blob: 61a39aa49c5df07a05af2e57b41f045f8a658b5e [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>
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -070035#include <linux/khugepaged.h>
Matt Mackall853ac432009-01-06 14:40:20 -080036
37static struct vfsmount *shm_mnt;
38
39#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
44 */
45
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070046#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070047#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010048#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080049#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/string.h>
52#include <linux/slab.h>
53#include <linux/backing-dev.h>
54#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070057#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070058#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070059#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070060#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <linux/security.h>
62#include <linux/swapops.h>
63#include <linux/mempolicy.h>
64#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000065#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070066#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070067#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080068#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040069#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070070#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070071#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070072#include <uapi/linux/memfd.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070073
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <asm/pgtable.h>
76
Mel Gormandd56b042015-11-06 16:28:43 -080077#include "internal.h"
78
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030079#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Linus Torvalds1da177e2005-04-16 15:20:36 -070082/* Pretend that each entry is of this size in directory's i_size */
83#define BOGO_DIRENT_SIZE 20
84
Hugh Dickins69f07ec2011-08-03 16:21:26 -070085/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86#define SHORT_SYMLINK_LEN 128
87
Hugh Dickins1aac1402012-05-29 15:06:42 -070088/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070089 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070092 */
93struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -070094 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -070095 pgoff_t start; /* start of range currently being fallocated */
96 pgoff_t next; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
99};
100
Andrew Mortonb76db732008-02-08 04:21:49 -0800101#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800102static unsigned long shmem_default_max_blocks(void)
103{
104 return totalram_pages / 2;
105}
106
107static unsigned long shmem_default_max_inodes(void)
108{
109 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
110}
Andrew Mortonb76db732008-02-08 04:21:49 -0800111#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800112
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700113static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700116static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700117 struct page **pagep, enum sgp_type sgp,
118 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700119
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700120int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700121 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700122{
123 return shmem_getpage_gfp(inode, index, pagep, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700124 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700125}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
128{
129 return sb->s_fs_info;
130}
131
132/*
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
137 */
138static inline int shmem_acct_size(unsigned long flags, loff_t size)
139{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000140 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000141 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
144static inline void shmem_unacct_size(unsigned long flags, loff_t size)
145{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000146 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 vm_unacct_memory(VM_ACCT(size));
148}
149
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700150static inline int shmem_reacct_size(unsigned long flags,
151 loff_t oldsize, loff_t newsize)
152{
153 if (!(flags & VM_NORESERVE)) {
154 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
155 return security_vm_enough_memory_mm(current->mm,
156 VM_ACCT(newsize) - VM_ACCT(oldsize));
157 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
158 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
159 }
160 return 0;
161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/*
164 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700165 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
168 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700169static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700171 if (!(flags & VM_NORESERVE))
172 return 0;
173
174 return security_vm_enough_memory_mm(current->mm,
175 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
178static inline void shmem_unacct_blocks(unsigned long flags, long pages)
179{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000180 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300181 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
Hugh Dickins759b9772007-03-05 00:30:28 -0800184static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700185static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800186static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800187static const struct inode_operations shmem_inode_operations;
188static const struct inode_operations shmem_dir_inode_operations;
189static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400190static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700191static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800194static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800196static int shmem_reserve_inode(struct super_block *sb)
197{
198 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
199 if (sbinfo->max_inodes) {
200 spin_lock(&sbinfo->stat_lock);
201 if (!sbinfo->free_inodes) {
202 spin_unlock(&sbinfo->stat_lock);
203 return -ENOSPC;
204 }
205 sbinfo->free_inodes--;
206 spin_unlock(&sbinfo->stat_lock);
207 }
208 return 0;
209}
210
211static void shmem_free_inode(struct super_block *sb)
212{
213 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
214 if (sbinfo->max_inodes) {
215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_inodes++;
217 spin_unlock(&sbinfo->stat_lock);
218 }
219}
220
Randy Dunlap46711812008-03-19 17:00:41 -0700221/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700222 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 * @inode: inode to recalc
224 *
225 * We have to calculate the free blocks since the mm can drop
226 * undirtied hole pages behind our back.
227 *
228 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
229 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
230 *
231 * It has to be called with the spinlock held.
232 */
233static void shmem_recalc_inode(struct inode *inode)
234{
235 struct shmem_inode_info *info = SHMEM_I(inode);
236 long freed;
237
238 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
239 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700240 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
241 if (sbinfo->max_blocks)
242 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700244 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247}
248
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700249bool shmem_charge(struct inode *inode, long pages)
250{
251 struct shmem_inode_info *info = SHMEM_I(inode);
252 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700253 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700254
255 if (shmem_acct_block(info->flags, pages))
256 return false;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700257 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700258 info->alloced += pages;
259 inode->i_blocks += pages * BLOCKS_PER_PAGE;
260 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700261 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700262 inode->i_mapping->nrpages += pages;
263
264 if (!sbinfo->max_blocks)
265 return true;
266 if (percpu_counter_compare(&sbinfo->used_blocks,
267 sbinfo->max_blocks - pages) > 0) {
268 inode->i_mapping->nrpages -= pages;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700269 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700270 info->alloced -= pages;
271 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700272 spin_unlock_irqrestore(&info->lock, flags);
Hugh Dickins71664662016-09-23 20:24:23 -0700273 shmem_unacct_blocks(info->flags, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700274 return false;
275 }
276 percpu_counter_add(&sbinfo->used_blocks, pages);
277 return true;
278}
279
280void shmem_uncharge(struct inode *inode, long pages)
281{
282 struct shmem_inode_info *info = SHMEM_I(inode);
283 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700284 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700285
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700286 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700287 info->alloced -= pages;
288 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
289 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700290 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700291
292 if (sbinfo->max_blocks)
293 percpu_counter_sub(&sbinfo->used_blocks, pages);
Hugh Dickins71664662016-09-23 20:24:23 -0700294 shmem_unacct_blocks(info->flags, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700295}
296
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700297/*
298 * Replace item expected in radix tree by a new item, while holding tree lock.
299 */
300static int shmem_radix_tree_replace(struct address_space *mapping,
301 pgoff_t index, void *expected, void *replacement)
302{
303 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700304 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700305
306 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700307 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700308 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700309 if (!pslot)
310 return -ENOENT;
311 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700312 if (item != expected)
313 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700314 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700315 return 0;
316}
317
318/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700319 * Sometimes, before we decide whether to proceed or to fail, we must check
320 * that an entry was not already brought back from swap by a racing thread.
321 *
322 * Checking page is not enough: by the time a SwapCache page is locked, it
323 * might be reused, and again be SwapCache, using the same swap as before.
324 */
325static bool shmem_confirm_swap(struct address_space *mapping,
326 pgoff_t index, swp_entry_t swap)
327{
328 void *item;
329
330 rcu_read_lock();
331 item = radix_tree_lookup(&mapping->page_tree, index);
332 rcu_read_unlock();
333 return item == swp_to_radix_entry(swap);
334}
335
336/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700337 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
338 *
339 * SHMEM_HUGE_NEVER:
340 * disables huge pages for the mount;
341 * SHMEM_HUGE_ALWAYS:
342 * enables huge pages for the mount;
343 * SHMEM_HUGE_WITHIN_SIZE:
344 * only allocate huge pages if the page will be fully within i_size,
345 * also respect fadvise()/madvise() hints;
346 * SHMEM_HUGE_ADVISE:
347 * only allocate huge pages if requested with fadvise()/madvise();
348 */
349
350#define SHMEM_HUGE_NEVER 0
351#define SHMEM_HUGE_ALWAYS 1
352#define SHMEM_HUGE_WITHIN_SIZE 2
353#define SHMEM_HUGE_ADVISE 3
354
355/*
356 * Special values.
357 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
358 *
359 * SHMEM_HUGE_DENY:
360 * disables huge on shm_mnt and all mounts, for emergency use;
361 * SHMEM_HUGE_FORCE:
362 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
363 *
364 */
365#define SHMEM_HUGE_DENY (-1)
366#define SHMEM_HUGE_FORCE (-2)
367
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700368#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700369/* ifdef here to avoid bloating shmem.o when not necessary */
370
371int shmem_huge __read_mostly;
372
Jérémy Lefaure8bec83b2016-12-12 16:43:23 -0800373#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700374static int shmem_parse_huge(const char *str)
375{
376 if (!strcmp(str, "never"))
377 return SHMEM_HUGE_NEVER;
378 if (!strcmp(str, "always"))
379 return SHMEM_HUGE_ALWAYS;
380 if (!strcmp(str, "within_size"))
381 return SHMEM_HUGE_WITHIN_SIZE;
382 if (!strcmp(str, "advise"))
383 return SHMEM_HUGE_ADVISE;
384 if (!strcmp(str, "deny"))
385 return SHMEM_HUGE_DENY;
386 if (!strcmp(str, "force"))
387 return SHMEM_HUGE_FORCE;
388 return -EINVAL;
389}
390
391static const char *shmem_format_huge(int huge)
392{
393 switch (huge) {
394 case SHMEM_HUGE_NEVER:
395 return "never";
396 case SHMEM_HUGE_ALWAYS:
397 return "always";
398 case SHMEM_HUGE_WITHIN_SIZE:
399 return "within_size";
400 case SHMEM_HUGE_ADVISE:
401 return "advise";
402 case SHMEM_HUGE_DENY:
403 return "deny";
404 case SHMEM_HUGE_FORCE:
405 return "force";
406 default:
407 VM_BUG_ON(1);
408 return "bad_val";
409 }
410}
Jérémy Lefaure8bec83b2016-12-12 16:43:23 -0800411#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700412
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700413static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
414 struct shrink_control *sc, unsigned long nr_to_split)
415{
416 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400417 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700418 struct inode *inode;
419 struct shmem_inode_info *info;
420 struct page *page;
421 unsigned long batch = sc ? sc->nr_to_scan : 128;
422 int removed = 0, split = 0;
423
424 if (list_empty(&sbinfo->shrinklist))
425 return SHRINK_STOP;
426
427 spin_lock(&sbinfo->shrinklist_lock);
428 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
429 info = list_entry(pos, struct shmem_inode_info, shrinklist);
430
431 /* pin the inode */
432 inode = igrab(&info->vfs_inode);
433
434 /* inode is about to be evicted */
435 if (!inode) {
436 list_del_init(&info->shrinklist);
437 removed++;
438 goto next;
439 }
440
441 /* Check if there's anything to gain */
442 if (round_up(inode->i_size, PAGE_SIZE) ==
443 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400444 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700445 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700446 goto next;
447 }
448
449 list_move(&info->shrinklist, &list);
450next:
451 if (!--batch)
452 break;
453 }
454 spin_unlock(&sbinfo->shrinklist_lock);
455
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400456 list_for_each_safe(pos, next, &to_remove) {
457 info = list_entry(pos, struct shmem_inode_info, shrinklist);
458 inode = &info->vfs_inode;
459 list_del_init(&info->shrinklist);
460 iput(inode);
461 }
462
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700463 list_for_each_safe(pos, next, &list) {
464 int ret;
465
466 info = list_entry(pos, struct shmem_inode_info, shrinklist);
467 inode = &info->vfs_inode;
468
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700469 if (nr_to_split && split >= nr_to_split)
470 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700471
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700472 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700473 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
474 if (!page)
475 goto drop;
476
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700477 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700478 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700479 put_page(page);
480 goto drop;
481 }
482
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700483 /*
484 * Leave the inode on the list if we failed to lock
485 * the page at this time.
486 *
487 * Waiting for the lock may lead to deadlock in the
488 * reclaim path.
489 */
490 if (!trylock_page(page)) {
491 put_page(page);
492 goto leave;
493 }
494
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700495 ret = split_huge_page(page);
496 unlock_page(page);
497 put_page(page);
498
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700499 /* If split failed leave the inode on the list */
500 if (ret)
501 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700502
503 split++;
504drop:
505 list_del_init(&info->shrinklist);
506 removed++;
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700507leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700508 iput(inode);
509 }
510
511 spin_lock(&sbinfo->shrinklist_lock);
512 list_splice_tail(&list, &sbinfo->shrinklist);
513 sbinfo->shrinklist_len -= removed;
514 spin_unlock(&sbinfo->shrinklist_lock);
515
516 return split;
517}
518
519static long shmem_unused_huge_scan(struct super_block *sb,
520 struct shrink_control *sc)
521{
522 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
523
524 if (!READ_ONCE(sbinfo->shrinklist_len))
525 return SHRINK_STOP;
526
527 return shmem_unused_huge_shrink(sbinfo, sc, 0);
528}
529
530static long shmem_unused_huge_count(struct super_block *sb,
531 struct shrink_control *sc)
532{
533 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
534 return READ_ONCE(sbinfo->shrinklist_len);
535}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700536#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700537
538#define shmem_huge SHMEM_HUGE_DENY
539
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700540static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
541 struct shrink_control *sc, unsigned long nr_to_split)
542{
543 return 0;
544}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700545#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700546
547/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700548 * Like add_to_page_cache_locked, but error if expected item has gone.
