blob: 9d32e1cb9f38f7dd508f1be0273d0721974b2c05 [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
373static int shmem_parse_huge(const char *str)
374{
375 if (!strcmp(str, "never"))
376 return SHMEM_HUGE_NEVER;
377 if (!strcmp(str, "always"))
378 return SHMEM_HUGE_ALWAYS;
379 if (!strcmp(str, "within_size"))
380 return SHMEM_HUGE_WITHIN_SIZE;
381 if (!strcmp(str, "advise"))
382 return SHMEM_HUGE_ADVISE;
383 if (!strcmp(str, "deny"))
384 return SHMEM_HUGE_DENY;
385 if (!strcmp(str, "force"))
386 return SHMEM_HUGE_FORCE;
387 return -EINVAL;
388}
389
390static const char *shmem_format_huge(int huge)
391{
392 switch (huge) {
393 case SHMEM_HUGE_NEVER:
394 return "never";
395 case SHMEM_HUGE_ALWAYS:
396 return "always";
397 case SHMEM_HUGE_WITHIN_SIZE:
398 return "within_size";
399 case SHMEM_HUGE_ADVISE:
400 return "advise";
401 case SHMEM_HUGE_DENY:
402 return "deny";
403 case SHMEM_HUGE_FORCE:
404 return "force";
405 default:
406 VM_BUG_ON(1);
407 return "bad_val";
408 }
409}
410
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700411static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
412 struct shrink_control *sc, unsigned long nr_to_split)
413{
414 LIST_HEAD(list), *pos, *next;
415 struct inode *inode;
416 struct shmem_inode_info *info;
417 struct page *page;
418 unsigned long batch = sc ? sc->nr_to_scan : 128;
419 int removed = 0, split = 0;
420
421 if (list_empty(&sbinfo->shrinklist))
422 return SHRINK_STOP;
423
424 spin_lock(&sbinfo->shrinklist_lock);
425 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
426 info = list_entry(pos, struct shmem_inode_info, shrinklist);
427
428 /* pin the inode */
429 inode = igrab(&info->vfs_inode);
430
431 /* inode is about to be evicted */
432 if (!inode) {
433 list_del_init(&info->shrinklist);
434 removed++;
435 goto next;
436 }
437
438 /* Check if there's anything to gain */
439 if (round_up(inode->i_size, PAGE_SIZE) ==
440 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
441 list_del_init(&info->shrinklist);
442 removed++;
443 iput(inode);
444 goto next;
445 }
446
447 list_move(&info->shrinklist, &list);
448next:
449 if (!--batch)
450 break;
451 }
452 spin_unlock(&sbinfo->shrinklist_lock);
453
454 list_for_each_safe(pos, next, &list) {
455 int ret;
456
457 info = list_entry(pos, struct shmem_inode_info, shrinklist);
458 inode = &info->vfs_inode;
459
460 if (nr_to_split && split >= nr_to_split) {
461 iput(inode);
462 continue;
463 }
464
465 page = find_lock_page(inode->i_mapping,
466 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
467 if (!page)
468 goto drop;
469
470 if (!PageTransHuge(page)) {
471 unlock_page(page);
472 put_page(page);
473 goto drop;
474 }
475
476 ret = split_huge_page(page);
477 unlock_page(page);
478 put_page(page);
479
480 if (ret) {
481 /* split failed: leave it on the list */
482 iput(inode);
483 continue;
484 }
485
486 split++;
487drop:
488 list_del_init(&info->shrinklist);
489 removed++;
490 iput(inode);
491 }
492
493 spin_lock(&sbinfo->shrinklist_lock);
494 list_splice_tail(&list, &sbinfo->shrinklist);
495 sbinfo->shrinklist_len -= removed;
496 spin_unlock(&sbinfo->shrinklist_lock);
497
498 return split;
499}
500
501static long shmem_unused_huge_scan(struct super_block *sb,
502 struct shrink_control *sc)
503{
504 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
505
506 if (!READ_ONCE(sbinfo->shrinklist_len))
507 return SHRINK_STOP;
508
509 return shmem_unused_huge_shrink(sbinfo, sc, 0);
510}
511
512static long shmem_unused_huge_count(struct super_block *sb,
513 struct shrink_control *sc)
514{
515 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
516 return READ_ONCE(sbinfo->shrinklist_len);
517}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700518#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700519
520#define shmem_huge SHMEM_HUGE_DENY
521
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700522static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
523 struct shrink_control *sc, unsigned long nr_to_split)
524{
525 return 0;
526}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700527#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700528
529/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700530 * Like add_to_page_cache_locked, but error if expected item has gone.
531 */
532static int shmem_add_to_page_cache(struct page *page,
533 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700534 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700535{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700536 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700537
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700538 VM_BUG_ON_PAGE(PageTail(page), page);
539 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800540 VM_BUG_ON_PAGE(!PageLocked(page), page);
541 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700542 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700543
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700544 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700545 page->mapping = mapping;
546 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700547
Hugh Dickinsb065b432012-07-11 14:02:48 -0700548 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700549 if (PageTransHuge(page)) {
550 void __rcu **results;
551 pgoff_t idx;
552 int i;
553
554 error = 0;
555 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
556 &results, &idx, index, 1) &&
557 idx < index + HPAGE_PMD_NR) {
558 error = -EEXIST;
559 }
560
561 if (!error) {
562 for (i = 0; i < HPAGE_PMD_NR; i++) {
563 error = radix_tree_insert(&mapping->page_tree,
564 index + i, page + i);
565 VM_BUG_ON(error);
566 }
567 count_vm_event(THP_FILE_ALLOC);
568 }
569 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700570 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700571 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700572 error = shmem_radix_tree_replace(mapping, index, expected,
573 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700574 }
575
Hugh Dickinsb065b432012-07-11 14:02:48 -0700576 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700577 mapping->nrpages += nr;
578 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700579 __inc_node_page_state(page, NR_SHMEM_THPS);
580 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
581 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700582 spin_unlock_irq(&mapping->tree_lock);
583 } else {
584 page->mapping = NULL;
585 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700586 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700587 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700588 return error;
589}
590
591/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700592 * Like delete_from_page_cache, but substitutes swap for page.
