info = SHMEM_I(inode);
1286 spin_lock(&info->lock);
1287 shmem_recalc_inode(inode);
1288 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001289 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001290 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001291 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001292 goto repeat;
1293 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294}
1295
1296static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1297{
Al Viro496ad9a2013-01-23 17:07:38 -05001298 struct inode *inode = file_inode(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001300 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001302 /*
1303 * Trinity finds that probing a hole which tmpfs is punching can
1304 * prevent the hole-punch from ever completing: which in turn
1305 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001306 * faulting pages into the hole while it's being punched. Although
1307 * shmem_undo_range() does remove the additions, it may be unable to
1308 * keep up, as each new page needs its own unmap_mapping_range() call,
1309 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1310 *
1311 * It does not matter if we sometimes reach this check just before the
1312 * hole-punch begins, so that one fault then races with the punch:
1313 * we just need to make racing faults a rare case.
1314 *
1315 * The implementation below would be much simpler if we just used a
1316 * standard mutex or completion: but we cannot take i_mutex in fault,
1317 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001318 */
1319 if (unlikely(inode->i_private)) {
1320 struct shmem_falloc *shmem_falloc;
1321
1322 spin_lock(&inode->i_lock);
1323 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001324 if (shmem_falloc &&
1325 shmem_falloc->waitq &&
1326 vmf->pgoff >= shmem_falloc->start &&
1327 vmf->pgoff < shmem_falloc->next) {
1328 wait_queue_head_t *shmem_falloc_waitq;
1329 DEFINE_WAIT(shmem_fault_wait);
1330
1331 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001332 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1333 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001334 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001335 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001336 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001337 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001338
1339 shmem_falloc_waitq = shmem_falloc->waitq;
1340 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1341 TASK_UNINTERRUPTIBLE);
1342 spin_unlock(&inode->i_lock);
1343 schedule();
1344
1345 /*
1346 * shmem_falloc_waitq points into the shmem_fallocate()
1347 * stack of the hole-punching task: shmem_falloc_waitq
1348 * is usually invalid by the time we reach here, but
1349 * finish_wait() does not dereference it in that case;
1350 * though i_lock needed lest racing with wake_up_all().
1351 */
1352 spin_lock(&inode->i_lock);
1353 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1354 spin_unlock(&inode->i_lock);
1355 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001356 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001357 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001358 }
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1361 if (error)
1362 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001363
Ying Han456f9982011-05-26 16:25:38 -07001364 if (ret & VM_FAULT_MAJOR) {
1365 count_vm_event(PGMAJFAULT);
1366 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1367 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001368 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369}
1370
1371#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001372static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
Al Viro496ad9a2013-01-23 17:07:38 -05001374 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001375 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001378static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1379 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380{
Al Viro496ad9a2013-01-23 17:07:38 -05001381 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001382 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001384 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1385 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386}
1387#endif
1388
1389int shmem_lock(struct file *file, int lock, struct user_struct *user)
1390{
Al Viro496ad9a2013-01-23 17:07:38 -05001391 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 struct shmem_inode_info *info = SHMEM_I(inode);
1393 int retval = -ENOMEM;
1394
1395 spin_lock(&info->lock);
1396 if (lock && !(info->flags & VM_LOCKED)) {
1397 if (!user_shm_lock(inode->i_size, user))
1398 goto out_nomem;
1399 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001400 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
1402 if (!lock && (info->flags & VM_LOCKED) && user) {
1403 user_shm_unlock(inode->i_size, user);
1404 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001405 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 }
1407 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409out_nomem:
1410 spin_unlock(&info->lock);
1411 return retval;
1412}
1413
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001414static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415{
1416 file_accessed(file);
1417 vma->vm_ops = &shmem_vm_ops;
1418 return 0;
1419}
1420
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001421static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001422 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423{
1424 struct inode *inode;
1425 struct shmem_inode_info *info;
1426 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1427
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001428 if (shmem_reserve_inode(sb))
1429 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 inode = new_inode(sb);
1432 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001433 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001434 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001437 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 info = SHMEM_I(inode);
1439 memset(info, 0, (char *)inode - (char *)info);
1440 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07001441 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001442 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001444 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001445 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
1447 switch (mode & S_IFMT) {
1448 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001449 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 init_special_inode(inode, mode, dev);
1451 break;
1452 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001453 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 inode->i_op = &shmem_inode_operations;
1455 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001456 mpol_shared_policy_init(&info->policy,
1457 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 break;
1459 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001460 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 /* Some things misbehave if size == 0 on a directory */
1462 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1463 inode->i_op = &shmem_dir_inode_operations;
1464 inode->i_fop = &simple_dir_operations;
1465 break;
1466 case S_IFLNK:
1467 /*
1468 * Must not load anything in the rbtree,
1469 * mpol_free_shared_policy will not be called.
1470 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001471 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 break;
1473 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001474 } else
1475 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 return inode;
1477}
1478
Johannes Weiner0cd61442014-04-03 14:47:46 -07001479bool shmem_mapping(struct address_space *mapping)
1480{
Sasha Levinf0774d82015-02-23 05:38:00 -05001481 if (!mapping->host)
1482 return false;
1483
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001484 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001485}
1486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001488static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001489static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001491#ifdef CONFIG_TMPFS_XATTR
1492static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1493#else
1494#define shmem_initxattrs NULL
1495#endif
1496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001498shmem_write_begin(struct file *file, struct address_space *mapping,
1499 loff_t pos, unsigned len, unsigned flags,
1500 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501{
Nick Piggin800d15a2007-10-16 01:25:03 -07001502 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07001503 struct shmem_inode_info *info = SHMEM_I(inode);
Nick Piggin800d15a2007-10-16 01:25:03 -07001504 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07001505
1506 /* i_mutex is held by caller */
1507 if (unlikely(info->seals)) {
1508 if (info->seals & F_SEAL_WRITE)
1509 return -EPERM;
1510 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1511 return -EPERM;
1512 }
1513
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001514 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
Nick Piggin800d15a2007-10-16 01:25:03 -07001515}
1516
1517static int
1518shmem_write_end(struct file *file, struct address_space *mapping,
1519 loff_t pos, unsigned len, unsigned copied,
1520 struct page *page, void *fsdata)
1521{
1522 struct inode *inode = mapping->host;
1523
Hugh Dickinsd3602442008-02-04 22:28:44 -08001524 if (pos + copied > inode->i_size)
1525 i_size_write(inode, pos + copied);
1526
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001527 if (!PageUptodate(page)) {
1528 if (copied < PAGE_CACHE_SIZE) {
1529 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1530 zero_user_segments(page, 0, from,
1531 from + copied, PAGE_CACHE_SIZE);
1532 }
1533 SetPageUptodate(page);
1534 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001535 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001536 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001537 page_cache_release(page);
1538
Nick Piggin800d15a2007-10-16 01:25:03 -07001539 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540}
1541
Al Viro2ba5bbe2014-04-02 20:00:02 -04001542static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543{
Al Viro6e58e792014-02-03 17:07:03 -05001544 struct file *file = iocb->ki_filp;
1545 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001547 pgoff_t index;
1548 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001549 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02001550 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05001551 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05001552 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001553
1554 /*
1555 * Might this read be for a stacking filesystem? Then when reading
1556 * holes of a sparse file, we actually need to allocate those pages,
1557 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1558 */
Al Viro777eda22014-12-17 04:46:46 -05001559 if (!iter_is_iovec(to))
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001560 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 index = *ppos >> PAGE_CACHE_SHIFT;
1563 offset = *ppos & ~PAGE_CACHE_MASK;
1564
1565 for (;;) {
1566 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001567 pgoff_t end_index;
1568 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 loff_t i_size = i_size_read(inode);
1570
1571 end_index = i_size >> PAGE_CACHE_SHIFT;
1572 if (index > end_index)
1573 break;
1574 if (index == end_index) {
1575 nr = i_size & ~PAGE_CACHE_MASK;
1576 if (nr <= offset)
1577 break;
1578 }
1579
Al Viro6e58e792014-02-03 17:07:03 -05001580 error = shmem_getpage(inode, index, &page, sgp, NULL);
1581 if (error) {
1582 if (error == -EINVAL)
1583 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 break;
1585 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001586 if (page)
1587 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 /*
1590 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001591 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 */
1593 nr = PAGE_CACHE_SIZE;
1594 i_size = i_size_read(inode);
1595 end_index = i_size >> PAGE_CACHE_SHIFT;
1596 if (index == end_index) {
1597 nr = i_size & ~PAGE_CACHE_MASK;
1598 if (nr <= offset) {
1599 if (page)
1600 page_cache_release(page);
1601 break;
1602 }
1603 }
1604 nr -= offset;
1605
1606 if (page) {
1607 /*
1608 * If users can be writing to this page using arbitrary
1609 * virtual addresses, take care about potential aliasing
1610 * before reading the page on the kernel side.