549 */
550static int shmem_add_to_page_cache(struct page *page,
551 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700552 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700553{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700554 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700555
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700556 VM_BUG_ON_PAGE(PageTail(page), page);
557 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800558 VM_BUG_ON_PAGE(!PageLocked(page), page);
559 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700560 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700561
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700562 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700563 page->mapping = mapping;
564 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700565
Hugh Dickinsb065b432012-07-11 14:02:48 -0700566 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700567 if (PageTransHuge(page)) {
568 void __rcu **results;
569 pgoff_t idx;
570 int i;
571
572 error = 0;
573 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
574 &results, &idx, index, 1) &&
575 idx < index + HPAGE_PMD_NR) {
576 error = -EEXIST;
577 }
578
579 if (!error) {
580 for (i = 0; i < HPAGE_PMD_NR; i++) {
581 error = radix_tree_insert(&mapping->page_tree,
582 index + i, page + i);
583 VM_BUG_ON(error);
584 }
585 count_vm_event(THP_FILE_ALLOC);
586 }
587 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700588 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700589 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700590 error = shmem_radix_tree_replace(mapping, index, expected,
591 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700592 }
593
Hugh Dickinsb065b432012-07-11 14:02:48 -0700594 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700595 mapping->nrpages += nr;
596 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700597 __inc_node_page_state(page, NR_SHMEM_THPS);
598 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
599 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700600 spin_unlock_irq(&mapping->tree_lock);
601 } else {
602 page->mapping = NULL;
603 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700604 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700605 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700606 return error;
607}
608
609/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700610 * Like delete_from_page_cache, but substitutes swap for page.
611 */
612static void shmem_delete_from_page_cache(struct page *page, void *radswap)
613{
614 struct address_space *mapping = page->mapping;
615 int error;
616
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700617 VM_BUG_ON_PAGE(PageCompound(page), page);
618
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700619 spin_lock_irq(&mapping->tree_lock);
620 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
621 page->mapping = NULL;
622 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700623 __dec_node_page_state(page, NR_FILE_PAGES);
624 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700625 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300626 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700627 BUG_ON(error);
628}
629
630/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700631 * Remove swap entry from radix tree, free the swap and its page cache.
632 */
633static int shmem_free_swap(struct address_space *mapping,
634 pgoff_t index, void *radswap)
635{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700636 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700637
638 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700639 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700640 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700641 if (old != radswap)
642 return -ENOENT;
643 free_swap_and_cache(radix_to_swp_entry(radswap));
644 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700645}
646
647/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800648 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800649 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800650 *
651 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
652 * as long as the inode doesn't go away and racy results are not a problem.
653 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800654unsigned long shmem_partial_swap_usage(struct address_space *mapping,
655 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800656{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800657 struct radix_tree_iter iter;
658 void **slot;
659 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800660 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800661
662 rcu_read_lock();
663
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800664 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
665 if (iter.index >= end)
666 break;
667
668 page = radix_tree_deref_slot(slot);
669
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700670 if (radix_tree_deref_retry(page)) {
671 slot = radix_tree_iter_retry(&iter);
672 continue;
673 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800674
675 if (radix_tree_exceptional_entry(page))
676 swapped++;
677
678 if (need_resched()) {
679 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700680 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800681 }
682 }
683
684 rcu_read_unlock();
685
686 return swapped << PAGE_SHIFT;
687}
688
689/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800690 * Determine (in bytes) how many of the shmem object's pages mapped by the
691 * given vma is swapped out.
692 *
693 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
694 * as long as the inode doesn't go away and racy results are not a problem.
695 */
696unsigned long shmem_swap_usage(struct vm_area_struct *vma)
697{
698 struct inode *inode = file_inode(vma->vm_file);
699 struct shmem_inode_info *info = SHMEM_I(inode);
700 struct address_space *mapping = inode->i_mapping;
701 unsigned long swapped;
702
703 /* Be careful as we don't hold info->lock */
704 swapped = READ_ONCE(info->swapped);
705
706 /*
707 * The easier cases are when the shmem object has nothing in swap, or
708 * the vma maps it whole. Then we can simply use the stats that we
709 * already track.
710 */
711 if (!swapped)
712 return 0;
713
714 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
715 return swapped << PAGE_SHIFT;
716
717 /* Here comes the more involved part */
718 return shmem_partial_swap_usage(mapping,
719 linear_page_index(vma, vma->vm_start),
720 linear_page_index(vma, vma->vm_end));
721}
722
723/*
Hugh Dickins24513262012-01-20 14:34:21 -0800724 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
725 */
726void shmem_unlock_mapping(struct address_space *mapping)
727{
728 struct pagevec pvec;
729 pgoff_t indices[PAGEVEC_SIZE];
730 pgoff_t index = 0;
731
732 pagevec_init(&pvec, 0);
733 /*
734 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
735 */
736 while (!mapping_unevictable(mapping)) {
737 /*
738 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
739 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
740 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700741 pvec.nr = find_get_entries(mapping, index,
742 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800743 if (!pvec.nr)
744 break;
745 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700746 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800747 check_move_unevictable_pages(pvec.pages, pvec.nr);
748 pagevec_release(&pvec);
749 cond_resched();
750 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700751}
752
753/*
754 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700755 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700756 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700757static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
758 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700760 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300762 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
763 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
764 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
765 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700766 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700767 pgoff_t indices[PAGEVEC_SIZE];
768 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700769 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700770 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700772 if (lend == -1)
773 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700774
775 pagevec_init(&pvec, 0);
776 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700777 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700778 pvec.nr = find_get_entries(mapping, index,
779 min(end - index, (pgoff_t)PAGEVEC_SIZE),
780 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700781 if (!pvec.nr)
782 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700783 for (i = 0; i < pagevec_count(&pvec); i++) {
784 struct page *page = pvec.pages[i];
785
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700786 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700787 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700788 break;
789
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700790 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700791 if (unfalloc)
792 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700793 nr_swaps_freed += !shmem_free_swap(mapping,
794 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700795 continue;
796 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700797
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700798 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
799
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700800 if (!trylock_page(page))
801 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700802
803 if (PageTransTail(page)) {
804 /* Middle of THP: zero out the page */
805 clear_highpage(page);
806 unlock_page(page);
807 continue;
808 } else if (PageTransHuge(page)) {
809 if (index == round_down(end, HPAGE_PMD_NR)) {
810 /*
811 * Range ends in the middle of THP:
812 * zero out the page
813 */
814 clear_highpage(page);
815 unlock_page(page);
816 continue;
817 }
818 index += HPAGE_PMD_NR - 1;
819 i += HPAGE_PMD_NR - 1;
820 }
821
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700822 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700823 VM_BUG_ON_PAGE(PageTail(page), page);
824 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800825 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700826 truncate_inode_page(mapping, page);
827 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700828 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700829 unlock_page(page);
830 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700831 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800832 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700833 cond_resched();
834 index++;
835 }
836
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700837 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700838 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700839 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700840 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300841 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700842 if (start > end) {
843 top = partial_end;
844 partial_end = 0;
845 }
846 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700847 set_page_dirty(page);
848 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300849 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700850 }
851 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700852 if (partial_end) {
853 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700854 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700855 if (page) {
856 zero_user_segment(page, 0, partial_end);
857 set_page_dirty(page);
858 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300859 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700860 }
861 }
862 if (start >= end)
863 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700864
865 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700866 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700867 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700868
869 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700870 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700871 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700872 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700873 /* If all gone or hole-punch or unfalloc, we're done */
874 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700875 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700876 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700877 index = start;
878 continue;
879 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700880 for (i = 0; i < pagevec_count(&pvec); i++) {
881 struct page *page = pvec.pages[i];
882
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700883 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700884 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700885 break;
886
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700887 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700888 if (unfalloc)
889 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700890 if (shmem_free_swap(mapping, index, page)) {
891 /* Swap was replaced by page: retry */
892 index--;
893 break;
894 }
895 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700896 continue;
897 }
898
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700899 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700900
901 if (PageTransTail(page)) {
902 /* Middle of THP: zero out the page */
903 clear_highpage(page);
904 unlock_page(page);
905 /*
906 * Partial thp truncate due 'start' in middle
907 * of THP: don't need to look on these pages
908 * again on !pvec.nr restart.
909 */
910 if (index != round_down(end, HPAGE_PMD_NR))
911 start++;
912 continue;
913 } else if (PageTransHuge(page)) {
914 if (index == round_down(end, HPAGE_PMD_NR)) {
915 /*
916 * Range ends in the middle of THP:
917 * zero out the page
918 */
919 clear_highpage(page);
920 unlock_page(page);
921 continue;
922 }
923 index += HPAGE_PMD_NR - 1;
924 i += HPAGE_PMD_NR - 1;
925 }
926
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700927 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700928 VM_BUG_ON_PAGE(PageTail(page), page);
929 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800930 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700931 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700932 } else {
933 /* Page was replaced by swap: retry */
934 unlock_page(page);
935 index--;
936 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700937 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700938 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700939 unlock_page(page);
940 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700941 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800942 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700943 index++;
944 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700945
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700946 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700947 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700949 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700950}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700952void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
953{
954 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700955 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700957EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Yu Zhao44a30222015-09-08 15:03:33 -0700959static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
960 struct kstat *stat)
961{
962 struct inode *inode = dentry->d_inode;
963 struct shmem_inode_info *info = SHMEM_I(inode);
964
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800965 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700966 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800967 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700968 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800969 }
Yu Zhao44a30222015-09-08 15:03:33 -0700970 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700971 return 0;
972}
973
Hugh Dickins94c1e622011-06-27 16:18:03 -0700974static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975{
David Howells75c3cfa2015-03-17 22:26:12 +0000976 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700977 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700978 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 int error;
980
Jan Kara31051c82016-05-26 16:55:18 +0200981 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200982 if (error)
983 return error;
984
Hugh Dickins94c1e622011-06-27 16:18:03 -0700985 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
986 loff_t oldsize = inode->i_size;
987 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000988
David Herrmann40e041a2014-08-08 14:25:27 -0700989 /* protected by i_mutex */
990 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
991 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
992 return -EPERM;
993
Hugh Dickins94c1e622011-06-27 16:18:03 -0700994 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700995 error = shmem_reacct_size(SHMEM_I(inode)->flags,
996 oldsize, newsize);
997 if (error)
998 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700999 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001000 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001001 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001002 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001003 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001004 if (oldsize > holebegin)
1005 unmap_mapping_range(inode->i_mapping,
1006 holebegin, 0, 1);
1007 if (info->alloced)
1008 shmem_truncate_range(inode,
1009 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001010 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001011 if (oldsize > holebegin)
1012 unmap_mapping_range(inode->i_mapping,
1013 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001014
1015 /*
1016 * Part of the huge page can be beyond i_size: subject
1017 * to shrink under memory pressure.
1018 */
1019 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1020 spin_lock(&sbinfo->shrinklist_lock);
Cong Wange2286912017-08-10 15:24:24 -07001021 /*
1022 * _careful to defend against unlocked access to
1023 * ->shrink_list in shmem_unused_huge_shrink()
1024 */
1025 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001026 list_add_tail(&info->shrinklist,
1027 &sbinfo->shrinklist);
1028 sbinfo->shrinklist_len++;
1029 }
1030 spin_unlock(&sbinfo->shrinklist_lock);
1031 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 }
1034
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001035 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001036 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001037 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 return error;
1039}
1040
Al Viro1f895f72010-06-05 19:10:41 -04001041static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001044 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001046 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 shmem_unacct_size(info->flags, inode->i_size);
1048 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001049 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001050 if (!list_empty(&info->shrinklist)) {
1051 spin_lock(&sbinfo->shrinklist_lock);
1052 if (!list_empty(&info->shrinklist)) {
1053 list_del_init(&info->shrinklist);
1054 sbinfo->shrinklist_len--;
1055 }
1056 spin_unlock(&sbinfo->shrinklist_lock);
1057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001059 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001061 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 }
Al Viro3ed47db2016-01-22 18:08:52 -05001063 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001064
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001065 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001066 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001067 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001068 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069}
1070
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001071/*
1072 * If swap found in inode, free it and move page from swapcache to filecache.