593 */
594static void shmem_delete_from_page_cache(struct page *page, void *radswap)
595{
596 struct address_space *mapping = page->mapping;
597 int error;
598
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700599 VM_BUG_ON_PAGE(PageCompound(page), page);
600
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700601 spin_lock_irq(&mapping->tree_lock);
602 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
603 page->mapping = NULL;
604 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700605 __dec_node_page_state(page, NR_FILE_PAGES);
606 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700607 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300608 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700609 BUG_ON(error);
610}
611
612/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700613 * Remove swap entry from radix tree, free the swap and its page cache.
614 */
615static int shmem_free_swap(struct address_space *mapping,
616 pgoff_t index, void *radswap)
617{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700618 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700619
620 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700621 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700622 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700623 if (old != radswap)
624 return -ENOENT;
625 free_swap_and_cache(radix_to_swp_entry(radswap));
626 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700627}
628
629/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800630 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800631 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800632 *
633 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
634 * as long as the inode doesn't go away and racy results are not a problem.
635 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800636unsigned long shmem_partial_swap_usage(struct address_space *mapping,
637 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800638{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800639 struct radix_tree_iter iter;
640 void **slot;
641 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800642 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800643
644 rcu_read_lock();
645
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800646 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
647 if (iter.index >= end)
648 break;
649
650 page = radix_tree_deref_slot(slot);
651
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700652 if (radix_tree_deref_retry(page)) {
653 slot = radix_tree_iter_retry(&iter);
654 continue;
655 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800656
657 if (radix_tree_exceptional_entry(page))
658 swapped++;
659
660 if (need_resched()) {
661 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700662 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800663 }
664 }
665
666 rcu_read_unlock();
667
668 return swapped << PAGE_SHIFT;
669}
670
671/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800672 * Determine (in bytes) how many of the shmem object's pages mapped by the
673 * given vma is swapped out.
674 *
675 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
676 * as long as the inode doesn't go away and racy results are not a problem.
677 */
678unsigned long shmem_swap_usage(struct vm_area_struct *vma)
679{
680 struct inode *inode = file_inode(vma->vm_file);
681 struct shmem_inode_info *info = SHMEM_I(inode);
682 struct address_space *mapping = inode->i_mapping;
683 unsigned long swapped;
684
685 /* Be careful as we don't hold info->lock */
686 swapped = READ_ONCE(info->swapped);
687
688 /*
689 * The easier cases are when the shmem object has nothing in swap, or
690 * the vma maps it whole. Then we can simply use the stats that we
691 * already track.
692 */
693 if (!swapped)
694 return 0;
695
696 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
697 return swapped << PAGE_SHIFT;
698
699 /* Here comes the more involved part */
700 return shmem_partial_swap_usage(mapping,
701 linear_page_index(vma, vma->vm_start),
702 linear_page_index(vma, vma->vm_end));
703}
704
705/*
Hugh Dickins24513262012-01-20 14:34:21 -0800706 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
707 */
708void shmem_unlock_mapping(struct address_space *mapping)
709{
710 struct pagevec pvec;
711 pgoff_t indices[PAGEVEC_SIZE];
712 pgoff_t index = 0;
713
714 pagevec_init(&pvec, 0);
715 /*
716 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
717 */
718 while (!mapping_unevictable(mapping)) {
719 /*
720 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
721 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
722 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700723 pvec.nr = find_get_entries(mapping, index,
724 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800725 if (!pvec.nr)
726 break;
727 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700728 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800729 check_move_unevictable_pages(pvec.pages, pvec.nr);
730 pagevec_release(&pvec);
731 cond_resched();
732 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700733}
734
735/*
736 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700737 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700738 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700739static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
740 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700742 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300744 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
745 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
746 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
747 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700748 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700749 pgoff_t indices[PAGEVEC_SIZE];
750 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700751 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700752 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700754 if (lend == -1)
755 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700756
757 pagevec_init(&pvec, 0);
758 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700759 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700760 pvec.nr = find_get_entries(mapping, index,
761 min(end - index, (pgoff_t)PAGEVEC_SIZE),
762 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700763 if (!pvec.nr)
764 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700765 for (i = 0; i < pagevec_count(&pvec); i++) {
766 struct page *page = pvec.pages[i];
767
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700768 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700769 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700770 break;
771
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700772 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700773 if (unfalloc)
774 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700775 nr_swaps_freed += !shmem_free_swap(mapping,
776 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700777 continue;
778 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700779
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700780 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
781
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700782 if (!trylock_page(page))
783 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700784
785 if (PageTransTail(page)) {
786 /* Middle of THP: zero out the page */
787 clear_highpage(page);
788 unlock_page(page);
789 continue;
790 } else if (PageTransHuge(page)) {
791 if (index == round_down(end, HPAGE_PMD_NR)) {
792 /*
793 * Range ends in the middle of THP:
794 * zero out the page
795 */
796 clear_highpage(page);
797 unlock_page(page);
798 continue;
799 }
800 index += HPAGE_PMD_NR - 1;
801 i += HPAGE_PMD_NR - 1;
802 }
803
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700804 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700805 VM_BUG_ON_PAGE(PageTail(page), page);
806 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800807 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700808 truncate_inode_page(mapping, page);
809 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700810 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700811 unlock_page(page);
812 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700813 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800814 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700815 cond_resched();
816 index++;
817 }
818
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700819 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700820 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700821 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700822 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300823 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700824 if (start > end) {
825 top = partial_end;
826 partial_end = 0;
827 }
828 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700829 set_page_dirty(page);
830 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300831 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700832 }
833 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700834 if (partial_end) {
835 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700836 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700837 if (page) {
838 zero_user_segment(page, 0, partial_end);
839 set_page_dirty(page);
840 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300841 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700842 }
843 }
844 if (start >= end)
845 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700846
847 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700848 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700849 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700850
851 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700852 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700853 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700854 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700855 /* If all gone or hole-punch or unfalloc, we're done */
856 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700857 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700858 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700859 index = start;
860 continue;
861 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700862 for (i = 0; i < pagevec_count(&pvec); i++) {
863 struct page *page = pvec.pages[i];
864
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700865 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700866 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700867 break;
868
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700869 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700870 if (unfalloc)
871 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700872 if (shmem_free_swap(mapping, index, page)) {
873 /* Swap was replaced by page: retry */
874 index--;
875 break;
876 }
877 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700878 continue;
879 }
880
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700881 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700882
883 if (PageTransTail(page)) {
884 /* Middle of THP: zero out the page */
885 clear_highpage(page);
886 unlock_page(page);
887 /*
888 * Partial thp truncate due 'start' in middle
889 * of THP: don't need to look on these pages
890 * again on !pvec.nr restart.