1611 */
1612 if (mapping_writably_mapped(mapping))
1613 flush_dcache_page(page);
1614 /*
1615 * Mark the page accessed if we read the beginning.
1616 */
1617 if (!offset)
1618 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001619 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001621 page_cache_get(page);
1622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624 /*
1625 * Ok, we have the page, and it's up-to-date, so
1626 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04001628 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05001629 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 offset += ret;
1631 index += offset >> PAGE_CACHE_SHIFT;
1632 offset &= ~PAGE_CACHE_MASK;
1633
1634 page_cache_release(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04001635 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 break;
Al Viro6e58e792014-02-03 17:07:03 -05001637 if (ret < nr) {
1638 error = -EFAULT;
1639 break;
1640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 cond_resched();
1642 }
1643
1644 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05001645 file_accessed(file);
1646 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647}
1648
Hugh Dickins708e3502011-07-25 17:12:32 -07001649static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1650 struct pipe_inode_info *pipe, size_t len,
1651 unsigned int flags)
1652{
1653 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001654 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001655 unsigned int loff, nr_pages, req_pages;
1656 struct page *pages[PIPE_DEF_BUFFERS];
1657 struct partial_page partial[PIPE_DEF_BUFFERS];
1658 struct page *page;
1659 pgoff_t index, end_index;
1660 loff_t isize, left;
1661 int error, page_nr;
1662 struct splice_pipe_desc spd = {
1663 .pages = pages,
1664 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001665 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001666 .flags = flags,
1667 .ops = &page_cache_pipe_buf_ops,
1668 .spd_release = spd_release_page,
1669 };
1670
Hugh Dickins71f0e072011-07-25 17:12:33 -07001671 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001672 if (unlikely(*ppos >= isize))
1673 return 0;
1674
1675 left = isize - *ppos;
1676 if (unlikely(left < len))
1677 len = left;
1678
1679 if (splice_grow_spd(pipe, &spd))
1680 return -ENOMEM;
1681
1682 index = *ppos >> PAGE_CACHE_SHIFT;
1683 loff = *ppos & ~PAGE_CACHE_MASK;
1684 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04001685 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07001686
Hugh Dickins708e3502011-07-25 17:12:32 -07001687 spd.nr_pages = find_get_pages_contig(mapping, index,
1688 nr_pages, spd.pages);
1689 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001690 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001691
Hugh Dickins708e3502011-07-25 17:12:32 -07001692 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001693 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1694 if (error)
1695 break;
1696 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001697 spd.pages[spd.nr_pages++] = page;
1698 index++;
1699 }
1700
Hugh Dickins708e3502011-07-25 17:12:32 -07001701 index = *ppos >> PAGE_CACHE_SHIFT;
1702 nr_pages = spd.nr_pages;
1703 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001704
Hugh Dickins708e3502011-07-25 17:12:32 -07001705 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1706 unsigned int this_len;
1707
1708 if (!len)
1709 break;
1710
Hugh Dickins708e3502011-07-25 17:12:32 -07001711 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1712 page = spd.pages[page_nr];
1713
Hugh Dickins71f0e072011-07-25 17:12:33 -07001714 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001715 error = shmem_getpage(inode, index, &page,
1716 SGP_CACHE, NULL);
1717 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001718 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001719 unlock_page(page);
1720 page_cache_release(spd.pages[page_nr]);
1721 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001722 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001723
1724 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001725 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1726 if (unlikely(!isize || index > end_index))
1727 break;
1728
Hugh Dickins708e3502011-07-25 17:12:32 -07001729 if (end_index == index) {
1730 unsigned int plen;
1731
Hugh Dickins708e3502011-07-25 17:12:32 -07001732 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1733 if (plen <= loff)
1734 break;
1735
Hugh Dickins708e3502011-07-25 17:12:32 -07001736 this_len = min(this_len, plen - loff);
1737 len = this_len;
1738 }
1739
1740 spd.partial[page_nr].offset = loff;
1741 spd.partial[page_nr].len = this_len;
1742 len -= this_len;
1743 loff = 0;
1744 spd.nr_pages++;
1745 index++;
1746 }
1747
Hugh Dickins708e3502011-07-25 17:12:32 -07001748 while (page_nr < nr_pages)
1749 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001750
1751 if (spd.nr_pages)
1752 error = splice_to_pipe(pipe, &spd);
1753
Eric Dumazet047fe362012-06-12 15:24:40 +02001754 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001755
1756 if (error > 0) {
1757 *ppos += error;
1758 file_accessed(in);
1759 }
1760 return error;
1761}
1762
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001763/*
1764 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1765 */
1766static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001767 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001768{
1769 struct page *page;
1770 struct pagevec pvec;
1771 pgoff_t indices[PAGEVEC_SIZE];
1772 bool done = false;
1773 int i;
1774
1775 pagevec_init(&pvec, 0);
1776 pvec.nr = 1; /* start small: we may be there already */
1777 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001778 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001779 pvec.nr, pvec.pages, indices);
1780 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001781 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001782 index = end;
1783 break;
1784 }
1785 for (i = 0; i < pvec.nr; i++, index++) {
1786 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001787 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001788 done = true;
1789 break;
1790 }
1791 index = indices[i];
1792 }
1793 page = pvec.pages[i];
1794 if (page && !radix_tree_exceptional_entry(page)) {
1795 if (!PageUptodate(page))
1796 page = NULL;
1797 }
1798 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001799 (page && whence == SEEK_DATA) ||
1800 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001801 done = true;