1073 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001074static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001075 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001077 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001078 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001079 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001080 gfp_t gfp;
1081 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001083 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001084 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001085 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001086 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001087
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001088 /*
1089 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001090 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001091 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001092 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001093 */
1094 if (shmem_swaplist.next != &info->swaplist)
1095 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001096
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001097 gfp = mapping_gfp_mask(mapping);
1098 if (shmem_should_replace_page(*pagep, gfp)) {
1099 mutex_unlock(&shmem_swaplist_mutex);
1100 error = shmem_replace_page(pagep, gfp, info, index);
1101 mutex_lock(&shmem_swaplist_mutex);
1102 /*
1103 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001104 * allocation, but the inode might have been freed while we
1105 * dropped it: although a racing shmem_evict_inode() cannot
1106 * complete without emptying the radix_tree, our page lock
1107 * on this swapcache page is not enough to prevent that -
1108 * free_swap_and_cache() of our swap entry will only
1109 * trylock_page(), removing swap from radix_tree whatever.
1110 *
1111 * We must not proceed to shmem_add_to_page_cache() if the
1112 * inode has been freed, but of course we cannot rely on
1113 * inode or mapping or info to check that. However, we can
1114 * safely check if our swap entry is still in use (and here
1115 * it can't have got reused for another page): if it's still
1116 * in use, then the inode cannot have been freed yet, and we
1117 * can safely proceed (if it's no longer in use, that tells
1118 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001119 */
1120 if (!page_swapcount(*pagep))
1121 error = -ENOENT;
1122 }
1123
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001124 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001125 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1126 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1127 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001128 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001129 if (!error)
1130 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001131 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001132 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001133 /*
1134 * Truncation and eviction use free_swap_and_cache(), which
1135 * only does trylock page: if we raced, best clean up here.
1136 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001137 delete_from_swap_cache(*pagep);
1138 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001139 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001140 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001141 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001142 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001143 swap_free(swap);
1144 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001146 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147}
1148
1149/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001150 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001152int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001154 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001156 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001157 int error = 0;
1158
1159 /*
1160 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001161 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001162 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001163 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001164 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001165
1166 /*
1167 * Charge page using GFP_KERNEL while we can wait, before taking
1168 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1169 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001170 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001171 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1172 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001173 if (error)
1174 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001175 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001176 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001178 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001179 list_for_each_safe(this, next, &shmem_swaplist) {
1180 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001181 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001182 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001183 else
1184 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001185 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001186 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001187 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001188 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001190 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001191
Johannes Weiner00501b52014-08-08 14:19:20 -07001192 if (error) {
1193 if (error != -ENOMEM)
1194 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001195 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001196 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001197 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001198out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001199 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001200 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001201 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
1204/*
1205 * Move the page from the page cache to the swap cache.
1206 */
1207static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1208{
1209 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001212 swp_entry_t swap;
1213 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001215 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 mapping = page->mapping;
1218 index = page->index;
1219 inode = mapping->host;
1220 info = SHMEM_I(inode);
1221 if (info->flags & VM_LOCKED)
1222 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001223 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 goto redirty;
1225
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001226 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001227 * Our capabilities prevent regular writeback or sync from ever calling
1228 * shmem_writepage; but a stacking filesystem might use ->writepage of
1229 * its underlying filesystem, in which case tmpfs should write out to
1230 * swap only in response to memory pressure, and not for the writeback
1231 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001232 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001233 if (!wbc->for_reclaim) {
1234 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1235 goto redirty;
1236 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001237
1238 /*
1239 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1240 * value into swapfile.c, the only way we can correctly account for a
1241 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001242 *
1243 * That's okay for a page already fallocated earlier, but if we have
1244 * not yet completed the fallocation, then (a) we want to keep track
1245 * of this page in case we have to undo it, and (b) it may not be a
1246 * good idea to continue anyway, once we're pushing into swap. So
1247 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001248 */
1249 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001250 if (inode->i_private) {
1251 struct shmem_falloc *shmem_falloc;
1252 spin_lock(&inode->i_lock);
1253 shmem_falloc = inode->i_private;
1254 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001255 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001256 index >= shmem_falloc->start &&
1257 index < shmem_falloc->next)
1258 shmem_falloc->nr_unswapped++;
1259 else
1260 shmem_falloc = NULL;
1261 spin_unlock(&inode->i_lock);
1262 if (shmem_falloc)
1263 goto redirty;
1264 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001265 clear_highpage(page);
1266 flush_dcache_page(page);
1267 SetPageUptodate(page);
1268 }
1269
Hugh Dickins48f170f2011-07-25 17:12:37 -07001270 swap = get_swap_page();
1271 if (!swap.val)
1272 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001273
Vladimir Davydov37e84352016-01-20 15:02:56 -08001274 if (mem_cgroup_try_charge_swap(page, swap))
1275 goto free_swap;
1276
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001277 /*
1278 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001279 * if it's not already there. Do it now before the page is
1280 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001281 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001282 * we've incremented swapped, because shmem_unuse_inode() will
1283 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001284 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001285 mutex_lock(&shmem_swaplist_mutex);
1286 if (list_empty(&info->swaplist))
1287 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001288
Hugh Dickins48f170f2011-07-25 17:12:37 -07001289 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001290 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001291 shmem_recalc_inode(inode);
1292 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001293 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001294
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001295 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001296 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1297
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001298 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001299 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001300 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 return 0;
1302 }
1303
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001304 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001305free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001306 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307redirty:
1308 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001309 if (wbc->for_reclaim)
1310 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1311 unlock_page(page);
1312 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
Hugh Dickins75edd342016-05-19 17:12:44 -07001315#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001316static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001317{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001318 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001319
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001320 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001321 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001322
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001323 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001324
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001325 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001326}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001327
1328static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1329{
1330 struct mempolicy *mpol = NULL;
1331 if (sbinfo->mpol) {
1332 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1333 mpol = sbinfo->mpol;
1334 mpol_get(mpol);
1335 spin_unlock(&sbinfo->stat_lock);
1336 }
1337 return mpol;
1338}
Hugh Dickins75edd342016-05-19 17:12:44 -07001339#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1340static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1341{
1342}
1343static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1344{
1345 return NULL;
1346}
1347#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1348#ifndef CONFIG_NUMA
1349#define vm_policy vm_private_data
1350#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001351
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001352static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1353 struct shmem_inode_info *info, pgoff_t index)
1354{
1355 /* Create a pseudo vma that just contains the policy */
1356 vma->vm_start = 0;
1357 /* Bias interleave by inode number to distribute better across nodes */
1358 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1359 vma->vm_ops = NULL;
1360 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1361}
1362
1363static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1364{
1365 /* Drop reference taken by mpol_shared_policy_lookup() */
1366 mpol_cond_put(vma->vm_policy);
1367}
1368
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001369static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1370 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001373 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001375 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001376 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001377 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001378
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001379 return page;
1380}
Mel Gorman18a2f372012-12-05 14:01:41 -08001381
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001382static struct page *shmem_alloc_hugepage(gfp_t gfp,
1383 struct shmem_inode_info *info, pgoff_t index)
1384{
1385 struct vm_area_struct pvma;
1386 struct inode *inode = &info->vfs_inode;
1387 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001388 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001389 void __rcu **results;
1390 struct page *page;
1391
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001392 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001393 return NULL;
1394
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001395 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001396 rcu_read_lock();
1397 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1398 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1399 rcu_read_unlock();
1400 return NULL;
1401 }
1402 rcu_read_unlock();
1403
1404 shmem_pseudo_vma_init(&pvma, info, hindex);
1405 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1406 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1407 shmem_pseudo_vma_destroy(&pvma);
1408 if (page)
1409 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001410 return page;
1411}
1412
1413static struct page *shmem_alloc_page(gfp_t gfp,
1414 struct shmem_inode_info *info, pgoff_t index)
1415{
1416 struct vm_area_struct pvma;
1417 struct page *page;
1418
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001419 shmem_pseudo_vma_init(&pvma, info, index);
1420 page = alloc_page_vma(gfp, &pvma, 0);
1421 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001422
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001423 return page;
1424}
1425
1426static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1427 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1428 pgoff_t index, bool huge)
1429{
1430 struct page *page;
1431 int nr;
1432 int err = -ENOSPC;
1433
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001434 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001435 huge = false;
1436 nr = huge ? HPAGE_PMD_NR : 1;
1437
1438 if (shmem_acct_block(info->flags, nr))
1439 goto failed;
1440 if (sbinfo->max_blocks) {
1441 if (percpu_counter_compare(&sbinfo->used_blocks,
1442 sbinfo->max_blocks - nr) > 0)
1443 goto unacct;
1444 percpu_counter_add(&sbinfo->used_blocks, nr);
1445 }
1446
1447 if (huge)
1448 page = shmem_alloc_hugepage(gfp, info, index);
1449 else
1450 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001451 if (page) {
1452 __SetPageLocked(page);
1453 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001454 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001455 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001456
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001457 err = -ENOMEM;
1458 if (sbinfo->max_blocks)
1459 percpu_counter_add(&sbinfo->used_blocks, -nr);
1460unacct:
1461 shmem_unacct_blocks(info->flags, nr);
1462failed:
1463 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001467 * When a page is moved from swapcache to shmem filecache (either by the
1468 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1469 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1470 * ignorance of the mapping it belongs to. If that mapping has special
1471 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1472 * we may need to copy to a suitable page before moving to filecache.
1473 *
1474 * In a future release, this may well be extended to respect cpuset and
1475 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1476 * but for now it is a simple matter of zone.
1477 */
1478static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1479{
1480 return page_zonenum(page) > gfp_zone(gfp);
1481}
1482
1483static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1484 struct shmem_inode_info *info, pgoff_t index)
1485{
1486 struct page *oldpage, *newpage;
1487 struct address_space *swap_mapping;
1488 pgoff_t swap_index;
1489 int error;
1490
1491 oldpage = *pagep;
1492 swap_index = page_private(oldpage);
1493 swap_mapping = page_mapping(oldpage);
1494
1495 /*
1496 * We have arrived here because our zones are constrained, so don't
1497 * limit chance of success by further cpuset and node constraints.
1498 */
1499 gfp &= ~GFP_CONSTRAINT_MASK;
1500 newpage = shmem_alloc_page(gfp, info, index);
1501 if (!newpage)
1502 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001503
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001504 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001505 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001506 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001507
Hugh Dickins9956edf2016-11-10 10:46:11 -08001508 __SetPageLocked(newpage);
1509 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001510 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001511 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001512 SetPageSwapCache(newpage);
1513
1514 /*
1515 * Our caller will very soon move newpage out of swapcache, but it's
1516 * a nice clean interface for us to replace oldpage by newpage there.
1517 */
1518 spin_lock_irq(&swap_mapping->tree_lock);
1519 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1520 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001521 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001522 __inc_node_page_state(newpage, NR_FILE_PAGES);
1523 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001524 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001525 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001526
Hugh Dickins0142ef62012-06-07 14:21:09 -07001527 if (unlikely(error)) {
1528 /*
1529 * Is this possible? I think not, now that our callers check
1530 * both PageSwapCache and page_private after getting page lock;
1531 * but be defensive. Reverse old to newpage for clear and free.
1532 */
1533 oldpage = newpage;
1534 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001535 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001536 lru_cache_add_anon(newpage);
1537 *pagep = newpage;
1538 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001539
1540 ClearPageSwapCache(oldpage);
1541 set_page_private(oldpage, 0);
1542
1543 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001544 put_page(oldpage);
1545 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001546 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001547}
1548
1549/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001550 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 *
1552 * If we allocate a new one we do not mark it dirty. That's up to the
1553 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001554 * entry since a page cannot live in both the swap and page cache.
1555 *
1556 * fault_mm and fault_type are only supplied by shmem_fault:
1557 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001559static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001560 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1561 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
1563 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann1dc68392016-12-12 16:42:28 -08001564 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001566 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001567 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001568 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001570 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001571 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001573 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001574 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001576 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001578 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1579 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001581 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001582 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001583 if (radix_tree_exceptional_entry(page)) {
1584 swap = radix_to_swp_entry(page);
1585 page = NULL;
1586 }
1587
Hugh Dickins75edd342016-05-19 17:12:44 -07001588 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001589 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001590 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001591 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001592 }
1593
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001594 if (page && sgp == SGP_WRITE)
1595 mark_page_accessed(page);
1596
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001597 /* fallocated page? */
1598 if (page && !PageUptodate(page)) {
1599 if (sgp != SGP_READ)
1600 goto clear;
1601 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001602 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001603 page = NULL;
1604 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001605 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001606 *pagep = page;
1607 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
1610 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001611 * Fast cache lookup did not find it:
1612 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001614 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001615 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001616
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 if (swap.val) {
1618 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001619 page = lookup_swap_cache(swap);
1620 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001621 /* Or update major stats only when swapin succeeds?? */
1622 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001623 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001624 count_vm_event(PGMAJFAULT);
1625 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1626 }
1627 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001628 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001629 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001630 error = -ENOMEM;
1631 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 }
1634
1635 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001636 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001637 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001638 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001639 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001640 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001641 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001642 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001644 goto failed;
1645 }
1646 wait_on_page_writeback(page);
1647
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001648 if (shmem_should_replace_page(page, gfp)) {
1649 error = shmem_replace_page(&page, gfp, info, index);
1650 if (error)
1651 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 }
1653
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001654 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001655 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001656 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001657 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001658 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001659 /*
1660 * We already confirmed swap under page lock, and make
1661 * no memory allocation here, so usually no possibility
1662 * of error; but free_swap_and_cache() only trylocks a
1663 * page, so it is just possible that the entry has been
1664 * truncated or holepunched since swap was confirmed.