891 */
892 if (index != round_down(end, HPAGE_PMD_NR))
893 start++;
894 continue;
895 } else if (PageTransHuge(page)) {
896 if (index == round_down(end, HPAGE_PMD_NR)) {
897 /*
898 * Range ends in the middle of THP:
899 * zero out the page
900 */
901 clear_highpage(page);
902 unlock_page(page);
903 continue;
904 }
905 index += HPAGE_PMD_NR - 1;
906 i += HPAGE_PMD_NR - 1;
907 }
908
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700909 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700910 VM_BUG_ON_PAGE(PageTail(page), page);
911 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800912 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700913 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700914 } else {
915 /* Page was replaced by swap: retry */
916 unlock_page(page);
917 index--;
918 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700919 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700920 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700921 unlock_page(page);
922 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700923 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800924 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700925 index++;
926 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700927
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700928 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700929 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700931 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700932}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700934void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
935{
936 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700937 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700939EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Yu Zhao44a30222015-09-08 15:03:33 -0700941static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
942 struct kstat *stat)
943{
944 struct inode *inode = dentry->d_inode;
945 struct shmem_inode_info *info = SHMEM_I(inode);
946
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800947 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700948 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800949 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700950 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800951 }
Yu Zhao44a30222015-09-08 15:03:33 -0700952 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700953 return 0;
954}
955
Hugh Dickins94c1e622011-06-27 16:18:03 -0700956static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957{
David Howells75c3cfa2015-03-17 22:26:12 +0000958 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700959 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700960 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 int error;
962
Jan Kara31051c82016-05-26 16:55:18 +0200963 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200964 if (error)
965 return error;
966
Hugh Dickins94c1e622011-06-27 16:18:03 -0700967 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
968 loff_t oldsize = inode->i_size;
969 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000970
David Herrmann40e041a2014-08-08 14:25:27 -0700971 /* protected by i_mutex */
972 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
973 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
974 return -EPERM;
975
Hugh Dickins94c1e622011-06-27 16:18:03 -0700976 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700977 error = shmem_reacct_size(SHMEM_I(inode)->flags,
978 oldsize, newsize);
979 if (error)
980 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700981 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700982 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700983 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700984 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700985 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800986 if (oldsize > holebegin)
987 unmap_mapping_range(inode->i_mapping,
988 holebegin, 0, 1);
989 if (info->alloced)
990 shmem_truncate_range(inode,
991 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700992 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800993 if (oldsize > holebegin)
994 unmap_mapping_range(inode->i_mapping,
995 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700996
997 /*
998 * Part of the huge page can be beyond i_size: subject
999 * to shrink under memory pressure.
1000 */
1001 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1002 spin_lock(&sbinfo->shrinklist_lock);
1003 if (list_empty(&info->shrinklist)) {
1004 list_add_tail(&info->shrinklist,
1005 &sbinfo->shrinklist);
1006 sbinfo->shrinklist_len++;
1007 }
1008 spin_unlock(&sbinfo->shrinklist_lock);
1009 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001010 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
1012
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001013 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001014 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001015 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 return error;
1017}
1018
Al Viro1f895f72010-06-05 19:10:41 -04001019static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001022 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001024 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 shmem_unacct_size(info->flags, inode->i_size);
1026 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001027 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001028 if (!list_empty(&info->shrinklist)) {
1029 spin_lock(&sbinfo->shrinklist_lock);
1030 if (!list_empty(&info->shrinklist)) {
1031 list_del_init(&info->shrinklist);
1032 sbinfo->shrinklist_len--;
1033 }
1034 spin_unlock(&sbinfo->shrinklist_lock);
1035 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001037 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001039 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
Al Viro3ed47db2016-01-22 18:08:52 -05001041 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001042
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001043 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001044 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001045 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001046 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047}
1048
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001049/*
1050 * If swap found in inode, free it and move page from swapcache to filecache.
1051 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001052static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001053 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001055 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001056 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001057 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001058 gfp_t gfp;
1059 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001061 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001062 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001063 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001064 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001065
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001066 /*
1067 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001068 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001069 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001070 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001071 */
1072 if (shmem_swaplist.next != &info->swaplist)
1073 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001074
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001075 gfp = mapping_gfp_mask(mapping);
1076 if (shmem_should_replace_page(*pagep, gfp)) {
1077 mutex_unlock(&shmem_swaplist_mutex);
1078 error = shmem_replace_page(pagep, gfp, info, index);
1079 mutex_lock(&shmem_swaplist_mutex);
1080 /*
1081 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001082 * allocation, but the inode might have been freed while we
1083 * dropped it: although a racing shmem_evict_inode() cannot
1084 * complete without emptying the radix_tree, our page lock
1085 * on this swapcache page is not enough to prevent that -
1086 * free_swap_and_cache() of our swap entry will only
1087 * trylock_page(), removing swap from radix_tree whatever.
1088 *
1089 * We must not proceed to shmem_add_to_page_cache() if the
1090 * inode has been freed, but of course we cannot rely on
1091 * inode or mapping or info to check that. However, we can
1092 * safely check if our swap entry is still in use (and here
1093 * it can't have got reused for another page): if it's still
1094 * in use, then the inode cannot have been freed yet, and we
1095 * can safely proceed (if it's no longer in use, that tells
1096 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001097 */
1098 if (!page_swapcount(*pagep))
1099 error = -ENOENT;
1100 }
1101
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001102 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001103 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1104 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1105 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001106 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001107 if (!error)
1108 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001109 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001110 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001111 /*
1112 * Truncation and eviction use free_swap_and_cache(), which
1113 * only does trylock page: if we raced, best clean up here.