1802 break;
1803 }
1804 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001805 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001806 pagevec_release(&pvec);
1807 pvec.nr = PAGEVEC_SIZE;
1808 cond_resched();
1809 }
1810 return index;
1811}
1812
Andrew Morton965c8e52012-12-17 15:59:39 -08001813static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001814{
1815 struct address_space *mapping = file->f_mapping;
1816 struct inode *inode = mapping->host;
1817 pgoff_t start, end;
1818 loff_t new_offset;
1819
Andrew Morton965c8e52012-12-17 15:59:39 -08001820 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1821 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001822 MAX_LFS_FILESIZE, i_size_read(inode));
1823 mutex_lock(&inode->i_mutex);
1824 /* We're holding i_mutex so we can access i_size directly */
1825
1826 if (offset < 0)
1827 offset = -EINVAL;
1828 else if (offset >= inode->i_size)
1829 offset = -ENXIO;
1830 else {
1831 start = offset >> PAGE_CACHE_SHIFT;
1832 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001833 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001834 new_offset <<= PAGE_CACHE_SHIFT;
1835 if (new_offset > offset) {
1836 if (new_offset < inode->i_size)
1837 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001838 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001839 offset = -ENXIO;
1840 else
1841 offset = inode->i_size;
1842 }
1843 }
1844
Hugh Dickins387aae62013-08-04 11:30:25 -07001845 if (offset >= 0)
1846 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001847 mutex_unlock(&inode->i_mutex);
1848 return offset;
1849}
1850
David Herrmann05f65b52014-08-08 14:25:36 -07001851/*
1852 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1853 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1854 */
1855#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1856#define LAST_SCAN 4 /* about 150ms max */
1857
1858static void shmem_tag_pins(struct address_space *mapping)
1859{
1860 struct radix_tree_iter iter;
1861 void **slot;
1862 pgoff_t start;
1863 struct page *page;
1864
1865 lru_add_drain();
1866 start = 0;
1867 rcu_read_lock();
1868
1869restart:
1870 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1871 page = radix_tree_deref_slot(slot);
1872 if (!page || radix_tree_exception(page)) {
1873 if (radix_tree_deref_retry(page))
1874 goto restart;
1875 } else if (page_count(page) - page_mapcount(page) > 1) {
1876 spin_lock_irq(&mapping->tree_lock);
1877 radix_tree_tag_set(&mapping->page_tree, iter.index,
1878 SHMEM_TAG_PINNED);
1879 spin_unlock_irq(&mapping->tree_lock);
1880 }
1881
1882 if (need_resched()) {
1883 cond_resched_rcu();
1884 start = iter.index + 1;
1885 goto restart;
1886 }
1887 }
1888 rcu_read_unlock();
1889}
1890
1891/*
1892 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1893 * via get_user_pages(), drivers might have some pending I/O without any active
1894 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1895 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1896 * them to be dropped.
1897 * The caller must guarantee that no new user will acquire writable references
1898 * to those pages to avoid races.
1899 */
David Herrmann40e041a2014-08-08 14:25:27 -07001900static int shmem_wait_for_pins(struct address_space *mapping)
1901{
David Herrmann05f65b52014-08-08 14:25:36 -07001902 struct radix_tree_iter iter;
1903 void **slot;
1904 pgoff_t start;
1905 struct page *page;
1906 int error, scan;
1907
1908 shmem_tag_pins(mapping);
1909
1910 error = 0;
1911 for (scan = 0; scan <= LAST_SCAN; scan++) {
1912 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1913 break;
1914
1915 if (!scan)
1916 lru_add_drain_all();
1917 else if (schedule_timeout_killable((HZ << scan) / 200))
1918 scan = LAST_SCAN;
1919
1920 start = 0;
1921 rcu_read_lock();
1922restart:
1923 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
1924 start, SHMEM_TAG_PINNED) {
1925
1926 page = radix_tree_deref_slot(slot);
1927 if (radix_tree_exception(page)) {
1928 if (radix_tree_deref_retry(page))
1929 goto restart;
1930
1931 page = NULL;
1932 }
1933
1934 if (page &&
1935 page_count(page) - page_mapcount(page) != 1) {
1936 if (scan < LAST_SCAN)
1937 goto continue_resched;
1938
1939 /*
1940 * On the last scan, we clean up all those tags
1941 * we inserted; but make a note that we still
1942 * found pages pinned.
1943 */
1944 error = -EBUSY;
1945 }
1946
1947 spin_lock_irq(&mapping->tree_lock);
1948 radix_tree_tag_clear(&mapping->page_tree,
1949 iter.index, SHMEM_TAG_PINNED);
1950 spin_unlock_irq(&mapping->tree_lock);
1951continue_resched:
1952 if (need_resched()) {
1953 cond_resched_rcu();
1954 start = iter.index + 1;
1955 goto restart;
1956 }
1957 }
1958 rcu_read_unlock();
1959 }
1960
1961 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07001962}
1963
1964#define F_ALL_SEALS (F_SEAL_SEAL | \
1965 F_SEAL_SHRINK | \
1966 F_SEAL_GROW | \
1967 F_SEAL_WRITE)
1968
1969int shmem_add_seals(struct file *file, unsigned int seals)
1970{
1971 struct inode *inode = file_inode(file);
1972 struct shmem_inode_info *info = SHMEM_I(inode);
1973 int error;
1974
1975 /*
1976 * SEALING
1977 * Sealing allows multiple parties to share a shmem-file but restrict
1978 * access to a specific subset of file operations. Seals can only be
1979 * added, but never removed. This way, mutually untrusted parties can
1980 * share common memory regions with a well-defined policy. A malicious
1981 * peer can thus never perform unwanted operations on a shared object.
1982 *
1983 * Seals are only supported on special shmem-files and always affect
1984 * the whole underlying inode. Once a seal is set, it may prevent some
1985 * kinds of access to the file. Currently, the following seals are
1986 * defined:
1987 * SEAL_SEAL: Prevent further seals from being set on this file
1988 * SEAL_SHRINK: Prevent the file from shrinking
1989 * SEAL_GROW: Prevent the file from growing
1990 * SEAL_WRITE: Prevent write access to the file
1991 *
1992 * As we don't require any trust relationship between two parties, we
1993 * must prevent seals from being removed. Therefore, sealing a file
1994 * only adds a given set of seals to the file, it never touches
1995 * existing seals. Furthermore, the "setting seals"-operation can be
1996 * sealed itself, which basically prevents any further seal from being
1997 * added.
1998 *
1999 * Semantics of sealing are only defined on volatile files. Only
2000 * anonymous shmem files support sealing. More importantly, seals are
2001 * never written to disk. Therefore, there's no plan to support it on
2002 * other file types.