1665 * shmem_undo_range() will have done some of the
1666 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001667 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001668 * Reset swap.val? No, leave it so "failed" goes back to
1669 * "repeat": reading a hole and writing should succeed.
1670 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001671 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001672 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001673 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001674 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001675 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001676 if (error)
1677 goto failed;
1678
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001679 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001680
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001681 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001682 info->swapped--;
1683 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001684 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001685
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001686 if (sgp == SGP_WRITE)
1687 mark_page_accessed(page);
1688
Hugh Dickins27ab7002011-07-25 17:12:36 -07001689 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001690 set_page_dirty(page);
1691 swap_free(swap);
1692
Hugh Dickins54af60422011-08-03 16:21:24 -07001693 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001694 /* shmem_symlink() */
1695 if (mapping->a_ops != &shmem_aops)
1696 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001697 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001698 goto alloc_nohuge;
1699 if (shmem_huge == SHMEM_HUGE_FORCE)
1700 goto alloc_huge;
1701 switch (sbinfo->huge) {
1702 loff_t i_size;
1703 pgoff_t off;
1704 case SHMEM_HUGE_NEVER:
1705 goto alloc_nohuge;
1706 case SHMEM_HUGE_WITHIN_SIZE:
1707 off = round_up(index, HPAGE_PMD_NR);
1708 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1709 if (i_size >= HPAGE_PMD_SIZE &&
1710 i_size >> PAGE_SHIFT >= off)
1711 goto alloc_huge;
1712 /* fallthrough */
1713 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001714 if (sgp_huge == SGP_HUGE)
1715 goto alloc_huge;
1716 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001717 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001720alloc_huge:
1721 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1722 index, true);
1723 if (IS_ERR(page)) {
1724alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1725 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001726 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001727 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001728 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001729 error = PTR_ERR(page);
1730 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001731 if (error != -ENOSPC)
1732 goto failed;
1733 /*
1734 * Try to reclaim some spece by splitting a huge page
1735 * beyond i_size on the filesystem.
1736 */
1737 while (retry--) {
1738 int ret;
1739 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1740 if (ret == SHRINK_STOP)
1741 break;
1742 if (ret)
1743 goto alloc_nohuge;
1744 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001745 goto failed;
1746 }
1747
1748 if (PageTransHuge(page))
1749 hindex = round_down(index, HPAGE_PMD_NR);
1750 else
1751 hindex = index;
1752
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001753 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001754 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001755
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001756 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001757 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001758 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001759 goto unacct;
1760 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1761 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001762 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001763 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001764 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001765 radix_tree_preload_end();
1766 }
1767 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001768 mem_cgroup_cancel_charge(page, memcg,
1769 PageTransHuge(page));
1770 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001771 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001772 mem_cgroup_commit_charge(page, memcg, false,
1773 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001774 lru_cache_add_anon(page);
1775
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001776 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001777 info->alloced += 1 << compound_order(page);
1778 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001779 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001780 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001781 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001782
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001783 if (PageTransHuge(page) &&
1784 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1785 hindex + HPAGE_PMD_NR - 1) {
1786 /*
1787 * Part of the huge page is beyond i_size: subject
1788 * to shrink under memory pressure.
1789 */
1790 spin_lock(&sbinfo->shrinklist_lock);
Cong Wange2286912017-08-10 15:24:24 -07001791 /*
1792 * _careful to defend against unlocked access to
1793 * ->shrink_list in shmem_unused_huge_shrink()
1794 */
1795 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001796 list_add_tail(&info->shrinklist,
1797 &sbinfo->shrinklist);
1798 sbinfo->shrinklist_len++;
1799 }
1800 spin_unlock(&sbinfo->shrinklist_lock);
1801 }
1802
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001803 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001804 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1805 */
1806 if (sgp == SGP_FALLOC)
1807 sgp = SGP_WRITE;
1808clear:
1809 /*
1810 * Let SGP_WRITE caller clear ends if write does not fill page;
1811 * but SGP_FALLOC on a page fallocated earlier must initialize
1812 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001813 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001814 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1815 struct page *head = compound_head(page);
1816 int i;
1817
1818 for (i = 0; i < (1 << compound_order(head)); i++) {
1819 clear_highpage(head + i);
1820 flush_dcache_page(head + i);
1821 }
1822 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001823 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001824 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001825
Hugh Dickins54af60422011-08-03 16:21:24 -07001826 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001827 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001828 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001829 if (alloced) {
1830 ClearPageDirty(page);
1831 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001832 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001833 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001834 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001835 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001836 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001837 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001838 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001839 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001840 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001841
Nick Piggind0217ac2007-07-19 01:47:03 -07001842 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001843 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001845unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001846 if (sbinfo->max_blocks)
1847 percpu_counter_sub(&sbinfo->used_blocks,
1848 1 << compound_order(page));
1849 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1850
1851 if (PageTransHuge(page)) {
1852 unlock_page(page);
1853 put_page(page);
1854 goto alloc_nohuge;
1855 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001856failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001857 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001858 error = -EEXIST;
1859unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001860 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001861 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001862 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001863 }
1864 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001865 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001866 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001867 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001868 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001869 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001870 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001871 goto repeat;
1872 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873}
1874
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001875/*
1876 * This is like autoremove_wake_function, but it removes the wait queue
1877 * entry unconditionally - even if something else had already woken the
1878 * target.
1879 */
1880static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1881{
1882 int ret = default_wake_function(wait, mode, sync, key);
1883 list_del_init(&wait->task_list);
1884 return ret;
1885}
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1888{
Al Viro496ad9a2013-01-23 17:07:38 -05001889 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001890 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001891 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001893 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001895 /*
1896 * Trinity finds that probing a hole which tmpfs is punching can
1897 * prevent the hole-punch from ever completing: which in turn
1898 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001899 * faulting pages into the hole while it's being punched. Although
1900 * shmem_undo_range() does remove the additions, it may be unable to
1901 * keep up, as each new page needs its own unmap_mapping_range() call,
1902 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1903 *
1904 * It does not matter if we sometimes reach this check just before the
1905 * hole-punch begins, so that one fault then races with the punch:
1906 * we just need to make racing faults a rare case.
1907 *
1908 * The implementation below would be much simpler if we just used a
1909 * standard mutex or completion: but we cannot take i_mutex in fault,
1910 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001911 */
1912 if (unlikely(inode->i_private)) {
1913 struct shmem_falloc *shmem_falloc;
1914
1915 spin_lock(&inode->i_lock);
1916 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001917 if (shmem_falloc &&
1918 shmem_falloc->waitq &&
1919 vmf->pgoff >= shmem_falloc->start &&
1920 vmf->pgoff < shmem_falloc->next) {
1921 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001922 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001923
1924 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001925 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1926 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001927 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001928 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001929 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001930 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001931
1932 shmem_falloc_waitq = shmem_falloc->waitq;
1933 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1934 TASK_UNINTERRUPTIBLE);
1935 spin_unlock(&inode->i_lock);
1936 schedule();
1937
1938 /*
1939 * shmem_falloc_waitq points into the shmem_fallocate()
1940 * stack of the hole-punching task: shmem_falloc_waitq
1941 * is usually invalid by the time we reach here, but
1942 * finish_wait() does not dereference it in that case;
1943 * though i_lock needed lest racing with wake_up_all().
1944 */
1945 spin_lock(&inode->i_lock);
1946 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1947 spin_unlock(&inode->i_lock);
1948 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001949 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001950 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001951 }
1952
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001953 sgp = SGP_CACHE;
1954 if (vma->vm_flags & VM_HUGEPAGE)
1955 sgp = SGP_HUGE;
1956 else if (vma->vm_flags & VM_NOHUGEPAGE)
1957 sgp = SGP_NOHUGE;
1958
1959 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001960 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 if (error)
1962 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001963 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964}
1965
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001966unsigned long shmem_get_unmapped_area(struct file *file,
1967 unsigned long uaddr, unsigned long len,
1968 unsigned long pgoff, unsigned long flags)
1969{
1970 unsigned long (*get_area)(struct file *,
1971 unsigned long, unsigned long, unsigned long, unsigned long);
1972 unsigned long addr;
1973 unsigned long offset;
1974 unsigned long inflated_len;
1975 unsigned long inflated_addr;
1976 unsigned long inflated_offset;
1977
1978 if (len > TASK_SIZE)
1979 return -ENOMEM;
1980
1981 get_area = current->mm->get_unmapped_area;
1982 addr = get_area(file, uaddr, len, pgoff, flags);
1983
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001984 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001985 return addr;
1986 if (IS_ERR_VALUE(addr))
1987 return addr;
1988 if (addr & ~PAGE_MASK)
1989 return addr;
1990 if (addr > TASK_SIZE - len)
1991 return addr;
1992
1993 if (shmem_huge == SHMEM_HUGE_DENY)
1994 return addr;
1995 if (len < HPAGE_PMD_SIZE)
1996 return addr;
1997 if (flags & MAP_FIXED)
1998 return addr;
1999 /*
2000 * Our priority is to support MAP_SHARED mapped hugely;
2001 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2002 * But if caller specified an address hint, respect that as before.
2003 */
2004 if (uaddr)
2005 return addr;
2006
2007 if (shmem_huge != SHMEM_HUGE_FORCE) {
2008 struct super_block *sb;
2009
2010 if (file) {
2011 VM_BUG_ON(file->f_op != &shmem_file_operations);
2012 sb = file_inode(file)->i_sb;
2013 } else {
2014 /*
2015 * Called directly from mm/mmap.c, or drivers/char/mem.c
2016 * for "/dev/zero", to create a shared anonymous object.
2017 */
2018 if (IS_ERR(shm_mnt))
2019 return addr;
2020 sb = shm_mnt->mnt_sb;
2021 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002022 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002023 return addr;
2024 }
2025
2026 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2027 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2028 return addr;
2029 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2030 return addr;
2031
2032 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2033 if (inflated_len > TASK_SIZE)
2034 return addr;
2035 if (inflated_len < len)
2036 return addr;
2037
2038 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2039 if (IS_ERR_VALUE(inflated_addr))
2040 return addr;
2041 if (inflated_addr & ~PAGE_MASK)
2042 return addr;
2043
2044 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2045 inflated_addr += offset - inflated_offset;
2046 if (inflated_offset > offset)
2047 inflated_addr += HPAGE_PMD_SIZE;
2048
2049 if (inflated_addr > TASK_SIZE - len)
2050 return addr;
2051 return inflated_addr;
2052}
2053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002055static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056{
Al Viro496ad9a2013-01-23 17:07:38 -05002057 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002058 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059}
2060
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002061static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2062 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063{
Al Viro496ad9a2013-01-23 17:07:38 -05002064 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002065 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002067 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2068 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069}
2070#endif
2071
2072int shmem_lock(struct file *file, int lock, struct user_struct *user)
2073{
Al Viro496ad9a2013-01-23 17:07:38 -05002074 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 struct shmem_inode_info *info = SHMEM_I(inode);
2076 int retval = -ENOMEM;
2077
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002078 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (lock && !(info->flags & VM_LOCKED)) {
2080 if (!user_shm_lock(inode->i_size, user))
2081 goto out_nomem;
2082 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002083 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 }
2085 if (!lock && (info->flags & VM_LOCKED) && user) {
2086 user_shm_unlock(inode->i_size, user);
2087 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002088 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 }
2090 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002091
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002093 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 return retval;
2095}
2096
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002097static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098{
2099 file_accessed(file);
2100 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002101 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002102 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2103 (vma->vm_end & HPAGE_PMD_MASK)) {
2104 khugepaged_enter(vma, vma->vm_flags);
2105 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return 0;
2107}
2108
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002109static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002110 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111{
2112 struct inode *inode;
2113 struct shmem_inode_info *info;
2114 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2115
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002116 if (shmem_reserve_inode(sb))
2117 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 inode = new_inode(sb);
2120 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002121 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002122 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002124 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002125 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 info = SHMEM_I(inode);
2127 memset(info, 0, (char *)inode - (char *)info);
2128 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002129 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002130 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002131 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002133 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002134 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
2136 switch (mode & S_IFMT) {
2137 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002138 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 init_special_inode(inode, mode, dev);
2140 break;
2141 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002142 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 inode->i_op = &shmem_inode_operations;
2144 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002145 mpol_shared_policy_init(&info->policy,
2146 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 break;
2148 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002149 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 /* Some things misbehave if size == 0 on a directory */
2151 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2152 inode->i_op = &shmem_dir_inode_operations;
2153 inode->i_fop = &simple_dir_operations;
2154 break;
2155 case S_IFLNK:
2156 /*
2157 * Must not load anything in the rbtree,
2158 * mpol_free_shared_policy will not be called.