1114 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001115 delete_from_swap_cache(*pagep);
1116 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001117 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001118 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001119 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001120 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001121 swap_free(swap);
1122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001124 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
1127/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001128 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001130int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001132 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001134 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001135 int error = 0;
1136
1137 /*
1138 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001139 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001140 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001141 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001142 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001143
1144 /*
1145 * Charge page using GFP_KERNEL while we can wait, before taking
1146 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1147 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001148 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001149 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1150 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001151 if (error)
1152 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001153 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001154 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001156 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001157 list_for_each_safe(this, next, &shmem_swaplist) {
1158 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001159 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001160 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001161 else
1162 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001163 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001164 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001165 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001166 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001168 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001169
Johannes Weiner00501b52014-08-08 14:19:20 -07001170 if (error) {
1171 if (error != -ENOMEM)
1172 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001173 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001174 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001175 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001176out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001177 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001178 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001179 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182/*
1183 * Move the page from the page cache to the swap cache.
1184 */
1185static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1186{
1187 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001190 swp_entry_t swap;
1191 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001193 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 mapping = page->mapping;
1196 index = page->index;
1197 inode = mapping->host;
1198 info = SHMEM_I(inode);
1199 if (info->flags & VM_LOCKED)
1200 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001201 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 goto redirty;
1203
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001204 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001205 * Our capabilities prevent regular writeback or sync from ever calling
1206 * shmem_writepage; but a stacking filesystem might use ->writepage of
1207 * its underlying filesystem, in which case tmpfs should write out to
1208 * swap only in response to memory pressure, and not for the writeback
1209 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001210 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001211 if (!wbc->for_reclaim) {
1212 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1213 goto redirty;
1214 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001215
1216 /*
1217 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1218 * value into swapfile.c, the only way we can correctly account for a
1219 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001220 *
1221 * That's okay for a page already fallocated earlier, but if we have
1222 * not yet completed the fallocation, then (a) we want to keep track
1223 * of this page in case we have to undo it, and (b) it may not be a
1224 * good idea to continue anyway, once we're pushing into swap. So
1225 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001226 */
1227 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001228 if (inode->i_private) {
1229 struct shmem_falloc *shmem_falloc;
1230 spin_lock(&inode->i_lock);
1231 shmem_falloc = inode->i_private;
1232 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001233 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001234 index >= shmem_falloc->start &&
1235 index < shmem_falloc->next)
1236 shmem_falloc->nr_unswapped++;
1237 else
1238 shmem_falloc = NULL;
1239 spin_unlock(&inode->i_lock);
1240 if (shmem_falloc)
1241 goto redirty;
1242 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001243 clear_highpage(page);
1244 flush_dcache_page(page);
1245 SetPageUptodate(page);
1246 }
1247
Hugh Dickins48f170f2011-07-25 17:12:37 -07001248 swap = get_swap_page();
1249 if (!swap.val)
1250 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001251
Vladimir Davydov37e84352016-01-20 15:02:56 -08001252 if (mem_cgroup_try_charge_swap(page, swap))
1253 goto free_swap;
1254
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001255 /*
1256 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001257 * if it's not already there. Do it now before the page is
1258 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001259 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001260 * we've incremented swapped, because shmem_unuse_inode() will
1261 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001262 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001263 mutex_lock(&shmem_swaplist_mutex);
1264 if (list_empty(&info->swaplist))
1265 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001266
Hugh Dickins48f170f2011-07-25 17:12:37 -07001267 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001268 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001269 shmem_recalc_inode(inode);
1270 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001271 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001272
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001273 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001274 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1275
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001276 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001277 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001278 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 return 0;
1280 }
1281
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001282 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001283free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001284 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285redirty:
1286 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001287 if (wbc->for_reclaim)
1288 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1289 unlock_page(page);
1290 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291}
1292
Hugh Dickins75edd342016-05-19 17:12:44 -07001293#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001294static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001295{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001296 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001297
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001298 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001299 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001300
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001301 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001302
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001303 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001304}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001305
1306static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1307{
1308 struct mempolicy *mpol = NULL;
1309 if (sbinfo->mpol) {
1310 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1311 mpol = sbinfo->mpol;
1312 mpol_get(mpol);
1313 spin_unlock(&sbinfo->stat_lock);
1314 }
1315 return mpol;
1316}
Hugh Dickins75edd342016-05-19 17:12:44 -07001317#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1318static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1319{
1320}
1321static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1322{
1323 return NULL;
1324}
1325#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1326#ifndef CONFIG_NUMA
1327#define vm_policy vm_private_data
1328#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001329
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001330static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1331 struct shmem_inode_info *info, pgoff_t index)
1332{
1333 /* Create a pseudo vma that just contains the policy */
1334 vma->vm_start = 0;
1335 /* Bias interleave by inode number to distribute better across nodes */
1336 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1337 vma->vm_ops = NULL;
1338 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1339}
1340
1341static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1342{
1343 /* Drop reference taken by mpol_shared_policy_lookup() */
1344 mpol_cond_put(vma->vm_policy);
1345}
1346
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001347static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1348 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001351 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001353 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001354 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001355 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001356
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001357 return page;
1358}
Mel Gorman18a2f372012-12-05 14:01:41 -08001359
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001360static struct page *shmem_alloc_hugepage(gfp_t gfp,
1361 struct shmem_inode_info *info, pgoff_t index)
1362{
1363 struct vm_area_struct pvma;
1364 struct inode *inode = &info->vfs_inode;
1365 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001366 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001367 void __rcu **results;
1368 struct page *page;
1369
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001370 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001371 return NULL;
1372
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001373 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001374 rcu_read_lock();
1375 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1376 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1377 rcu_read_unlock();
1378 return NULL;
1379 }
1380 rcu_read_unlock();
1381
1382 shmem_pseudo_vma_init(&pvma, info, hindex);
1383 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1384 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1385 shmem_pseudo_vma_destroy(&pvma);
1386 if (page)
1387 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001388 return page;
1389}
1390
1391static struct page *shmem_alloc_page(gfp_t gfp,
1392 struct shmem_inode_info *info, pgoff_t index)
1393{
1394 struct vm_area_struct pvma;
1395 struct page *page;
1396
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001397 shmem_pseudo_vma_init(&pvma, info, index);
1398 page = alloc_page_vma(gfp, &pvma, 0);
1399 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001400
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001401 return page;
1402}
1403
1404static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1405 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1406 pgoff_t index, bool huge)
1407{
1408 struct page *page;
1409 int nr;
1410 int err = -ENOSPC;
1411
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001412 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001413 huge = false;
1414 nr = huge ? HPAGE_PMD_NR : 1;
1415
1416 if (shmem_acct_block(info->flags, nr))
1417 goto failed;
1418 if (sbinfo->max_blocks) {
1419 if (percpu_counter_compare(&sbinfo->used_blocks,
1420 sbinfo->max_blocks - nr) > 0)
1421 goto unacct;
1422 percpu_counter_add(&sbinfo->used_blocks, nr);
1423 }
1424
1425 if (huge)
1426 page = shmem_alloc_hugepage(gfp, info, index);
1427 else
1428 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001429 if (page) {
1430 __SetPageLocked(page);
1431 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001432 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001433 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001434
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001435 err = -ENOMEM;
1436 if (sbinfo->max_blocks)
1437 percpu_counter_add(&sbinfo->used_blocks, -nr);
1438unacct:
1439 shmem_unacct_blocks(info->flags, nr);
1440failed:
1441 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001445 * When a page is moved from swapcache to shmem filecache (either by the
1446 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1447 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1448 * ignorance of the mapping it belongs to. If that mapping has special
1449 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1450 * we may need to copy to a suitable page before moving to filecache.