2003 */
2004
2005 if (file->f_op != &shmem_file_operations)
2006 return -EINVAL;
2007 if (!(file->f_mode & FMODE_WRITE))
2008 return -EPERM;
2009 if (seals & ~(unsigned int)F_ALL_SEALS)
2010 return -EINVAL;
2011
2012 mutex_lock(&inode->i_mutex);
2013
2014 if (info->seals & F_SEAL_SEAL) {
2015 error = -EPERM;
2016 goto unlock;
2017 }
2018
2019 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2020 error = mapping_deny_writable(file->f_mapping);
2021 if (error)
2022 goto unlock;
2023
2024 error = shmem_wait_for_pins(file->f_mapping);
2025 if (error) {
2026 mapping_allow_writable(file->f_mapping);
2027 goto unlock;
2028 }
2029 }
2030
2031 info->seals |= seals;
2032 error = 0;
2033
2034unlock:
2035 mutex_unlock(&inode->i_mutex);
2036 return error;
2037}
2038EXPORT_SYMBOL_GPL(shmem_add_seals);
2039
2040int shmem_get_seals(struct file *file)
2041{
2042 if (file->f_op != &shmem_file_operations)
2043 return -EINVAL;
2044
2045 return SHMEM_I(file_inode(file))->seals;
2046}
2047EXPORT_SYMBOL_GPL(shmem_get_seals);
2048
2049long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2050{
2051 long error;
2052
2053 switch (cmd) {
2054 case F_ADD_SEALS:
2055 /* disallow upper 32bit */
2056 if (arg > UINT_MAX)
2057 return -EINVAL;
2058
2059 error = shmem_add_seals(file, arg);
2060 break;
2061 case F_GET_SEALS:
2062 error = shmem_get_seals(file);
2063 break;
2064 default:
2065 error = -EINVAL;
2066 break;
2067 }
2068
2069 return error;
2070}
2071
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002072static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2073 loff_t len)
2074{
Al Viro496ad9a2013-01-23 17:07:38 -05002075 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002076 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002077 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002078 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002079 pgoff_t start, index, end;
2080 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002081
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002082 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2083 return -EOPNOTSUPP;
2084
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002085 mutex_lock(&inode->i_mutex);
2086
2087 if (mode & FALLOC_FL_PUNCH_HOLE) {
2088 struct address_space *mapping = file->f_mapping;
2089 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2090 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002091 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002092
David Herrmann40e041a2014-08-08 14:25:27 -07002093 /* protected by i_mutex */
2094 if (info->seals & F_SEAL_WRITE) {
2095 error = -EPERM;
2096 goto out;
2097 }
2098
Hugh Dickins8e205f72014-07-23 14:00:10 -07002099 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002100 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2101 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2102 spin_lock(&inode->i_lock);
2103 inode->i_private = &shmem_falloc;
2104 spin_unlock(&inode->i_lock);
2105
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002106 if ((u64)unmap_end > (u64)unmap_start)
2107 unmap_mapping_range(mapping, unmap_start,
2108 1 + unmap_end - unmap_start, 0);
2109 shmem_truncate_range(inode, offset, offset + len - 1);
2110 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002111
2112 spin_lock(&inode->i_lock);
2113 inode->i_private = NULL;
2114 wake_up_all(&shmem_falloc_waitq);
2115 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002116 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002117 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002118 }
2119
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002120 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2121 error = inode_newsize_ok(inode, offset + len);
2122 if (error)
2123 goto out;
2124
David Herrmann40e041a2014-08-08 14:25:27 -07002125 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2126 error = -EPERM;
2127 goto out;
2128 }
2129
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002130 start = offset >> PAGE_CACHE_SHIFT;
2131 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2132 /* Try to avoid a swapstorm if len is impossible to satisfy */
2133 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2134 error = -ENOSPC;
2135 goto out;
2136 }
2137
Hugh Dickins8e205f72014-07-23 14:00:10 -07002138 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002139 shmem_falloc.start = start;
2140 shmem_falloc.next = start;
2141 shmem_falloc.nr_falloced = 0;
2142 shmem_falloc.nr_unswapped = 0;
2143 spin_lock(&inode->i_lock);
2144 inode->i_private = &shmem_falloc;
2145 spin_unlock(&inode->i_lock);
2146
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002147 for (index = start; index < end; index++) {
2148 struct page *page;
2149
2150 /*
2151 * Good, the fallocate(2) manpage permits EINTR: we may have
2152 * been interrupted because we are using up too much memory.
2153 */
2154 if (signal_pending(current))
2155 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002156 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2157 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002158 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002159 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002160 NULL);
2161 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002162 /* Remove the !PageUptodate pages we added */
2163 shmem_undo_range(inode,
2164 (loff_t)start << PAGE_CACHE_SHIFT,
2165 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002166 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002167 }
2168
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002169 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002170 * Inform shmem_writepage() how far we have reached.
2171 * No need for lock or barrier: we have the page lock.
2172 */
2173 shmem_falloc.next++;
2174 if (!PageUptodate(page))
2175 shmem_falloc.nr_falloced++;
2176
2177 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002178 * If !PageUptodate, leave it that way so that freeable pages
2179 * can be recognized if we need to rollback on error later.
2180 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002181 * than free the pages we are allocating (and SGP_CACHE pages
2182 * might still be clean: we now need to mark those dirty too).
2183 */
2184 set_page_dirty(page);
2185 unlock_page(page);
2186 page_cache_release(page);
2187 cond_resched();
2188 }
2189
2190 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2191 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002192 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002193undone:
2194 spin_lock(&inode->i_lock);
2195 inode->i_private = NULL;
2196 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002197out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002198 mutex_unlock(&inode->i_mutex);
2199 return error;
2200}
2201
David Howells726c3342006-06-23 02:02:58 -07002202static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203{
David Howells726c3342006-06-23 02:02:58 -07002204 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206 buf->f_type = TMPFS_MAGIC;
2207 buf->f_bsize = PAGE_CACHE_SIZE;
2208 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002209 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002211 buf->f_bavail =
2212 buf->f_bfree = sbinfo->max_blocks -
2213 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002214 }
2215 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 buf->f_files = sbinfo->max_inodes;
2217 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 }
2219 /* else leave those fields 0 like simple_statfs */
2220 return 0;
2221}
2222
2223/*
2224 * File creation. Allocate an inode, and we're done..
2225 */
2226static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002227shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002229 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 int error = -ENOSPC;
2231
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002232 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002234 error = simple_acl_create(dir, inode);
2235 if (error)
2236 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002237 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002238 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002239 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002240 if (error && error != -EOPNOTSUPP)
2241 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002242
Al Viro718deb62009-12-16 19:35:36 -05002243 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 dir->i_size += BOGO_DIRENT_SIZE;
2245 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2246 d_instantiate(dentry, inode);
2247 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 }
2249 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002250out_iput:
2251 iput(inode);
2252 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253}
2254
Al Viro60545d02013-06-07 01:20:27 -04002255static int
2256shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2257{
2258 struct inode *inode;
2259 int error = -ENOSPC;
2260
2261 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2262 if (inode) {
2263 error = security_inode_init_security(inode, dir,
2264 NULL,
2265 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002266 if (error && error != -EOPNOTSUPP)
2267 goto out_iput;
2268 error = simple_acl_create(dir, inode);
2269 if (error)
2270 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002271 d_tmpfile(dentry, inode);
2272 }
2273 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002274out_iput:
2275 iput(inode);
2276 return error;
Al Viro60545d02013-06-07 01:20:27 -04002277}
2278
Al Viro18bb1db2011-07-26 01:41:39 -04002279static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280{
2281 int error;
2282
2283 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2284 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002285 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 return 0;
2287}
2288
Al Viro4acdaf22011-07-26 01:42:34 -04002289static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002290 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
2292 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2293}
2294
2295/*
2296 * Link a file..
2297 */
2298static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2299{
David Howells75c3cfa2015-03-17 22:26:12 +00002300 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002301 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
2303 /*
2304 * No ordinary (disk based) filesystem counts links as inodes;
2305 * but each new link needs a new dentry, pinning lowmem, and
2306 * tmpfs dentries cannot be pruned until they are unlinked.