2159 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002160 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 break;
2162 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002163 } else
2164 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 return inode;
2166}
2167
Johannes Weiner0cd61442014-04-03 14:47:46 -07002168bool shmem_mapping(struct address_space *mapping)
2169{
Sasha Levinf0774d82015-02-23 05:38:00 -05002170 if (!mapping->host)
2171 return false;
2172
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002173 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002174}
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002177static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002178static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002180#ifdef CONFIG_TMPFS_XATTR
2181static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2182#else
2183#define shmem_initxattrs NULL
2184#endif
2185
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002187shmem_write_begin(struct file *file, struct address_space *mapping,
2188 loff_t pos, unsigned len, unsigned flags,
2189 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
Nick Piggin800d15a2007-10-16 01:25:03 -07002191 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002192 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002193 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002194
2195 /* i_mutex is held by caller */
2196 if (unlikely(info->seals)) {
2197 if (info->seals & F_SEAL_WRITE)
2198 return -EPERM;
2199 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2200 return -EPERM;
2201 }
2202
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002203 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002204}
2205
2206static int
2207shmem_write_end(struct file *file, struct address_space *mapping,
2208 loff_t pos, unsigned len, unsigned copied,
2209 struct page *page, void *fsdata)
2210{
2211 struct inode *inode = mapping->host;
2212
Hugh Dickinsd3602442008-02-04 22:28:44 -08002213 if (pos + copied > inode->i_size)
2214 i_size_write(inode, pos + copied);
2215
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002216 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002217 struct page *head = compound_head(page);
2218 if (PageTransCompound(page)) {
2219 int i;
2220
2221 for (i = 0; i < HPAGE_PMD_NR; i++) {
2222 if (head + i == page)
2223 continue;
2224 clear_highpage(head + i);
2225 flush_dcache_page(head + i);
2226 }
2227 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002228 if (copied < PAGE_SIZE) {
2229 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002230 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002231 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002232 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002233 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002234 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002235 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002236 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002237 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002238
Nick Piggin800d15a2007-10-16 01:25:03 -07002239 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240}
2241
Al Viro2ba5bbe2014-04-02 20:00:02 -04002242static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243{
Al Viro6e58e792014-02-03 17:07:03 -05002244 struct file *file = iocb->ki_filp;
2245 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002247 pgoff_t index;
2248 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002249 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002250 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002251 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002252 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002253
2254 /*
2255 * Might this read be for a stacking filesystem? Then when reading
2256 * holes of a sparse file, we actually need to allocate those pages,
2257 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2258 */
Al Viro777eda22014-12-17 04:46:46 -05002259 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002260 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002262 index = *ppos >> PAGE_SHIFT;
2263 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265 for (;;) {
2266 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002267 pgoff_t end_index;
2268 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 loff_t i_size = i_size_read(inode);
2270
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002271 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (index > end_index)
2273 break;
2274 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002275 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 if (nr <= offset)
2277 break;
2278 }
2279
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002280 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002281 if (error) {
2282 if (error == -EINVAL)
2283 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 break;
2285 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002286 if (page) {
2287 if (sgp == SGP_CACHE)
2288 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002289 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292 /*
2293 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002294 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002296 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002298 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002300 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 if (nr <= offset) {
2302 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002303 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 break;
2305 }
2306 }
2307 nr -= offset;
2308
2309 if (page) {
2310 /*
2311 * If users can be writing to this page using arbitrary
2312 * virtual addresses, take care about potential aliasing
2313 * before reading the page on the kernel side.
2314 */
2315 if (mapping_writably_mapped(mapping))
2316 flush_dcache_page(page);
2317 /*
2318 * Mark the page accessed if we read the beginning.
2319 */
2320 if (!offset)
2321 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002322 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002324 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
2327 /*
2328 * Ok, we have the page, and it's up-to-date, so
2329 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002331 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002332 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002334 index += offset >> PAGE_SHIFT;
2335 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002337 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002338 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 break;
Al Viro6e58e792014-02-03 17:07:03 -05002340 if (ret < nr) {
2341 error = -EFAULT;
2342 break;
2343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 cond_resched();
2345 }
2346
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002347 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002348 file_accessed(file);
2349 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350}
2351
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002352/*
2353 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2354 */
2355static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002356 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002357{
2358 struct page *page;
2359 struct pagevec pvec;
2360 pgoff_t indices[PAGEVEC_SIZE];
2361 bool done = false;
2362 int i;
2363
2364 pagevec_init(&pvec, 0);
2365 pvec.nr = 1; /* start small: we may be there already */
2366 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002367 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002368 pvec.nr, pvec.pages, indices);
2369 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002370 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002371 index = end;
2372 break;
2373 }
2374 for (i = 0; i < pvec.nr; i++, index++) {
2375 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002376 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002377 done = true;
2378 break;
2379 }
2380 index = indices[i];
2381 }
2382 page = pvec.pages[i];
2383 if (page && !radix_tree_exceptional_entry(page)) {
2384 if (!PageUptodate(page))
2385 page = NULL;
2386 }
2387 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002388 (page && whence == SEEK_DATA) ||
2389 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002390 done = true;
2391 break;
2392 }
2393 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002394 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002395 pagevec_release(&pvec);
2396 pvec.nr = PAGEVEC_SIZE;
2397 cond_resched();
2398 }
2399 return index;
2400}
2401
Andrew Morton965c8e52012-12-17 15:59:39 -08002402static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002403{
2404 struct address_space *mapping = file->f_mapping;
2405 struct inode *inode = mapping->host;
2406 pgoff_t start, end;
2407 loff_t new_offset;
2408
Andrew Morton965c8e52012-12-17 15:59:39 -08002409 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2410 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002411 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002412 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002413 /* We're holding i_mutex so we can access i_size directly */
2414
2415 if (offset < 0)
2416 offset = -EINVAL;
2417 else if (offset >= inode->i_size)
2418 offset = -ENXIO;
2419 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002420 start = offset >> PAGE_SHIFT;
2421 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002422 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002423 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002424 if (new_offset > offset) {
2425 if (new_offset < inode->i_size)
2426 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002427 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002428 offset = -ENXIO;
2429 else
2430 offset = inode->i_size;
2431 }
2432 }
2433
Hugh Dickins387aae62013-08-04 11:30:25 -07002434 if (offset >= 0)
2435 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002436 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002437 return offset;
2438}
2439
David Herrmann05f65b52014-08-08 14:25:36 -07002440/*
2441 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2442 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2443 */
2444#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2445#define LAST_SCAN 4 /* about 150ms max */
2446
2447static void shmem_tag_pins(struct address_space *mapping)
2448{
2449 struct radix_tree_iter iter;
2450 void **slot;
2451 pgoff_t start;
2452 struct page *page;
2453
2454 lru_add_drain();
2455 start = 0;
2456 rcu_read_lock();
2457
David Herrmann05f65b52014-08-08 14:25:36 -07002458 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2459 page = radix_tree_deref_slot(slot);
2460 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002461 if (radix_tree_deref_retry(page)) {
2462 slot = radix_tree_iter_retry(&iter);
2463 continue;
2464 }
David Herrmann05f65b52014-08-08 14:25:36 -07002465 } else if (page_count(page) - page_mapcount(page) > 1) {
2466 spin_lock_irq(&mapping->tree_lock);
2467 radix_tree_tag_set(&mapping->page_tree, iter.index,
2468 SHMEM_TAG_PINNED);
2469 spin_unlock_irq(&mapping->tree_lock);
2470 }
2471
2472 if (need_resched()) {
2473 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002474 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002475 }
2476 }
2477 rcu_read_unlock();
2478}
2479
2480/*
2481 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2482 * via get_user_pages(), drivers might have some pending I/O without any active
2483 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2484 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2485 * them to be dropped.
2486 * The caller must guarantee that no new user will acquire writable references
2487 * to those pages to avoid races.
2488 */
David Herrmann40e041a2014-08-08 14:25:27 -07002489static int shmem_wait_for_pins(struct address_space *mapping)
2490{
David Herrmann05f65b52014-08-08 14:25:36 -07002491 struct radix_tree_iter iter;
2492 void **slot;
2493 pgoff_t start;
2494 struct page *page;
2495 int error, scan;
2496
2497 shmem_tag_pins(mapping);
2498
2499 error = 0;
2500 for (scan = 0; scan <= LAST_SCAN; scan++) {
2501 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2502 break;
2503
2504 if (!scan)
2505 lru_add_drain_all();
2506 else if (schedule_timeout_killable((HZ << scan) / 200))
2507 scan = LAST_SCAN;
2508
2509 start = 0;
2510 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002511 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2512 start, SHMEM_TAG_PINNED) {
2513
2514 page = radix_tree_deref_slot(slot);
2515 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002516 if (radix_tree_deref_retry(page)) {
2517 slot = radix_tree_iter_retry(&iter);
2518 continue;
2519 }
David Herrmann05f65b52014-08-08 14:25:36 -07002520
2521 page = NULL;
2522 }
2523
2524 if (page &&
2525 page_count(page) - page_mapcount(page) != 1) {
2526 if (scan < LAST_SCAN)
2527 goto continue_resched;
2528
2529 /*
2530 * On the last scan, we clean up all those tags
2531 * we inserted; but make a note that we still
2532 * found pages pinned.
2533 */
2534 error = -EBUSY;
2535 }
2536
2537 spin_lock_irq(&mapping->tree_lock);
2538 radix_tree_tag_clear(&mapping->page_tree,
2539 iter.index, SHMEM_TAG_PINNED);
2540 spin_unlock_irq(&mapping->tree_lock);
2541continue_resched:
2542 if (need_resched()) {
2543 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002544 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002545 }
2546 }
2547 rcu_read_unlock();
2548 }
2549
2550 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002551}
2552
2553#define F_ALL_SEALS (F_SEAL_SEAL | \
2554 F_SEAL_SHRINK | \
2555 F_SEAL_GROW | \
2556 F_SEAL_WRITE)
2557
2558int shmem_add_seals(struct file *file, unsigned int seals)
2559{
2560 struct inode *inode = file_inode(file);
2561 struct shmem_inode_info *info = SHMEM_I(inode);
2562 int error;
2563
2564 /*
2565 * SEALING
2566 * Sealing allows multiple parties to share a shmem-file but restrict
2567 * access to a specific subset of file operations. Seals can only be
2568 * added, but never removed. This way, mutually untrusted parties can
2569 * share common memory regions with a well-defined policy. A malicious
2570 * peer can thus never perform unwanted operations on a shared object.
2571 *
2572 * Seals are only supported on special shmem-files and always affect
2573 * the whole underlying inode. Once a seal is set, it may prevent some
2574 * kinds of access to the file. Currently, the following seals are
2575 * defined:
2576 * SEAL_SEAL: Prevent further seals from being set on this file
2577 * SEAL_SHRINK: Prevent the file from shrinking
2578 * SEAL_GROW: Prevent the file from growing
2579 * SEAL_WRITE: Prevent write access to the file
2580 *
2581 * As we don't require any trust relationship between two parties, we
2582 * must prevent seals from being removed. Therefore, sealing a file
2583 * only adds a given set of seals to the file, it never touches
2584 * existing seals. Furthermore, the "setting seals"-operation can be
2585 * sealed itself, which basically prevents any further seal from being
2586 * added.
2587 *
2588 * Semantics of sealing are only defined on volatile files. Only
2589 * anonymous shmem files support sealing. More importantly, seals are
2590 * never written to disk. Therefore, there's no plan to support it on
2591 * other file types.