1451 *
1452 * In a future release, this may well be extended to respect cpuset and
1453 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1454 * but for now it is a simple matter of zone.
1455 */
1456static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1457{
1458 return page_zonenum(page) > gfp_zone(gfp);
1459}
1460
1461static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1462 struct shmem_inode_info *info, pgoff_t index)
1463{
1464 struct page *oldpage, *newpage;
1465 struct address_space *swap_mapping;
1466 pgoff_t swap_index;
1467 int error;
1468
1469 oldpage = *pagep;
1470 swap_index = page_private(oldpage);
1471 swap_mapping = page_mapping(oldpage);
1472
1473 /*
1474 * We have arrived here because our zones are constrained, so don't
1475 * limit chance of success by further cpuset and node constraints.
1476 */
1477 gfp &= ~GFP_CONSTRAINT_MASK;
1478 newpage = shmem_alloc_page(gfp, info, index);
1479 if (!newpage)
1480 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001481
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001482 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001483 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001484 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001485
Hugh Dickins9956edf2016-11-10 10:46:11 -08001486 __SetPageLocked(newpage);
1487 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001488 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001489 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001490 SetPageSwapCache(newpage);
1491
1492 /*
1493 * Our caller will very soon move newpage out of swapcache, but it's
1494 * a nice clean interface for us to replace oldpage by newpage there.
1495 */
1496 spin_lock_irq(&swap_mapping->tree_lock);
1497 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1498 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001499 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001500 __inc_node_page_state(newpage, NR_FILE_PAGES);
1501 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001502 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001503 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001504
Hugh Dickins0142ef62012-06-07 14:21:09 -07001505 if (unlikely(error)) {
1506 /*
1507 * Is this possible? I think not, now that our callers check
1508 * both PageSwapCache and page_private after getting page lock;
1509 * but be defensive. Reverse old to newpage for clear and free.
1510 */
1511 oldpage = newpage;
1512 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001513 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001514 lru_cache_add_anon(newpage);
1515 *pagep = newpage;
1516 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001517
1518 ClearPageSwapCache(oldpage);
1519 set_page_private(oldpage, 0);
1520
1521 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001522 put_page(oldpage);
1523 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001524 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001525}
1526
1527/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001528 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
1530 * If we allocate a new one we do not mark it dirty. That's up to the
1531 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001532 * entry since a page cannot live in both the swap and page cache.
1533 *
1534 * fault_mm and fault_type are only supplied by shmem_fault:
1535 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001537static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001538 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1539 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
1541 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001542 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001544 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001545 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001546 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001548 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001549 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001551 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001552 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001554 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001556 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1557 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001559 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001560 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001561 if (radix_tree_exceptional_entry(page)) {
1562 swap = radix_to_swp_entry(page);
1563 page = NULL;
1564 }
1565
Hugh Dickins75edd342016-05-19 17:12:44 -07001566 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001567 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001568 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001569 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001570 }
1571
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001572 if (page && sgp == SGP_WRITE)
1573 mark_page_accessed(page);
1574
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001575 /* fallocated page? */
1576 if (page && !PageUptodate(page)) {
1577 if (sgp != SGP_READ)
1578 goto clear;
1579 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001580 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001581 page = NULL;
1582 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001583 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001584 *pagep = page;
1585 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001589 * Fast cache lookup did not find it:
1590 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001592 info = SHMEM_I(inode);
1593 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001594 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 if (swap.val) {
1597 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001598 page = lookup_swap_cache(swap);
1599 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001600 /* Or update major stats only when swapin succeeds?? */
1601 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001602 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001603 count_vm_event(PGMAJFAULT);
1604 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1605 }
1606 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001607 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001608 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001609 error = -ENOMEM;
1610 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 }
1613
1614 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001615 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001616 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001617 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001618 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001619 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001620 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001621 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001623 goto failed;
1624 }
1625 wait_on_page_writeback(page);
1626
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001627 if (shmem_should_replace_page(page, gfp)) {
1628 error = shmem_replace_page(&page, gfp, info, index);
1629 if (error)
1630 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 }
1632
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001633 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001634 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001635 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001636 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001637 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001638 /*
1639 * We already confirmed swap under page lock, and make
1640 * no memory allocation here, so usually no possibility
1641 * of error; but free_swap_and_cache() only trylocks a
1642 * page, so it is just possible that the entry has been
1643 * truncated or holepunched since swap was confirmed.
1644 * shmem_undo_range() will have done some of the
1645 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001646 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001647 * Reset swap.val? No, leave it so "failed" goes back to
1648 * "repeat": reading a hole and writing should succeed.