2307 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002308 ret = shmem_reserve_inode(inode->i_sb);
2309 if (ret)
2310 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
2312 dir->i_size += BOGO_DIRENT_SIZE;
2313 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002314 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002315 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 dget(dentry); /* Extra pinning count for the created dentry */
2317 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002318out:
2319 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320}
2321
2322static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2323{
David Howells75c3cfa2015-03-17 22:26:12 +00002324 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002326 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2327 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
2329 dir->i_size -= BOGO_DIRENT_SIZE;
2330 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002331 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 dput(dentry); /* Undo the count from "create" - this does all the work */
2333 return 0;
2334}
2335
2336static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2337{
2338 if (!simple_empty(dentry))
2339 return -ENOTEMPTY;
2340
David Howells75c3cfa2015-03-17 22:26:12 +00002341 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002342 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 return shmem_unlink(dir, dentry);
2344}
2345
Miklos Szeredi37456772014-07-23 15:15:34 +02002346static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2347{
David Howellse36cb0b2015-01-29 12:02:35 +00002348 bool old_is_dir = d_is_dir(old_dentry);
2349 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002350
2351 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2352 if (old_is_dir) {
2353 drop_nlink(old_dir);
2354 inc_nlink(new_dir);
2355 } else {
2356 drop_nlink(new_dir);
2357 inc_nlink(old_dir);
2358 }
2359 }
2360 old_dir->i_ctime = old_dir->i_mtime =
2361 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002362 d_inode(old_dentry)->i_ctime =
2363 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
Miklos Szeredi37456772014-07-23 15:15:34 +02002364
2365 return 0;
2366}
2367
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002368static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2369{
2370 struct dentry *whiteout;
2371 int error;
2372
2373 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2374 if (!whiteout)
2375 return -ENOMEM;
2376
2377 error = shmem_mknod(old_dir, whiteout,
2378 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2379 dput(whiteout);
2380 if (error)
2381 return error;
2382
2383 /*
2384 * Cheat and hash the whiteout while the old dentry is still in
2385 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2386 *
2387 * d_lookup() will consistently find one of them at this point,
2388 * not sure which one, but that isn't even important.
2389 */
2390 d_rehash(whiteout);
2391 return 0;
2392}
2393
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394/*
2395 * The VFS layer already does all the dentry stuff for rename,
2396 * we just have to decrement the usage count for the target if
2397 * it exists so that the VFS layer correctly free's it when it
2398 * gets overwritten.
2399 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002400static 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 -07002401{
David Howells75c3cfa2015-03-17 22:26:12 +00002402 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 int they_are_dirs = S_ISDIR(inode->i_mode);
2404
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002405 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002406 return -EINVAL;
2407
Miklos Szeredi37456772014-07-23 15:15:34 +02002408 if (flags & RENAME_EXCHANGE)
2409 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2410
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 if (!simple_empty(new_dentry))
2412 return -ENOTEMPTY;
2413
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002414 if (flags & RENAME_WHITEOUT) {
2415 int error;
2416
2417 error = shmem_whiteout(old_dir, old_dentry);
2418 if (error)
2419 return error;
2420 }
2421
David Howells75c3cfa2015-03-17 22:26:12 +00002422 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002424 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002425 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002426 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002427 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002429 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002430 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 }
2432
2433 old_dir->i_size -= BOGO_DIRENT_SIZE;
2434 new_dir->i_size += BOGO_DIRENT_SIZE;
2435 old_dir->i_ctime = old_dir->i_mtime =
2436 new_dir->i_ctime = new_dir->i_mtime =
2437 inode->i_ctime = CURRENT_TIME;
2438 return 0;
2439}
2440
2441static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2442{
2443 int error;
2444 int len;
2445 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002446 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 char *kaddr;
2448 struct shmem_inode_info *info;
2449
2450 len = strlen(symname) + 1;
2451 if (len > PAGE_CACHE_SIZE)
2452 return -ENAMETOOLONG;
2453
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002454 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 if (!inode)
2456 return -ENOSPC;
2457
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002458 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002459 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002460 if (error) {
2461 if (error != -EOPNOTSUPP) {
2462 iput(inode);
2463 return error;
2464 }
2465 error = 0;
2466 }
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 info = SHMEM_I(inode);
2469 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002470 if (len <= SHORT_SYMLINK_LEN) {
2471 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2472 if (!info->symlink) {
2473 iput(inode);
2474 return -ENOMEM;
2475 }
2476 inode->i_op = &shmem_short_symlink_operations;
Al Viro60380f12015-05-02 10:24:43 -04002477 inode->i_link = info->symlink;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 } else {
2479 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2480 if (error) {
2481 iput(inode);
2482 return error;
2483 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002484 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002486 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002488 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002489 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002491 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 page_cache_release(page);
2493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 dir->i_size += BOGO_DIRENT_SIZE;
2495 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2496 d_instantiate(dentry, inode);
2497 dget(dentry);
2498 return 0;
2499}
2500
Al Viro6e771372015-05-02 13:37:52 -04002501static const char *shmem_follow_link(struct dentry *dentry, void **cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502{
2503 struct page *page = NULL;
David Howells75c3cfa2015-03-17 22:26:12 +00002504 int error = shmem_getpage(d_inode(dentry), 0, &page, SGP_READ, NULL);
Al Viro680baac2015-05-02 13:32:22 -04002505 if (error)
2506 return ERR_PTR(error);
2507 unlock_page(page);
2508 *cookie = page;
2509 return kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510}
2511
Al Viro5f2c4172015-05-07 11:14:26 -04002512static void shmem_put_link(struct inode *unused, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513{
Al Viro680baac2015-05-02 13:32:22 -04002514 struct page *page = cookie;
2515 kunmap(page);
2516 mark_page_accessed(page);
2517 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518}
2519
Eric Parisb09e0fa2011-05-24 17:12:39 -07002520#ifdef CONFIG_TMPFS_XATTR
2521/*
2522 * Superblocks without xattr inode operations may get some security.* xattr
2523 * support from the LSM "for free". As soon as we have any other xattrs
2524 * like ACLs, we also need to implement the security.* handlers at
2525 * filesystem level, though.
2526 */
2527
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002528/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002529 * Callback for security_inode_init_security() for acquiring xattrs.
2530 */
2531static int shmem_initxattrs(struct inode *inode,
2532 const struct xattr *xattr_array,
2533 void *fs_info)
2534{
2535 struct shmem_inode_info *info = SHMEM_I(inode);
2536 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002537 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002538 size_t len;
2539
2540 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002541 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002542 if (!new_xattr)
2543 return -ENOMEM;
2544
2545 len = strlen(xattr->name) + 1;
2546 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2547 GFP_KERNEL);
2548 if (!new_xattr->name) {
2549 kfree(new_xattr);
2550 return -ENOMEM;
2551 }
2552
2553 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2554 XATTR_SECURITY_PREFIX_LEN);
2555 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2556 xattr->name, len);
2557
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002558 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002559 }
2560
2561 return 0;
2562}
2563
Eric Parisb09e0fa2011-05-24 17:12:39 -07002564static const struct xattr_handler *shmem_xattr_handlers[] = {
2565#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002566 &posix_acl_access_xattr_handler,
2567 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002568#endif
2569 NULL
2570};
2571
2572static int shmem_xattr_validate(const char *name)
2573{
2574 struct { const char *prefix; size_t len; } arr[] = {
2575 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2576 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2577 };
2578 int i;
2579
2580 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2581 size_t preflen = arr[i].len;
2582 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2583 if (!name[preflen])
2584 return -EINVAL;
2585 return 0;
2586 }
2587 }
2588 return -EOPNOTSUPP;
2589}
2590
2591static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2592 void *buffer, size_t size)
2593{
David Howells75c3cfa2015-03-17 22:26:12 +00002594 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Eric Parisb09e0fa2011-05-24 17:12:39 -07002595 int err;
2596
2597 /*
2598 * If this is a request for a synthetic attribute in the system.*
2599 * namespace use the generic infrastructure to resolve a handler
2600 * for it via sb->s_xattr.