2592 */
2593
2594 if (file->f_op != &shmem_file_operations)
2595 return -EINVAL;
2596 if (!(file->f_mode & FMODE_WRITE))
2597 return -EPERM;
2598 if (seals & ~(unsigned int)F_ALL_SEALS)
2599 return -EINVAL;
2600
Al Viro59551022016-01-22 15:40:57 -05002601 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002602
2603 if (info->seals & F_SEAL_SEAL) {
2604 error = -EPERM;
2605 goto unlock;
2606 }
2607
2608 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2609 error = mapping_deny_writable(file->f_mapping);
2610 if (error)
2611 goto unlock;
2612
2613 error = shmem_wait_for_pins(file->f_mapping);
2614 if (error) {
2615 mapping_allow_writable(file->f_mapping);
2616 goto unlock;
2617 }
2618 }
2619
2620 info->seals |= seals;
2621 error = 0;
2622
2623unlock:
Al Viro59551022016-01-22 15:40:57 -05002624 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002625 return error;
2626}
2627EXPORT_SYMBOL_GPL(shmem_add_seals);
2628
2629int shmem_get_seals(struct file *file)
2630{
2631 if (file->f_op != &shmem_file_operations)
2632 return -EINVAL;
2633
2634 return SHMEM_I(file_inode(file))->seals;
2635}
2636EXPORT_SYMBOL_GPL(shmem_get_seals);
2637
2638long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2639{
2640 long error;
2641
2642 switch (cmd) {
2643 case F_ADD_SEALS:
2644 /* disallow upper 32bit */
2645 if (arg > UINT_MAX)
2646 return -EINVAL;
2647
2648 error = shmem_add_seals(file, arg);
2649 break;
2650 case F_GET_SEALS:
2651 error = shmem_get_seals(file);
2652 break;
2653 default:
2654 error = -EINVAL;
2655 break;
2656 }
2657
2658 return error;
2659}
2660
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002661static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2662 loff_t len)
2663{
Al Viro496ad9a2013-01-23 17:07:38 -05002664 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002665 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002666 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002667 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002668 pgoff_t start, index, end;
2669 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002670
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002671 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2672 return -EOPNOTSUPP;
2673
Al Viro59551022016-01-22 15:40:57 -05002674 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002675
2676 if (mode & FALLOC_FL_PUNCH_HOLE) {
2677 struct address_space *mapping = file->f_mapping;
2678 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2679 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002680 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002681
David Herrmann40e041a2014-08-08 14:25:27 -07002682 /* protected by i_mutex */
2683 if (info->seals & F_SEAL_WRITE) {
2684 error = -EPERM;
2685 goto out;
2686 }
2687
Hugh Dickins8e205f72014-07-23 14:00:10 -07002688 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002689 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2690 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2691 spin_lock(&inode->i_lock);
2692 inode->i_private = &shmem_falloc;
2693 spin_unlock(&inode->i_lock);
2694
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002695 if ((u64)unmap_end > (u64)unmap_start)
2696 unmap_mapping_range(mapping, unmap_start,
2697 1 + unmap_end - unmap_start, 0);
2698 shmem_truncate_range(inode, offset, offset + len - 1);
2699 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002700
2701 spin_lock(&inode->i_lock);
2702 inode->i_private = NULL;
2703 wake_up_all(&shmem_falloc_waitq);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002704 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002705 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002706 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002707 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002708 }
2709
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002710 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2711 error = inode_newsize_ok(inode, offset + len);
2712 if (error)
2713 goto out;
2714
David Herrmann40e041a2014-08-08 14:25:27 -07002715 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2716 error = -EPERM;
2717 goto out;
2718 }
2719
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002720 start = offset >> PAGE_SHIFT;
2721 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002722 /* Try to avoid a swapstorm if len is impossible to satisfy */
2723 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2724 error = -ENOSPC;
2725 goto out;
2726 }
2727
Hugh Dickins8e205f72014-07-23 14:00:10 -07002728 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002729 shmem_falloc.start = start;
2730 shmem_falloc.next = start;
2731 shmem_falloc.nr_falloced = 0;
2732 shmem_falloc.nr_unswapped = 0;
2733 spin_lock(&inode->i_lock);
2734 inode->i_private = &shmem_falloc;
2735 spin_unlock(&inode->i_lock);
2736
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002737 for (index = start; index < end; index++) {
2738 struct page *page;
2739
2740 /*
2741 * Good, the fallocate(2) manpage permits EINTR: we may have
2742 * been interrupted because we are using up too much memory.
2743 */
2744 if (signal_pending(current))
2745 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002746 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2747 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002748 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002749 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002750 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002751 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002752 if (index > start) {
2753 shmem_undo_range(inode,
2754 (loff_t)start << PAGE_SHIFT,
2755 ((loff_t)index << PAGE_SHIFT) - 1, true);
2756 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002757 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002758 }
2759
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002760 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002761 * Inform shmem_writepage() how far we have reached.
2762 * No need for lock or barrier: we have the page lock.
2763 */
2764 shmem_falloc.next++;
2765 if (!PageUptodate(page))
2766 shmem_falloc.nr_falloced++;
2767
2768 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002769 * If !PageUptodate, leave it that way so that freeable pages
2770 * can be recognized if we need to rollback on error later.
2771 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002772 * than free the pages we are allocating (and SGP_CACHE pages
2773 * might still be clean: we now need to mark those dirty too).
2774 */
2775 set_page_dirty(page);
2776 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002777 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002778 cond_resched();
2779 }
2780
2781 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2782 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002783 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002784undone:
2785 spin_lock(&inode->i_lock);
2786 inode->i_private = NULL;
2787 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002788out:
Al Viro59551022016-01-22 15:40:57 -05002789 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002790 return error;
2791}
2792
David Howells726c3342006-06-23 02:02:58 -07002793static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794{
David Howells726c3342006-06-23 02:02:58 -07002795 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
2797 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002798 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002800 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002802 buf->f_bavail =
2803 buf->f_bfree = sbinfo->max_blocks -
2804 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002805 }
2806 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 buf->f_files = sbinfo->max_inodes;
2808 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 }
2810 /* else leave those fields 0 like simple_statfs */
2811 return 0;
2812}
2813
2814/*
2815 * File creation. Allocate an inode, and we're done..
2816 */
2817static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002818shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002820 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 int error = -ENOSPC;
2822
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002823 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002825 error = simple_acl_create(dir, inode);
2826 if (error)
2827 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002828 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002829 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002830 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002831 if (error && error != -EOPNOTSUPP)
2832 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002833
Al Viro718deb62009-12-16 19:35:36 -05002834 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002836 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 d_instantiate(dentry, inode);
2838 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 }
2840 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002841out_iput:
2842 iput(inode);
2843 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844}
2845
Al Viro60545d02013-06-07 01:20:27 -04002846static int
2847shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2848{
2849 struct inode *inode;
2850 int error = -ENOSPC;
2851
2852 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2853 if (inode) {
2854 error = security_inode_init_security(inode, dir,
2855 NULL,
2856 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002857 if (error && error != -EOPNOTSUPP)
2858 goto out_iput;
2859 error = simple_acl_create(dir, inode);
2860 if (error)
2861 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002862 d_tmpfile(dentry, inode);
2863 }
2864 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002865out_iput:
2866 iput(inode);
2867 return error;
Al Viro60545d02013-06-07 01:20:27 -04002868}
2869
Al Viro18bb1db2011-07-26 01:41:39 -04002870static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871{
2872 int error;
2873
2874 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2875 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002876 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 return 0;
2878}
2879
Al Viro4acdaf22011-07-26 01:42:34 -04002880static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002881 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882{
2883 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2884}
2885
2886/*
2887 * Link a file..
2888 */
2889static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2890{
David Howells75c3cfa2015-03-17 22:26:12 +00002891 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002892 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
2894 /*
2895 * No ordinary (disk based) filesystem counts links as inodes;
2896 * but each new link needs a new dentry, pinning lowmem, and
2897 * tmpfs dentries cannot be pruned until they are unlinked.
2898 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002899 ret = shmem_reserve_inode(inode->i_sb);
2900 if (ret)
2901 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
2903 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002904 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002905 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002906 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 dget(dentry); /* Extra pinning count for the created dentry */
2908 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002909out:
2910 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911}
2912
2913static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2914{
David Howells75c3cfa2015-03-17 22:26:12 +00002915 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002917 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2918 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
2920 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002921 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002922 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 dput(dentry); /* Undo the count from "create" - this does all the work */
2924 return 0;
2925}
2926
2927static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2928{
2929 if (!simple_empty(dentry))
2930 return -ENOTEMPTY;
2931
David Howells75c3cfa2015-03-17 22:26:12 +00002932 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002933 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 return shmem_unlink(dir, dentry);
2935}
2936
Miklos Szeredi37456772014-07-23 15:15:34 +02002937static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2938{
David Howellse36cb0b2015-01-29 12:02:35 +00002939 bool old_is_dir = d_is_dir(old_dentry);
2940 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002941
2942 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2943 if (old_is_dir) {
2944 drop_nlink(old_dir);
2945 inc_nlink(new_dir);
2946 } else {
2947 drop_nlink(new_dir);
2948 inc_nlink(old_dir);
2949 }
2950 }
2951 old_dir->i_ctime = old_dir->i_mtime =
2952 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002953 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002954 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02002955
2956 return 0;
2957}
2958
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002959static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2960{
2961 struct dentry *whiteout;
2962 int error;
2963
2964 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2965 if (!whiteout)
2966 return -ENOMEM;
2967
2968 error = shmem_mknod(old_dir, whiteout,
2969 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2970 dput(whiteout);
2971 if (error)
2972 return error;
2973
2974 /*
2975 * Cheat and hash the whiteout while the old dentry is still in
2976 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2977 *
2978 * d_lookup() will consistently find one of them at this point,
2979 * not sure which one, but that isn't even important.
2980 */
2981 d_rehash(whiteout);
2982 return 0;
2983}
2984
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985/*
2986 * The VFS layer already does all the dentry stuff for rename,
2987 * we just have to decrement the usage count for the target if
2988 * it exists so that the VFS layer correctly free's it when it
2989 * gets overwritten.
2990 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002991static 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 -07002992{
David Howells75c3cfa2015-03-17 22:26:12 +00002993 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 int they_are_dirs = S_ISDIR(inode->i_mode);
2995
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002996 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002997 return -EINVAL;
2998
Miklos Szeredi37456772014-07-23 15:15:34 +02002999 if (flags & RENAME_EXCHANGE)
3000 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3001
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 if (!simple_empty(new_dentry))
3003 return -ENOTEMPTY;
3004
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003005 if (flags & RENAME_WHITEOUT) {
3006 int error;
3007
3008 error = shmem_whiteout(old_dir, old_dentry);
3009 if (error)
3010 return error;
3011 }
3012
David Howells75c3cfa2015-03-17 22:26:12 +00003013 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003015 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003016 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003017 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003020 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003021 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 }
3023
3024 old_dir->i_size -= BOGO_DIRENT_SIZE;
3025 new_dir->i_size += BOGO_DIRENT_SIZE;
3026 old_dir->i_ctime = old_dir->i_mtime =
3027 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003028 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 return 0;
3030}
3031
3032static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3033{
3034 int error;
3035 int len;
3036 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003037 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 struct shmem_inode_info *info;
3039
3040 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003041 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 return -ENAMETOOLONG;
3043
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003044 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 if (!inode)
3046 return -ENOSPC;
3047
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003048 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003049 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003050 if (error) {
3051 if (error != -EOPNOTSUPP) {
3052 iput(inode);
3053 return error;
3054 }
3055 error = 0;
3056 }
3057
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 info = SHMEM_I(inode);
3059 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003060 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003061 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3062 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003063 iput(inode);
3064 return -ENOMEM;
3065 }
3066 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003068 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003069 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 if (error) {
3071 iput(inode);
3072 return error;
3073 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003074 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003076 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003077 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003079 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003080 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003083 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 d_instantiate(dentry, inode);
3085 dget(dentry);
3086 return 0;
3087}
3088
Al Virofceef392015-12-29 15:58:39 -05003089static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090{
Al Virofceef392015-12-29 15:58:39 -05003091 mark_page_accessed(arg);
3092 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093}
3094
Al Viro6b255392015-11-17 10:20:54 -05003095static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003096 struct inode *inode,
3097 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003100 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003101 if (!dentry) {
3102 page = find_get_page(inode->i_mapping, 0);
3103 if (!page)
3104 return ERR_PTR(-ECHILD);
3105 if (!PageUptodate(page)) {
3106 put_page(page);
3107 return ERR_PTR(-ECHILD);
3108 }
3109 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003110 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003111 if (error)
3112 return ERR_PTR(error);
3113 unlock_page(page);
3114 }
Al Virofceef392015-12-29 15:58:39 -05003115 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003116 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117}
3118
Eric Parisb09e0fa2011-05-24 17:12:39 -07003119#ifdef CONFIG_TMPFS_XATTR
3120/*
3121 * Superblocks without xattr inode operations may get some security.* xattr
3122 * support from the LSM "for free". As soon as we have any other xattrs
3123 * like ACLs, we also need to implement the security.* handlers at
3124 * filesystem level, though.