1649 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001650 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001651 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001652 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001653 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001654 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001655 if (error)
1656 goto failed;
1657
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001658 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001659
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001660 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001661 info->swapped--;
1662 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001663 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001664
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001665 if (sgp == SGP_WRITE)
1666 mark_page_accessed(page);
1667
Hugh Dickins27ab7002011-07-25 17:12:36 -07001668 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001669 set_page_dirty(page);
1670 swap_free(swap);
1671
Hugh Dickins54af60422011-08-03 16:21:24 -07001672 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001673 /* shmem_symlink() */
1674 if (mapping->a_ops != &shmem_aops)
1675 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001676 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001677 goto alloc_nohuge;
1678 if (shmem_huge == SHMEM_HUGE_FORCE)
1679 goto alloc_huge;
1680 switch (sbinfo->huge) {
1681 loff_t i_size;
1682 pgoff_t off;
1683 case SHMEM_HUGE_NEVER:
1684 goto alloc_nohuge;
1685 case SHMEM_HUGE_WITHIN_SIZE:
1686 off = round_up(index, HPAGE_PMD_NR);
1687 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1688 if (i_size >= HPAGE_PMD_SIZE &&
1689 i_size >> PAGE_SHIFT >= off)
1690 goto alloc_huge;
1691 /* fallthrough */
1692 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001693 if (sgp_huge == SGP_HUGE)
1694 goto alloc_huge;
1695 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001696 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001699alloc_huge:
1700 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1701 index, true);
1702 if (IS_ERR(page)) {
1703alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1704 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001705 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001706 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001707 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001708 error = PTR_ERR(page);
1709 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001710 if (error != -ENOSPC)
1711 goto failed;
1712 /*
1713 * Try to reclaim some spece by splitting a huge page
1714 * beyond i_size on the filesystem.
1715 */
1716 while (retry--) {
1717 int ret;
1718 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1719 if (ret == SHRINK_STOP)
1720 break;
1721 if (ret)
1722 goto alloc_nohuge;
1723 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001724 goto failed;
1725 }
1726
1727 if (PageTransHuge(page))
1728 hindex = round_down(index, HPAGE_PMD_NR);
1729 else
1730 hindex = index;
1731
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001732 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001733 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001734
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001735 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001736 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001737 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001738 goto unacct;
1739 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1740 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001741 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001742 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001743 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001744 radix_tree_preload_end();
1745 }
1746 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001747 mem_cgroup_cancel_charge(page, memcg,
1748 PageTransHuge(page));
1749 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001750 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001751 mem_cgroup_commit_charge(page, memcg, false,
1752 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001753 lru_cache_add_anon(page);
1754
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001755 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001756 info->alloced += 1 << compound_order(page);
1757 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001758 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001759 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001760 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001761
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001762 if (PageTransHuge(page) &&
1763 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1764 hindex + HPAGE_PMD_NR - 1) {
1765 /*
1766 * Part of the huge page is beyond i_size: subject
1767 * to shrink under memory pressure.
1768 */
1769 spin_lock(&sbinfo->shrinklist_lock);
1770 if (list_empty(&info->shrinklist)) {
1771 list_add_tail(&info->shrinklist,
1772 &sbinfo->shrinklist);
1773 sbinfo->shrinklist_len++;
1774 }
1775 spin_unlock(&sbinfo->shrinklist_lock);
1776 }
1777
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001778 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001779 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1780 */
1781 if (sgp == SGP_FALLOC)
1782 sgp = SGP_WRITE;
1783clear:
1784 /*
1785 * Let SGP_WRITE caller clear ends if write does not fill page;
1786 * but SGP_FALLOC on a page fallocated earlier must initialize
1787 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001788 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001789 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1790 struct page *head = compound_head(page);
1791 int i;
1792
1793 for (i = 0; i < (1 << compound_order(head)); i++) {
1794 clear_highpage(head + i);
1795 flush_dcache_page(head + i);
1796 }
1797 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001798 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001799 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001800
Hugh Dickins54af60422011-08-03 16:21:24 -07001801 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001802 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001803 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001804 if (alloced) {
1805 ClearPageDirty(page);
1806 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001807 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001808 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001809 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001810 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001811 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001812 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001813 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001814 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001815 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001816
Nick Piggind0217ac2007-07-19 01:47:03 -07001817 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001818 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001820unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001821 if (sbinfo->max_blocks)
1822 percpu_counter_sub(&sbinfo->used_blocks,
1823 1 << compound_order(page));
1824 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1825
1826 if (PageTransHuge(page)) {
1827 unlock_page(page);
1828 put_page(page);
1829 goto alloc_nohuge;
1830 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001831failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001832 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001833 error = -EEXIST;
1834unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001835 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001836 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001837 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001838 }
1839 if (error == -ENOSPC && !once++) {
1840 info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001841 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001842 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001843 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001844 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001845 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001846 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001847 goto repeat;
1848 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849}
1850
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001851/*
1852 * This is like autoremove_wake_function, but it removes the wait queue
1853 * entry unconditionally - even if something else had already woken the
1854 * target.
1855 */
1856static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1857{
1858 int ret = default_wake_function(wait, mode, sync, key);
1859 list_del_init(&wait->task_list);
1860 return ret;
1861}
1862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1864{
Al Viro496ad9a2013-01-23 17:07:38 -05001865 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001866 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001867 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001869 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001871 /*
1872 * Trinity finds that probing a hole which tmpfs is punching can
1873 * prevent the hole-punch from ever completing: which in turn
1874 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001875 * faulting pages into the hole while it's being punched. Although
1876 * shmem_undo_range() does remove the additions, it may be unable to
1877 * keep up, as each new page needs its own unmap_mapping_range() call,
1878 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1879 *
1880 * It does not matter if we sometimes reach this check just before the
1881 * hole-punch begins, so that one fault then races with the punch:
1882 * we just need to make racing faults a rare case.
1883 *
1884 * The implementation below would be much simpler if we just used a
1885 * standard mutex or completion: but we cannot take i_mutex in fault,
1886 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001887 */
1888 if (unlikely(inode->i_private)) {
1889 struct shmem_falloc *shmem_falloc;
1890
1891 spin_lock(&inode->i_lock);
1892 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001893 if (shmem_falloc &&
1894 shmem_falloc->waitq &&
1895 vmf->pgoff >= shmem_falloc->start &&
1896 vmf->pgoff < shmem_falloc->next) {
1897 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001898 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001899
1900 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001901 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1902 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001903 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001904 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001905 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001906 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001907
1908 shmem_falloc_waitq = shmem_falloc->waitq;
1909 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1910 TASK_UNINTERRUPTIBLE);
1911 spin_unlock(&inode->i_lock);
1912 schedule();
1913
1914 /*
1915 * shmem_falloc_waitq points into the shmem_fallocate()
1916 * stack of the hole-punching task: shmem_falloc_waitq
1917 * is usually invalid by the time we reach here, but
1918 * finish_wait() does not dereference it in that case;
1919 * though i_lock needed lest racing with wake_up_all().