2601 */
2602 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2603 return generic_getxattr(dentry, name, buffer, size);
2604
2605 err = shmem_xattr_validate(name);
2606 if (err)
2607 return err;
2608
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002609 return simple_xattr_get(&info->xattrs, name, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002610}
2611
2612static int shmem_setxattr(struct dentry *dentry, const char *name,
2613 const void *value, size_t size, int flags)
2614{
David Howells75c3cfa2015-03-17 22:26:12 +00002615 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
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_setxattr(dentry, name, value, size, flags);
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_set(&info->xattrs, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002631}
2632
2633static int shmem_removexattr(struct dentry *dentry, const char *name)
2634{
David Howells75c3cfa2015-03-17 22:26:12 +00002635 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Eric Parisb09e0fa2011-05-24 17:12:39 -07002636 int err;
2637
2638 /*
2639 * If this is a request for a synthetic attribute in the system.*
2640 * namespace use the generic infrastructure to resolve a handler
2641 * for it via sb->s_xattr.
2642 */
2643 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2644 return generic_removexattr(dentry, name);
2645
2646 err = shmem_xattr_validate(name);
2647 if (err)
2648 return err;
2649
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002650 return simple_xattr_remove(&info->xattrs, name);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002651}
2652
2653static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2654{
David Howells75c3cfa2015-03-17 22:26:12 +00002655 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002656 return simple_xattr_list(&info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002657}
2658#endif /* CONFIG_TMPFS_XATTR */
2659
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002660static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 .readlink = generic_readlink,
Al Viro60380f12015-05-02 10:24:43 -04002662 .follow_link = simple_follow_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002663#ifdef CONFIG_TMPFS_XATTR
2664 .setxattr = shmem_setxattr,
2665 .getxattr = shmem_getxattr,
2666 .listxattr = shmem_listxattr,
2667 .removexattr = shmem_removexattr,
2668#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669};
2670
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002671static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 .readlink = generic_readlink,
2673 .follow_link = shmem_follow_link,
2674 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002675#ifdef CONFIG_TMPFS_XATTR
2676 .setxattr = shmem_setxattr,
2677 .getxattr = shmem_getxattr,
2678 .listxattr = shmem_listxattr,
2679 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002680#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002681};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002682
David M. Grimes91828a42006-10-17 00:09:45 -07002683static struct dentry *shmem_get_parent(struct dentry *child)
2684{
2685 return ERR_PTR(-ESTALE);
2686}
2687
2688static int shmem_match(struct inode *ino, void *vfh)
2689{
2690 __u32 *fh = vfh;
2691 __u64 inum = fh[2];
2692 inum = (inum << 32) | fh[1];
2693 return ino->i_ino == inum && fh[0] == ino->i_generation;
2694}
2695
Christoph Hellwig480b1162007-10-21 16:42:13 -07002696static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2697 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002698{
David M. Grimes91828a42006-10-17 00:09:45 -07002699 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002700 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002701 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002702
Christoph Hellwig480b1162007-10-21 16:42:13 -07002703 if (fh_len < 3)
2704 return NULL;
2705
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002706 inum = fid->raw[2];
2707 inum = (inum << 32) | fid->raw[1];
2708
Christoph Hellwig480b1162007-10-21 16:42:13 -07002709 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2710 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002711 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002712 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002713 iput(inode);
2714 }
2715
Christoph Hellwig480b1162007-10-21 16:42:13 -07002716 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002717}
2718
Al Virob0b03822012-04-02 14:34:06 -04002719static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2720 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002721{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302722 if (*len < 3) {
2723 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002724 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302725 }
David M. Grimes91828a42006-10-17 00:09:45 -07002726
Al Viro1d3382cb2010-10-23 15:19:20 -04002727 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002728 /* Unfortunately insert_inode_hash is not idempotent,
2729 * so as we hash inodes here rather than at creation
2730 * time, we need a lock to ensure we only try
2731 * to do it once
2732 */
2733 static DEFINE_SPINLOCK(lock);
2734 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002735 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002736 __insert_inode_hash(inode,
2737 inode->i_ino + inode->i_generation);
2738 spin_unlock(&lock);
2739 }
2740
2741 fh[0] = inode->i_generation;
2742 fh[1] = inode->i_ino;
2743 fh[2] = ((__u64)inode->i_ino) >> 32;
2744
2745 *len = 3;
2746 return 1;
2747}
2748
Christoph Hellwig39655162007-10-21 16:42:17 -07002749static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002750 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002751 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002752 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002753};
2754
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002755static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2756 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757{
2758 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002759 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002760 uid_t uid;
2761 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002763 while (options != NULL) {
2764 this_char = options;
2765 for (;;) {
2766 /*
2767 * NUL-terminate this option: unfortunately,
2768 * mount options form a comma-separated list,
2769 * but mpol's nodelist may also contain commas.
2770 */
2771 options = strchr(options, ',');
2772 if (options == NULL)
2773 break;
2774 options++;
2775 if (!isdigit(*options)) {
2776 options[-1] = '\0';
2777 break;
2778 }
2779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 if (!*this_char)
2781 continue;
2782 if ((value = strchr(this_char,'=')) != NULL) {
2783 *value++ = 0;
2784 } else {
2785 printk(KERN_ERR
2786 "tmpfs: No value for mount option '%s'\n",
2787 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002788 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 }
2790
2791 if (!strcmp(this_char,"size")) {
2792 unsigned long long size;
2793 size = memparse(value,&rest);
2794 if (*rest == '%') {
2795 size <<= PAGE_SHIFT;
2796 size *= totalram_pages;
2797 do_div(size, 100);
2798 rest++;
2799 }
2800 if (*rest)
2801 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002802 sbinfo->max_blocks =
2803 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002805 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 if (*rest)
2807 goto bad_val;
2808 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002809 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 if (*rest)
2811 goto bad_val;
2812 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002813 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002815 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (*rest)
2817 goto bad_val;
2818 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002819 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002821 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 if (*rest)
2823 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002824 sbinfo->uid = make_kuid(current_user_ns(), uid);
2825 if (!uid_valid(sbinfo->uid))
2826 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002828 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002830 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 if (*rest)
2832 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002833 sbinfo->gid = make_kgid(current_user_ns(), gid);
2834 if (!gid_valid(sbinfo->gid))
2835 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002836 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002837 mpol_put(mpol);
2838 mpol = NULL;
2839 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002840 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 } else {
2842 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2843 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002844 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 }
2846 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002847 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 return 0;
2849
2850bad_val:
2851 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2852 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002853error:
2854 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 return 1;
2856
2857}
2858
2859static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2860{
2861 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002862 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002863 unsigned long inodes;
2864 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865
Greg Thelen5f001102013-02-22 16:36:01 -08002866 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002867 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002868 return error;
2869
2870 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002871 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002872 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002873 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002874 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002875 goto out;
2876 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002877 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002878 * but we must separately disallow unlimited->limited, because
2879 * in that case we have no record of how much is already in use.