3125 */
3126
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003127/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003128 * Callback for security_inode_init_security() for acquiring xattrs.
3129 */
3130static int shmem_initxattrs(struct inode *inode,
3131 const struct xattr *xattr_array,
3132 void *fs_info)
3133{
3134 struct shmem_inode_info *info = SHMEM_I(inode);
3135 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003136 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003137 size_t len;
3138
3139 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003140 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003141 if (!new_xattr)
3142 return -ENOMEM;
3143
3144 len = strlen(xattr->name) + 1;
3145 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3146 GFP_KERNEL);
3147 if (!new_xattr->name) {
3148 kfree(new_xattr);
3149 return -ENOMEM;
3150 }
3151
3152 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3153 XATTR_SECURITY_PREFIX_LEN);
3154 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3155 xattr->name, len);
3156
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003157 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003158 }
3159
3160 return 0;
3161}
3162
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003163static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003164 struct dentry *unused, struct inode *inode,
3165 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003166{
Al Virob2968212016-04-10 20:48:24 -04003167 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003168
3169 name = xattr_full_name(handler, name);
3170 return simple_xattr_get(&info->xattrs, name, buffer, size);
3171}
3172
3173static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003174 struct dentry *unused, struct inode *inode,
3175 const char *name, const void *value,
3176 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003177{
Al Viro59301222016-05-27 10:19:30 -04003178 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003179
3180 name = xattr_full_name(handler, name);
3181 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3182}
3183
3184static const struct xattr_handler shmem_security_xattr_handler = {
3185 .prefix = XATTR_SECURITY_PREFIX,
3186 .get = shmem_xattr_handler_get,
3187 .set = shmem_xattr_handler_set,
3188};
3189
3190static const struct xattr_handler shmem_trusted_xattr_handler = {
3191 .prefix = XATTR_TRUSTED_PREFIX,
3192 .get = shmem_xattr_handler_get,
3193 .set = shmem_xattr_handler_set,
3194};
3195
Eric Parisb09e0fa2011-05-24 17:12:39 -07003196static const struct xattr_handler *shmem_xattr_handlers[] = {
3197#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003198 &posix_acl_access_xattr_handler,
3199 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003200#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003201 &shmem_security_xattr_handler,
3202 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003203 NULL
3204};
3205
Eric Parisb09e0fa2011-05-24 17:12:39 -07003206static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3207{
David Howells75c3cfa2015-03-17 22:26:12 +00003208 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003209 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003210}
3211#endif /* CONFIG_TMPFS_XATTR */
3212
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003213static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003215 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003216#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003217 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003218#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219};
3220
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003221static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003223 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003224#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003225 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003226#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003227};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003228
David M. Grimes91828a42006-10-17 00:09:45 -07003229static struct dentry *shmem_get_parent(struct dentry *child)
3230{
3231 return ERR_PTR(-ESTALE);
3232}
3233
3234static int shmem_match(struct inode *ino, void *vfh)
3235{
3236 __u32 *fh = vfh;
3237 __u64 inum = fh[2];
3238 inum = (inum << 32) | fh[1];
3239 return ino->i_ino == inum && fh[0] == ino->i_generation;
3240}
3241
Christoph Hellwig480b1162007-10-21 16:42:13 -07003242static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3243 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003244{
David M. Grimes91828a42006-10-17 00:09:45 -07003245 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003246 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003247 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003248
Christoph Hellwig480b1162007-10-21 16:42:13 -07003249 if (fh_len < 3)
3250 return NULL;
3251
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003252 inum = fid->raw[2];
3253 inum = (inum << 32) | fid->raw[1];
3254
Christoph Hellwig480b1162007-10-21 16:42:13 -07003255 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3256 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003257 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003258 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003259 iput(inode);
3260 }
3261
Christoph Hellwig480b1162007-10-21 16:42:13 -07003262 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003263}
3264
Al Virob0b03822012-04-02 14:34:06 -04003265static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3266 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003267{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303268 if (*len < 3) {
3269 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003270 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303271 }
David M. Grimes91828a42006-10-17 00:09:45 -07003272
Al Viro1d3382cb2010-10-23 15:19:20 -04003273 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003274 /* Unfortunately insert_inode_hash is not idempotent,
3275 * so as we hash inodes here rather than at creation
3276 * time, we need a lock to ensure we only try
3277 * to do it once
3278 */
3279 static DEFINE_SPINLOCK(lock);
3280 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003281 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003282 __insert_inode_hash(inode,
3283 inode->i_ino + inode->i_generation);
3284 spin_unlock(&lock);
3285 }
3286
3287 fh[0] = inode->i_generation;
3288 fh[1] = inode->i_ino;
3289 fh[2] = ((__u64)inode->i_ino) >> 32;
3290
3291 *len = 3;
3292 return 1;
3293}
3294
Christoph Hellwig39655162007-10-21 16:42:17 -07003295static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003296 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003297 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003298 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003299};
3300
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003301static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3302 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303{
3304 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003305 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003306 uid_t uid;
3307 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003309 while (options != NULL) {
3310 this_char = options;
3311 for (;;) {
3312 /*
3313 * NUL-terminate this option: unfortunately,
3314 * mount options form a comma-separated list,
3315 * but mpol's nodelist may also contain commas.
3316 */
3317 options = strchr(options, ',');
3318 if (options == NULL)
3319 break;
3320 options++;
3321 if (!isdigit(*options)) {
3322 options[-1] = '\0';
3323 break;
3324 }
3325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 if (!*this_char)
3327 continue;
3328 if ((value = strchr(this_char,'=')) != NULL) {
3329 *value++ = 0;
3330 } else {
Joe Perches11705322016-03-17 14:19:50 -07003331 pr_err("tmpfs: No value for mount option '%s'\n",
3332 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003333 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 }
3335
3336 if (!strcmp(this_char,"size")) {
3337 unsigned long long size;
3338 size = memparse(value,&rest);
3339 if (*rest == '%') {
3340 size <<= PAGE_SHIFT;
3341 size *= totalram_pages;
3342 do_div(size, 100);
3343 rest++;
3344 }
3345 if (*rest)
3346 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003347 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003348 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003350 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 if (*rest)
3352 goto bad_val;
3353 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003354 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 if (*rest)
3356 goto bad_val;
3357 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003358 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003360 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 if (*rest)
3362 goto bad_val;
3363 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003364 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003366 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 if (*rest)
3368 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003369 sbinfo->uid = make_kuid(current_user_ns(), uid);
3370 if (!uid_valid(sbinfo->uid))
3371 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003373 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003375 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 if (*rest)
3377 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003378 sbinfo->gid = make_kgid(current_user_ns(), gid);
3379 if (!gid_valid(sbinfo->gid))
3380 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003381#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003382 } else if (!strcmp(this_char, "huge")) {
3383 int huge;
3384 huge = shmem_parse_huge(value);
3385 if (huge < 0)
3386 goto bad_val;
3387 if (!has_transparent_hugepage() &&
3388 huge != SHMEM_HUGE_NEVER)
3389 goto bad_val;
3390 sbinfo->huge = huge;
3391#endif
3392#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003393 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003394 mpol_put(mpol);
3395 mpol = NULL;
3396 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003397 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003398#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 } else {
Joe Perches11705322016-03-17 14:19:50 -07003400 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003401 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 }
3403 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003404 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 return 0;
3406
3407bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003408 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003410error:
3411 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412 return 1;
3413
3414}
3415
3416static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3417{
3418 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003419 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003420 unsigned long inodes;
3421 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422
Greg Thelen5f001102013-02-22 16:36:01 -08003423 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003424 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003425 return error;
3426
3427 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003428 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003429 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003430 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003431 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003432 goto out;
3433 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003434 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003435 * but we must separately disallow unlimited->limited, because
3436 * in that case we have no record of how much is already in use.
3437 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003438 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003439 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003440 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003441 goto out;
3442
3443 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003444 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003445 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003446 sbinfo->max_inodes = config.max_inodes;
3447 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003448
Greg Thelen5f001102013-02-22 16:36:01 -08003449 /*
3450 * Preserve previous mempolicy unless mpol remount option was specified.
3451 */
3452 if (config.mpol) {
3453 mpol_put(sbinfo->mpol);
3454 sbinfo->mpol = config.mpol; /* transfers initial ref */
3455 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003456out:
3457 spin_unlock(&sbinfo->stat_lock);
3458 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003460
Al Viro34c80b12011-12-08 21:32:45 -05003461static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003462{
Al Viro34c80b12011-12-08 21:32:45 -05003463 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003464
3465 if (sbinfo->max_blocks != shmem_default_max_blocks())
3466 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003467 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003468 if (sbinfo->max_inodes != shmem_default_max_inodes())
3469 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3470 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003471 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003472 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3473 seq_printf(seq, ",uid=%u",
3474 from_kuid_munged(&init_user_ns, sbinfo->uid));
3475 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3476 seq_printf(seq, ",gid=%u",
3477 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003478#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003479 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3480 if (sbinfo->huge)
3481 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3482#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003483 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003484 return 0;
3485}
David Herrmann9183df22014-08-08 14:25:29 -07003486
3487#define MFD_NAME_PREFIX "memfd:"
3488#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3489#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3490
3491#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3492
3493SYSCALL_DEFINE2(memfd_create,
3494 const char __user *, uname,
3495 unsigned int, flags)
3496{
3497 struct shmem_inode_info *info;
3498 struct file *file;
3499 int fd, error;
3500 char *name;
3501 long len;
3502
3503 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3504 return -EINVAL;
3505
3506 /* length includes terminating zero */
3507 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3508 if (len <= 0)
3509 return -EFAULT;
3510 if (len > MFD_NAME_MAX_LEN + 1)
3511 return -EINVAL;
3512
3513 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3514 if (!name)
3515 return -ENOMEM;
3516
3517 strcpy(name, MFD_NAME_PREFIX);
3518 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3519 error = -EFAULT;
3520 goto err_name;
3521 }
3522
3523 /* terminating-zero may have changed after strnlen_user() returned */
3524 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3525 error = -EFAULT;
3526 goto err_name;
3527 }
3528
3529 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3530 if (fd < 0) {
3531 error = fd;
3532 goto err_name;
3533 }
3534
3535 file = shmem_file_setup(name, 0, VM_NORESERVE);
3536 if (IS_ERR(file)) {
3537 error = PTR_ERR(file);
3538 goto err_fd;
3539 }
3540 info = SHMEM_I(file_inode(file));
3541 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3542 file->f_flags |= O_RDWR | O_LARGEFILE;
3543 if (flags & MFD_ALLOW_SEALING)
3544 info->seals &= ~F_SEAL_SEAL;
3545
3546 fd_install(fd, file);
3547 kfree(name);
3548 return fd;
3549
3550err_fd:
3551 put_unused_fd(fd);
3552err_name:
3553 kfree(name);
3554 return error;
3555}
3556
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003557#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558
3559static void shmem_put_super(struct super_block *sb)
3560{
Hugh Dickins602586a2010-08-17 15:23:56 -07003561 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3562
3563 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003564 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003565 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566 sb->s_fs_info = NULL;
3567}
3568
Kay Sievers2b2af542009-04-30 15:23:42 +02003569int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570{
3571 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003572 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003573 int err = -ENOMEM;
3574
3575 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003576 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003577 L1_CACHE_BYTES), GFP_KERNEL);
3578 if (!sbinfo)
3579 return -ENOMEM;
3580
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003581 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003582 sbinfo->uid = current_fsuid();
3583 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003584 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003586#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 /*
3588 * Per default we only allow half of the physical ram per
3589 * tmpfs instance, limiting inodes to one per page of lowmem;
3590 * but the internal instance is left unlimited.