1920 */
1921 spin_lock(&inode->i_lock);
1922 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1923 spin_unlock(&inode->i_lock);
1924 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001925 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001926 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001927 }
1928
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001929 sgp = SGP_CACHE;
1930 if (vma->vm_flags & VM_HUGEPAGE)
1931 sgp = SGP_HUGE;
1932 else if (vma->vm_flags & VM_NOHUGEPAGE)
1933 sgp = SGP_NOHUGE;
1934
1935 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001936 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 if (error)
1938 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001939 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
1941
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001942unsigned long shmem_get_unmapped_area(struct file *file,
1943 unsigned long uaddr, unsigned long len,
1944 unsigned long pgoff, unsigned long flags)
1945{
1946 unsigned long (*get_area)(struct file *,
1947 unsigned long, unsigned long, unsigned long, unsigned long);
1948 unsigned long addr;
1949 unsigned long offset;
1950 unsigned long inflated_len;
1951 unsigned long inflated_addr;
1952 unsigned long inflated_offset;
1953
1954 if (len > TASK_SIZE)
1955 return -ENOMEM;
1956
1957 get_area = current->mm->get_unmapped_area;
1958 addr = get_area(file, uaddr, len, pgoff, flags);
1959
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001960 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001961 return addr;
1962 if (IS_ERR_VALUE(addr))
1963 return addr;
1964 if (addr & ~PAGE_MASK)
1965 return addr;
1966 if (addr > TASK_SIZE - len)
1967 return addr;
1968
1969 if (shmem_huge == SHMEM_HUGE_DENY)
1970 return addr;
1971 if (len < HPAGE_PMD_SIZE)
1972 return addr;
1973 if (flags & MAP_FIXED)
1974 return addr;
1975 /*
1976 * Our priority is to support MAP_SHARED mapped hugely;
1977 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1978 * But if caller specified an address hint, respect that as before.
1979 */
1980 if (uaddr)
1981 return addr;
1982
1983 if (shmem_huge != SHMEM_HUGE_FORCE) {
1984 struct super_block *sb;
1985
1986 if (file) {
1987 VM_BUG_ON(file->f_op != &shmem_file_operations);
1988 sb = file_inode(file)->i_sb;
1989 } else {
1990 /*
1991 * Called directly from mm/mmap.c, or drivers/char/mem.c
1992 * for "/dev/zero", to create a shared anonymous object.
1993 */
1994 if (IS_ERR(shm_mnt))
1995 return addr;
1996 sb = shm_mnt->mnt_sb;
1997 }
Toshi Kani3089bf62016-09-23 20:21:56 -07001998 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001999 return addr;
2000 }
2001
2002 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2003 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2004 return addr;
2005 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2006 return addr;
2007
2008 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2009 if (inflated_len > TASK_SIZE)
2010 return addr;
2011 if (inflated_len < len)
2012 return addr;
2013
2014 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2015 if (IS_ERR_VALUE(inflated_addr))
2016 return addr;
2017 if (inflated_addr & ~PAGE_MASK)
2018 return addr;
2019
2020 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2021 inflated_addr += offset - inflated_offset;
2022 if (inflated_offset > offset)
2023 inflated_addr += HPAGE_PMD_SIZE;
2024
2025 if (inflated_addr > TASK_SIZE - len)
2026 return addr;
2027 return inflated_addr;
2028}
2029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002031static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032{
Al Viro496ad9a2013-01-23 17:07:38 -05002033 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002034 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035}
2036
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002037static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2038 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039{
Al Viro496ad9a2013-01-23 17:07:38 -05002040 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002041 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002043 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2044 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045}
2046#endif
2047
2048int shmem_lock(struct file *file, int lock, struct user_struct *user)
2049{
Al Viro496ad9a2013-01-23 17:07:38 -05002050 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 struct shmem_inode_info *info = SHMEM_I(inode);
2052 int retval = -ENOMEM;
2053
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002054 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 if (lock && !(info->flags & VM_LOCKED)) {
2056 if (!user_shm_lock(inode->i_size, user))
2057 goto out_nomem;
2058 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002059 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 }
2061 if (!lock && (info->flags & VM_LOCKED) && user) {
2062 user_shm_unlock(inode->i_size, user);
2063 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002064 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
2066 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002067
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002069 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 return retval;
2071}
2072
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002073static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074{
2075 file_accessed(file);
2076 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002077 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002078 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2079 (vma->vm_end & HPAGE_PMD_MASK)) {
2080 khugepaged_enter(vma, vma->vm_flags);
2081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 return 0;
2083}
2084
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002085static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002086 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087{
2088 struct inode *inode;
2089 struct shmem_inode_info *info;
2090 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2091
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002092 if (shmem_reserve_inode(sb))
2093 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
2095 inode = new_inode(sb);
2096 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002097 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002098 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002100 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002101 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 info = SHMEM_I(inode);
2103 memset(info, 0, (char *)inode - (char *)info);
2104 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002105 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002106 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002107 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002109 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002110 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
2112 switch (mode & S_IFMT) {
2113 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002114 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 init_special_inode(inode, mode, dev);
2116 break;
2117 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002118 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 inode->i_op = &shmem_inode_operations;
2120 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002121 mpol_shared_policy_init(&info->policy,
2122 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 break;
2124 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002125 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 /* Some things misbehave if size == 0 on a directory */
2127 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2128 inode->i_op = &shmem_dir_inode_operations;
2129 inode->i_fop = &simple_dir_operations;
2130 break;
2131 case S_IFLNK:
2132 /*
2133 * Must not load anything in the rbtree,
2134 * mpol_free_shared_policy will not be called.