2880 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002881 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002882 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002883 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002884 goto out;
2885
2886 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002887 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002888 sbinfo->max_inodes = config.max_inodes;
2889 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002890
Greg Thelen5f001102013-02-22 16:36:01 -08002891 /*
2892 * Preserve previous mempolicy unless mpol remount option was specified.
2893 */
2894 if (config.mpol) {
2895 mpol_put(sbinfo->mpol);
2896 sbinfo->mpol = config.mpol; /* transfers initial ref */
2897 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002898out:
2899 spin_unlock(&sbinfo->stat_lock);
2900 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002902
Al Viro34c80b12011-12-08 21:32:45 -05002903static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002904{
Al Viro34c80b12011-12-08 21:32:45 -05002905 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002906
2907 if (sbinfo->max_blocks != shmem_default_max_blocks())
2908 seq_printf(seq, ",size=%luk",
2909 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2910 if (sbinfo->max_inodes != shmem_default_max_inodes())
2911 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2912 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002913 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002914 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2915 seq_printf(seq, ",uid=%u",
2916 from_kuid_munged(&init_user_ns, sbinfo->uid));
2917 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2918 seq_printf(seq, ",gid=%u",
2919 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002920 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002921 return 0;
2922}
David Herrmann9183df22014-08-08 14:25:29 -07002923
2924#define MFD_NAME_PREFIX "memfd:"
2925#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2926#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2927
2928#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2929
2930SYSCALL_DEFINE2(memfd_create,
2931 const char __user *, uname,
2932 unsigned int, flags)
2933{
2934 struct shmem_inode_info *info;
2935 struct file *file;
2936 int fd, error;
2937 char *name;
2938 long len;
2939
2940 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2941 return -EINVAL;
2942
2943 /* length includes terminating zero */
2944 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2945 if (len <= 0)
2946 return -EFAULT;
2947 if (len > MFD_NAME_MAX_LEN + 1)
2948 return -EINVAL;
2949
2950 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2951 if (!name)
2952 return -ENOMEM;
2953
2954 strcpy(name, MFD_NAME_PREFIX);
2955 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2956 error = -EFAULT;
2957 goto err_name;
2958 }
2959
2960 /* terminating-zero may have changed after strnlen_user() returned */
2961 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
2962 error = -EFAULT;
2963 goto err_name;
2964 }
2965
2966 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
2967 if (fd < 0) {
2968 error = fd;
2969 goto err_name;
2970 }
2971
2972 file = shmem_file_setup(name, 0, VM_NORESERVE);
2973 if (IS_ERR(file)) {
2974 error = PTR_ERR(file);
2975 goto err_fd;
2976 }
2977 info = SHMEM_I(file_inode(file));
2978 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
2979 file->f_flags |= O_RDWR | O_LARGEFILE;
2980 if (flags & MFD_ALLOW_SEALING)
2981 info->seals &= ~F_SEAL_SEAL;
2982
2983 fd_install(fd, file);
2984 kfree(name);
2985 return fd;
2986
2987err_fd:
2988 put_unused_fd(fd);
2989err_name:
2990 kfree(name);
2991 return error;
2992}
2993
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002994#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
2996static void shmem_put_super(struct super_block *sb)
2997{
Hugh Dickins602586a2010-08-17 15:23:56 -07002998 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2999
3000 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003001 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003002 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 sb->s_fs_info = NULL;
3004}
3005
Kay Sievers2b2af542009-04-30 15:23:42 +02003006int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007{
3008 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003009 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003010 int err = -ENOMEM;
3011
3012 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003013 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003014 L1_CACHE_BYTES), GFP_KERNEL);
3015 if (!sbinfo)
3016 return -ENOMEM;
3017
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003018 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003019 sbinfo->uid = current_fsuid();
3020 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003021 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003023#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 /*
3025 * Per default we only allow half of the physical ram per
3026 * tmpfs instance, limiting inodes to one per page of lowmem;
3027 * but the internal instance is left unlimited.
3028 */
Al Viroca4e0512013-08-31 12:57:10 -04003029 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003030 sbinfo->max_blocks = shmem_default_max_blocks();
3031 sbinfo->max_inodes = shmem_default_max_inodes();
3032 if (shmem_parse_options(data, sbinfo, false)) {
3033 err = -EINVAL;
3034 goto failed;
3035 }
Al Viroca4e0512013-08-31 12:57:10 -04003036 } else {
3037 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 }
David M. Grimes91828a42006-10-17 00:09:45 -07003039 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003040 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041#else
3042 sb->s_flags |= MS_NOUSER;
3043#endif
3044
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003045 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003046 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003047 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003048 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003049
Hugh Dickins285b2c42011-08-03 16:21:20 -07003050 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 sb->s_blocksize = PAGE_CACHE_SIZE;
3052 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
3053 sb->s_magic = TMPFS_MAGIC;
3054 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003055 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003056#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003057 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003058#endif
3059#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003060 sb->s_flags |= MS_POSIXACL;
3061#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003062
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003063 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 if (!inode)
3065 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003066 inode->i_uid = sbinfo->uid;
3067 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003068 sb->s_root = d_make_root(inode);
3069 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003070 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 return 0;
3072
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073failed:
3074 shmem_put_super(sb);
3075 return err;
3076}
3077
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003078static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079
3080static struct inode *shmem_alloc_inode(struct super_block *sb)
3081{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003082 struct shmem_inode_info *info;
3083 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3084 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003086 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087}
3088
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003089static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003090{
3091 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003092 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3093}
3094
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095static void shmem_destroy_inode(struct inode *inode)
3096{
Al Viro09208d12011-07-26 03:15:03 -04003097 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003099 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100}
3101
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003102static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003104 struct shmem_inode_info *info = foo;
3105 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106}
3107
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003108static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109{
3110 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3111 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003112 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 return 0;
3114}
3115
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003116static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003118 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119}
3120
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003121static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003123 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003125 .write_begin = shmem_write_begin,
3126 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003128#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003129 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003130#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003131 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132};
3133
Helge Deller15ad7cd2006-12-06 20:40:36 -08003134static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 .mmap = shmem_mmap,
3136#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003137 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003138 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003139 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003140 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003141 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003142 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003143 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144#endif
3145};
3146
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003147static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003148 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003149 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003150#ifdef CONFIG_TMPFS_XATTR
3151 .setxattr = shmem_setxattr,
3152 .getxattr = shmem_getxattr,
3153 .listxattr = shmem_listxattr,
3154 .removexattr = shmem_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003155 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003156#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157};
3158
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003159static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160#ifdef CONFIG_TMPFS
3161 .create = shmem_create,
3162 .lookup = simple_lookup,
3163 .link = shmem_link,
3164 .unlink = shmem_unlink,
3165 .symlink = shmem_symlink,
3166 .mkdir = shmem_mkdir,
3167 .rmdir = shmem_rmdir,
3168 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003169 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003170 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003172#ifdef CONFIG_TMPFS_XATTR
3173 .setxattr = shmem_setxattr,
3174 .getxattr = shmem_getxattr,
3175 .listxattr = shmem_listxattr,
3176 .removexattr = shmem_removexattr,
3177#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003178#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003179 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003180 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003181#endif
3182};
3183