3591 */
Al Viroca4e0512013-08-31 12:57:10 -04003592 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003593 sbinfo->max_blocks = shmem_default_max_blocks();
3594 sbinfo->max_inodes = shmem_default_max_inodes();
3595 if (shmem_parse_options(data, sbinfo, false)) {
3596 err = -EINVAL;
3597 goto failed;
3598 }
Al Viroca4e0512013-08-31 12:57:10 -04003599 } else {
3600 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 }
David M. Grimes91828a42006-10-17 00:09:45 -07003602 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003603 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604#else
3605 sb->s_flags |= MS_NOUSER;
3606#endif
3607
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003608 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003609 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003610 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003611 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003612 spin_lock_init(&sbinfo->shrinklist_lock);
3613 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003614
Hugh Dickins285b2c42011-08-03 16:21:20 -07003615 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003616 sb->s_blocksize = PAGE_SIZE;
3617 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 sb->s_magic = TMPFS_MAGIC;
3619 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003620 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003621#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003622 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003623#endif
3624#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003625 sb->s_flags |= MS_POSIXACL;
3626#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003627
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003628 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 if (!inode)
3630 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003631 inode->i_uid = sbinfo->uid;
3632 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003633 sb->s_root = d_make_root(inode);
3634 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003635 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 return 0;
3637
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638failed:
3639 shmem_put_super(sb);
3640 return err;
3641}
3642
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003643static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644
3645static struct inode *shmem_alloc_inode(struct super_block *sb)
3646{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003647 struct shmem_inode_info *info;
3648 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3649 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003651 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652}
3653
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003654static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003655{
3656 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003657 if (S_ISLNK(inode->i_mode))
3658 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003659 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3660}
3661
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662static void shmem_destroy_inode(struct inode *inode)
3663{
Al Viro09208d12011-07-26 03:15:03 -04003664 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003666 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667}
3668
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003669static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003671 struct shmem_inode_info *info = foo;
3672 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673}
3674
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003675static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676{
3677 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3678 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003679 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 return 0;
3681}
3682
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003683static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003685 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686}
3687
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003688static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003690 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003692 .write_begin = shmem_write_begin,
3693 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003695#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003696 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003697#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003698 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699};
3700
Helge Deller15ad7cd2006-12-06 20:40:36 -08003701static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003703 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003705 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003706 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003707 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003708 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003709 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003710 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003711 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712#endif
3713};
3714
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003715static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003716 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003717 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003718#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003719 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003720 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003721#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722};
3723
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003724static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725#ifdef CONFIG_TMPFS
3726 .create = shmem_create,
3727 .lookup = simple_lookup,
3728 .link = shmem_link,
3729 .unlink = shmem_unlink,
3730 .symlink = shmem_symlink,
3731 .mkdir = shmem_mkdir,
3732 .rmdir = shmem_rmdir,
3733 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003734 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003735 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003737#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003738 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003739#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003740#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003741 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003742 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003743#endif
3744};
3745
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003746static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003747#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003748 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003749#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003750#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003751 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003752 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003753#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754};
3755
Hugh Dickins759b9772007-03-05 00:30:28 -08003756static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 .alloc_inode = shmem_alloc_inode,
3758 .destroy_inode = shmem_destroy_inode,
3759#ifdef CONFIG_TMPFS
3760 .statfs = shmem_statfs,
3761 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003762 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763#endif
Al Viro1f895f72010-06-05 19:10:41 -04003764 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 .drop_inode = generic_delete_inode,
3766 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003767#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3768 .nr_cached_objects = shmem_unused_huge_count,
3769 .free_cached_objects = shmem_unused_huge_scan,
3770#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771};
3772
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003773static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003774 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003775 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776#ifdef CONFIG_NUMA
3777 .set_policy = shmem_set_policy,
3778 .get_policy = shmem_get_policy,
3779#endif
3780};
3781
Al Viro3c26ff62010-07-25 11:46:36 +04003782static struct dentry *shmem_mount(struct file_system_type *fs_type,
3783 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784{
Al Viro3c26ff62010-07-25 11:46:36 +04003785 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786}
3787
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003788static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789 .owner = THIS_MODULE,
3790 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003791 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003793 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003796int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797{
3798 int error;
3799
Rob Landley16203a72013-09-11 14:26:12 -07003800 /* If rootfs called this, don't re-init */
3801 if (shmem_inode_cachep)
3802 return 0;
3803
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003804 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805 if (error)
3806 goto out3;
3807
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003808 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003810 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 goto out2;
3812 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003813
Al Viroca4e0512013-08-31 12:57:10 -04003814 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 if (IS_ERR(shm_mnt)) {
3816 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003817 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 goto out1;
3819 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003820
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003821#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5d8b3cc2017-08-25 15:55:33 -07003822 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003823 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3824 else
3825 shmem_huge = 0; /* just in case it was patched */
3826#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 return 0;
3828
3829out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003830 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003832 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833out3:
3834 shm_mnt = ERR_PTR(error);
3835 return error;
3836}
Matt Mackall853ac432009-01-06 14:40:20 -08003837
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003838#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003839static ssize_t shmem_enabled_show(struct kobject *kobj,
3840 struct kobj_attribute *attr, char *buf)
3841{
3842 int values[] = {
3843 SHMEM_HUGE_ALWAYS,
3844 SHMEM_HUGE_WITHIN_SIZE,
3845 SHMEM_HUGE_ADVISE,
3846 SHMEM_HUGE_NEVER,
3847 SHMEM_HUGE_DENY,
3848 SHMEM_HUGE_FORCE,
3849 };
3850 int i, count;
3851
3852 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3853 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3854
3855 count += sprintf(buf + count, fmt,
3856 shmem_format_huge(values[i]));
3857 }
3858 buf[count - 1] = '\n';
3859 return count;
3860}
3861
3862static ssize_t shmem_enabled_store(struct kobject *kobj,
3863 struct kobj_attribute *attr, const char *buf, size_t count)
3864{
3865 char tmp[16];
3866 int huge;
3867
3868 if (count + 1 > sizeof(tmp))
3869 return -EINVAL;
3870 memcpy(tmp, buf, count);
3871 tmp[count] = '\0';
3872 if (count && tmp[count - 1] == '\n')
3873 tmp[count - 1] = '\0';
3874
3875 huge = shmem_parse_huge(tmp);
3876 if (huge == -EINVAL)
3877 return -EINVAL;
3878 if (!has_transparent_hugepage() &&
3879 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3880 return -EINVAL;
3881
3882 shmem_huge = huge;
Kirill A. Shutemov5d8b3cc2017-08-25 15:55:33 -07003883 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003884 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3885 return count;
3886}
3887
3888struct kobj_attribute shmem_enabled_attr =
3889 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003890#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003891
Arnd Bergmann3b337192016-08-10 16:27:44 -07003892#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003893bool shmem_huge_enabled(struct vm_area_struct *vma)
3894{
3895 struct inode *inode = file_inode(vma->vm_file);
3896 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3897 loff_t i_size;
3898 pgoff_t off;
3899
3900 if (shmem_huge == SHMEM_HUGE_FORCE)
3901 return true;
3902 if (shmem_huge == SHMEM_HUGE_DENY)
3903 return false;
3904 switch (sbinfo->huge) {
3905 case SHMEM_HUGE_NEVER:
3906 return false;
3907 case SHMEM_HUGE_ALWAYS:
3908 return true;
3909 case SHMEM_HUGE_WITHIN_SIZE:
3910 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3911 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3912 if (i_size >= HPAGE_PMD_SIZE &&
3913 i_size >> PAGE_SHIFT >= off)
3914 return true;
3915 case SHMEM_HUGE_ADVISE:
3916 /* TODO: implement fadvise() hints */
3917 return (vma->vm_flags & VM_HUGEPAGE);
3918 default:
3919 VM_BUG_ON(1);
3920 return false;
3921 }
3922}
Arnd Bergmann3b337192016-08-10 16:27:44 -07003923#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003924
Matt Mackall853ac432009-01-06 14:40:20 -08003925#else /* !CONFIG_SHMEM */
3926
3927/*
3928 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3929 *
3930 * This is intended for small system where the benefits of the full
3931 * shmem code (swap-backed and resource-limited) are outweighed by
3932 * their complexity. On systems without swap this code should be
3933 * effectively equivalent, but much lighter weight.
3934 */
3935
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003936static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003937 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003938 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003939 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003940 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003941};
3942
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003943int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003944{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003945 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003946
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003947 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003948 BUG_ON(IS_ERR(shm_mnt));
3949
3950 return 0;
3951}
3952
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003953int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003954{
3955 return 0;
3956}
3957
Hugh Dickins3f96b792009-09-21 17:03:37 -07003958int shmem_lock(struct file *file, int lock, struct user_struct *user)
3959{
3960 return 0;
3961}
3962
Hugh Dickins24513262012-01-20 14:34:21 -08003963void shmem_unlock_mapping(struct address_space *mapping)
3964{
3965}
3966
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003967#ifdef CONFIG_MMU
3968unsigned long shmem_get_unmapped_area(struct file *file,
3969 unsigned long addr, unsigned long len,
3970 unsigned long pgoff, unsigned long flags)
3971{
3972 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3973}
3974#endif
3975
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003976void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003977{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003978 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003979}
3980EXPORT_SYMBOL_GPL(shmem_truncate_range);
3981
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003982#define shmem_vm_ops generic_file_vm_ops
3983#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003984#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003985#define shmem_acct_size(flags, size) 0
3986#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003987
3988#endif /* CONFIG_SHMEM */
3989
3990/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991
Rasmus Villemoes19938e32016-10-07 17:01:01 -07003992static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003993 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003994};
3995
Eric Parisc7277092013-12-02 11:24:19 +00003996static struct file *__shmem_file_setup(const char *name, loff_t size,
3997 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998{
Al Viro6b4d0b22013-02-14 21:37:26 -05003999 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04004001 struct path path;
Al Viro34515382013-02-14 22:38:02 -05004002 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 struct qstr this;
4004
4005 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05004006 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007
Hugh Dickins285b2c42011-08-03 16:21:20 -07004008 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 return ERR_PTR(-EINVAL);
4010
4011 if (shmem_acct_size(flags, size))
4012 return ERR_PTR(-ENOMEM);
4013
Al Viro6b4d0b22013-02-14 21:37:26 -05004014 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 this.name = name;
4016 this.len = strlen(name);
4017 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004018 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004019 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004020 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004021 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004023 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024
Al Viro6b4d0b22013-02-14 21:37:26 -05004025 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004026 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004027 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004028 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029
Eric Parisc7277092013-12-02 11:24:19 +00004030 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004031 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004033 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004034 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4035 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004036 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004037
Al Viro6b4d0b22013-02-14 21:37:26 -05004038 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004039 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004040 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004041 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004042
Al Viro6b4d0b22013-02-14 21:37:26 -05004043 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045put_memory:
4046 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004047put_path:
4048 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004049 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050}
Eric Parisc7277092013-12-02 11:24:19 +00004051
4052/**
4053 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4054 * kernel internal. There will be NO LSM permission checks against the
4055 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004056 * higher layer. The users are the big_key and shm implementations. LSM
4057 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004058 * @name: name for dentry (to be seen in /proc/<pid>/maps
4059 * @size: size to be set for the file
4060 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4061 */
4062struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4063{
4064 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4065}
4066
4067/**
4068 * shmem_file_setup - get an unlinked file living in tmpfs
4069 * @name: name for dentry (to be seen in /proc/<pid>/maps
4070 * @size: size to be set for the file
4071 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4072 */
4073struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4074{
4075 return __shmem_file_setup(name, size, flags, 0);
4076}
Keith Packard395e0dd2008-06-20 00:08:06 -07004077EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078
John Stultz6f193672011-12-15 18:51:04 -08004079void shmem_set_file(struct vm_area_struct *vma, struct file *file)
4080{
4081 if (vma->vm_file)
4082 fput(vma->vm_file);
4083 vma->vm_file = file;
4084 vma->vm_ops = &shmem_vm_ops;
4085}
4086
Randy Dunlap46711812008-03-19 17:00:41 -07004087/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4090 */
4091int shmem_zero_setup(struct vm_area_struct *vma)
4092{
4093 struct file *file;
4094 loff_t size = vma->vm_end - vma->vm_start;
4095
Hugh Dickins66fc1302015-06-14 09:48:09 -07004096 /*
4097 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4098 * between XFS directory reading and selinux: since this file is only
4099 * accessible to the user through its mapping, use S_PRIVATE flag to
4100 * bypass file security, in the same way as shmem_kernel_file_setup().
4101 */
4102 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103 if (IS_ERR(file))
4104 return PTR_ERR(file);
4105
John Stultz6f193672011-12-15 18:51:04 -08004106 shmem_set_file(vma, file);
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004107
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004108 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004109 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4110 (vma->vm_end & HPAGE_PMD_MASK)) {
4111 khugepaged_enter(vma, vma->vm_flags);
4112 }
4113
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114 return 0;
4115}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004116
4117/**
4118 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4119 * @mapping: the page's address_space
4120 * @index: the page index
4121 * @gfp: the page allocator flags to use if allocating
4122 *
4123 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4124 * with any new page allocations done using the specified allocation flags.
4125 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4126 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4127 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4128 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004129 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4130 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004131 */
4132struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4133 pgoff_t index, gfp_t gfp)
4134{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004135#ifdef CONFIG_SHMEM
4136 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004137 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004138 int error;
4139
4140 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004141 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4142 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004143 if (error)
4144 page = ERR_PTR(error);
4145 else
4146 unlock_page(page);
4147 return page;
4148#else
4149 /*
4150 * The tiny !SHMEM case uses ramfs without swap
4151 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004152 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004153#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004154}
4155EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);