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003184static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003185#ifdef CONFIG_TMPFS_XATTR
3186 .setxattr = shmem_setxattr,
3187 .getxattr = shmem_getxattr,
3188 .listxattr = shmem_listxattr,
3189 .removexattr = shmem_removexattr,
3190#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003191#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003192 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003193 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003194#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195};
3196
Hugh Dickins759b9772007-03-05 00:30:28 -08003197static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 .alloc_inode = shmem_alloc_inode,
3199 .destroy_inode = shmem_destroy_inode,
3200#ifdef CONFIG_TMPFS
3201 .statfs = shmem_statfs,
3202 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003203 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204#endif
Al Viro1f895f72010-06-05 19:10:41 -04003205 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 .drop_inode = generic_delete_inode,
3207 .put_super = shmem_put_super,
3208};
3209
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003210static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003211 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003212 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213#ifdef CONFIG_NUMA
3214 .set_policy = shmem_set_policy,
3215 .get_policy = shmem_get_policy,
3216#endif
3217};
3218
Al Viro3c26ff62010-07-25 11:46:36 +04003219static struct dentry *shmem_mount(struct file_system_type *fs_type,
3220 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221{
Al Viro3c26ff62010-07-25 11:46:36 +04003222 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223}
3224
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003225static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 .owner = THIS_MODULE,
3227 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003228 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003230 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003233int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234{
3235 int error;
3236
Rob Landley16203a72013-09-11 14:26:12 -07003237 /* If rootfs called this, don't re-init */
3238 if (shmem_inode_cachep)
3239 return 0;
3240
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003241 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 if (error)
3243 goto out3;
3244
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003245 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 if (error) {
3247 printk(KERN_ERR "Could not register tmpfs\n");
3248 goto out2;
3249 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003250
Al Viroca4e0512013-08-31 12:57:10 -04003251 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 if (IS_ERR(shm_mnt)) {
3253 error = PTR_ERR(shm_mnt);
3254 printk(KERN_ERR "Could not kern_mount tmpfs\n");
3255 goto out1;
3256 }
3257 return 0;
3258
3259out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003260 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003262 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263out3:
3264 shm_mnt = ERR_PTR(error);
3265 return error;
3266}
Matt Mackall853ac432009-01-06 14:40:20 -08003267
3268#else /* !CONFIG_SHMEM */
3269
3270/*
3271 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3272 *
3273 * This is intended for small system where the benefits of the full
3274 * shmem code (swap-backed and resource-limited) are outweighed by
3275 * their complexity. On systems without swap this code should be
3276 * effectively equivalent, but much lighter weight.
3277 */
3278
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003279static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003280 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003281 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003282 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003283 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003284};
3285
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003286int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003287{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003288 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003289
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003290 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003291 BUG_ON(IS_ERR(shm_mnt));
3292
3293 return 0;
3294}
3295
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003296int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003297{
3298 return 0;
3299}
3300
Hugh Dickins3f96b792009-09-21 17:03:37 -07003301int shmem_lock(struct file *file, int lock, struct user_struct *user)
3302{
3303 return 0;
3304}
3305
Hugh Dickins24513262012-01-20 14:34:21 -08003306void shmem_unlock_mapping(struct address_space *mapping)
3307{
3308}
3309
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003310void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003311{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003312 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003313}
3314EXPORT_SYMBOL_GPL(shmem_truncate_range);
3315
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003316#define shmem_vm_ops generic_file_vm_ops
3317#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003318#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003319#define shmem_acct_size(flags, size) 0
3320#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003321
3322#endif /* CONFIG_SHMEM */
3323
3324/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325
Al Viro34515382013-02-14 22:38:02 -05003326static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003327 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003328};
3329
Eric Parisc7277092013-12-02 11:24:19 +00003330static struct file *__shmem_file_setup(const char *name, loff_t size,
3331 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332{
Al Viro6b4d0b22013-02-14 21:37:26 -05003333 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003335 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003336 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 struct qstr this;
3338
3339 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003340 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341
Hugh Dickins285b2c42011-08-03 16:21:20 -07003342 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 return ERR_PTR(-EINVAL);
3344
3345 if (shmem_acct_size(flags, size))
3346 return ERR_PTR(-ENOMEM);
3347
Al Viro6b4d0b22013-02-14 21:37:26 -05003348 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 this.name = name;
3350 this.len = strlen(name);
3351 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003352 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003353 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003354 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003355 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003357 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358
Al Viro6b4d0b22013-02-14 21:37:26 -05003359 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05003360 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003362 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363
Eric Parisc7277092013-12-02 11:24:19 +00003364 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04003365 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003367 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05003368 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3369 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003370 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003371
Al Viro6b4d0b22013-02-14 21:37:26 -05003372 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04003373 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05003374 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003375 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003376
Al Viro6b4d0b22013-02-14 21:37:26 -05003377 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379put_memory:
3380 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003381put_path:
3382 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05003383 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384}
Eric Parisc7277092013-12-02 11:24:19 +00003385
3386/**
3387 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3388 * kernel internal. There will be NO LSM permission checks against the
3389 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07003390 * higher layer. The users are the big_key and shm implementations. LSM
3391 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00003392 * @name: name for dentry (to be seen in /proc/<pid>/maps
3393 * @size: size to be set for the file
3394 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3395 */
3396struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3397{
3398 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3399}
3400
3401/**
3402 * shmem_file_setup - get an unlinked file living in tmpfs
3403 * @name: name for dentry (to be seen in /proc/<pid>/maps
3404 * @size: size to be set for the file
3405 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3406 */
3407struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3408{
3409 return __shmem_file_setup(name, size, flags, 0);
3410}
Keith Packard395e0dd2008-06-20 00:08:06 -07003411EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Randy Dunlap46711812008-03-19 17:00:41 -07003413/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3416 */
3417int shmem_zero_setup(struct vm_area_struct *vma)
3418{
3419 struct file *file;
3420 loff_t size = vma->vm_end - vma->vm_start;
3421
Hugh Dickins66fc1302015-06-14 09:48:09 -07003422 /*
3423 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3424 * between XFS directory reading and selinux: since this file is only
3425 * accessible to the user through its mapping, use S_PRIVATE flag to
3426 * bypass file security, in the same way as shmem_kernel_file_setup().
3427 */
3428 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 if (IS_ERR(file))
3430 return PTR_ERR(file);
3431
3432 if (vma->vm_file)
3433 fput(vma->vm_file);
3434 vma->vm_file = file;
3435 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 return 0;
3437}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003438
3439/**
3440 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3441 * @mapping: the page's address_space
3442 * @index: the page index
3443 * @gfp: the page allocator flags to use if allocating
3444 *
3445 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3446 * with any new page allocations done using the specified allocation flags.
3447 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3448 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3449 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3450 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003451 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3452 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003453 */
3454struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3455 pgoff_t index, gfp_t gfp)
3456{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003457#ifdef CONFIG_SHMEM
3458 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003459 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003460 int error;
3461
3462 BUG_ON(mapping->a_ops != &shmem_aops);
3463 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3464 if (error)
3465 page = ERR_PTR(error);
3466 else
3467 unlock_page(page);
3468 return page;
3469#else
3470 /*
3471 * The tiny !SHMEM case uses ramfs without swap
3472 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003473 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003474#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003475}
3476EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);