2135 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002136 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 break;
2138 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002139 } else
2140 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 return inode;
2142}
2143
Johannes Weiner0cd61442014-04-03 14:47:46 -07002144bool shmem_mapping(struct address_space *mapping)
2145{
Sasha Levinf0774d82015-02-23 05:38:00 -05002146 if (!mapping->host)
2147 return false;
2148
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002149 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002150}
2151
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002153static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002154static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002156#ifdef CONFIG_TMPFS_XATTR
2157static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2158#else
2159#define shmem_initxattrs NULL
2160#endif
2161
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002163shmem_write_begin(struct file *file, struct address_space *mapping,
2164 loff_t pos, unsigned len, unsigned flags,
2165 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166{
Nick Piggin800d15a2007-10-16 01:25:03 -07002167 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002168 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002169 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002170
2171 /* i_mutex is held by caller */
2172 if (unlikely(info->seals)) {
2173 if (info->seals & F_SEAL_WRITE)
2174 return -EPERM;
2175 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2176 return -EPERM;
2177 }
2178
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002179 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002180}
2181
2182static int
2183shmem_write_end(struct file *file, struct address_space *mapping,
2184 loff_t pos, unsigned len, unsigned copied,
2185 struct page *page, void *fsdata)
2186{
2187 struct inode *inode = mapping->host;
2188
Hugh Dickinsd3602442008-02-04 22:28:44 -08002189 if (pos + copied > inode->i_size)
2190 i_size_write(inode, pos + copied);
2191
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002192 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002193 struct page *head = compound_head(page);
2194 if (PageTransCompound(page)) {
2195 int i;
2196
2197 for (i = 0; i < HPAGE_PMD_NR; i++) {
2198 if (head + i == page)
2199 continue;
2200 clear_highpage(head + i);
2201 flush_dcache_page(head + i);
2202 }
2203 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002204 if (copied < PAGE_SIZE) {
2205 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002206 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002207 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002208 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002209 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002210 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002211 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002212 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002213 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002214
Nick Piggin800d15a2007-10-16 01:25:03 -07002215 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216}
2217
Al Viro2ba5bbe2014-04-02 20:00:02 -04002218static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
Al Viro6e58e792014-02-03 17:07:03 -05002220 struct file *file = iocb->ki_filp;
2221 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002223 pgoff_t index;
2224 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002225 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002226 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002227 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002228 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002229
2230 /*
2231 * Might this read be for a stacking filesystem? Then when reading
2232 * holes of a sparse file, we actually need to allocate those pages,
2233 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2234 */
Al Viro777eda22014-12-17 04:46:46 -05002235 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002236 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002238 index = *ppos >> PAGE_SHIFT;
2239 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
2241 for (;;) {
2242 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002243 pgoff_t end_index;
2244 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 loff_t i_size = i_size_read(inode);
2246
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002247 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 if (index > end_index)
2249 break;
2250 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002251 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 if (nr <= offset)
2253 break;
2254 }
2255
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002256 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002257 if (error) {
2258 if (error == -EINVAL)
2259 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 break;
2261 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002262 if (page) {
2263 if (sgp == SGP_CACHE)
2264 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002265 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
2268 /*
2269 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002270 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002272 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002274 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002276 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 if (nr <= offset) {
2278 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002279 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 break;
2281 }
2282 }
2283 nr -= offset;
2284
2285 if (page) {
2286 /*
2287 * If users can be writing to this page using arbitrary
2288 * virtual addresses, take care about potential aliasing
2289 * before reading the page on the kernel side.
2290 */
2291 if (mapping_writably_mapped(mapping))
2292 flush_dcache_page(page);
2293 /*
2294 * Mark the page accessed if we read the beginning.
2295 */
2296 if (!offset)
2297 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002298 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002300 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
2303 /*
2304 * Ok, we have the page, and it's up-to-date, so
2305 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002307 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002308 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002310 index += offset >> PAGE_SHIFT;
2311 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002313 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002314 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 break;
Al Viro6e58e792014-02-03 17:07:03 -05002316 if (ret < nr) {
2317 error = -EFAULT;
2318 break;
2319 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 cond_resched();
2321 }
2322
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002323 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002324 file_accessed(file);
2325 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326}
2327
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002328/*
2329 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2330 */
2331static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002332 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002333{
2334 struct page *page;
2335 struct pagevec pvec;
2336 pgoff_t indices[PAGEVEC_SIZE];
2337 bool done = false;
2338 int i;
2339
2340 pagevec_init(&pvec, 0);
2341 pvec.nr = 1; /* start small: we may be there already */
2342 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002343 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002344 pvec.nr, pvec.pages, indices);
2345 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002346 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002347 index = end;
2348 break;
2349 }
2350 for (i = 0; i < pvec.nr; i++, index++) {
2351 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002352 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002353 done = true;
2354 break;
2355 }
2356 index = indices[i];
2357 }
2358 page = pvec.pages[i];
2359 if (page && !radix_tree_exceptional_entry(page)) {
2360 if (!PageUptodate(page))
2361 page = NULL;
2362 }
2363 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002364 (page && whence == SEEK_DATA) ||
2365 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002366 done = true;
2367 break;
2368 }
2369 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002370 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002371 pagevec_release(&pvec);
2372 pvec.nr = PAGEVEC_SIZE;
2373 cond_resched();
2374 }
2375 return index;
2376}
2377
Andrew Morton965c8e52012-12-17 15:59:39 -08002378static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002379{
2380 struct address_space *mapping = file->f_mapping;
2381 struct inode *inode = mapping->host;
2382 pgoff_t start, end;
2383 loff_t new_offset;
2384
Andrew Morton965c8e52012-12-17 15:59:39 -08002385 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2386 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002387 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002388 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002389 /* We're holding i_mutex so we can access i_size directly */
2390
2391 if (offset < 0)
2392 offset = -EINVAL;
2393 else if (offset >= inode->i_size)
2394 offset = -ENXIO;
2395 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002396 start = offset >> PAGE_SHIFT;
2397 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002398 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002399 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002400 if (new_offset > offset) {
2401 if (new_offset < inode->i_size)
2402 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002403 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002404 offset = -ENXIO;
2405 else
2406 offset = inode->i_size;
2407 }
2408 }
2409
Hugh Dickins387aae62013-08-04 11:30:25 -07002410 if (offset >= 0)
2411 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002412 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002413 return offset;
2414}
2415
David Herrmann05f65b52014-08-08 14:25:36 -07002416/*
2417 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2418 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2419 */
2420#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2421#define LAST_SCAN 4 /* about 150ms max */
2422
2423static void shmem_tag_pins(struct address_space *mapping)
2424{
2425 struct radix_tree_iter iter;
2426 void **slot;
2427 pgoff_t start;
2428 struct page *page;
2429
2430 lru_add_drain();
2431 start = 0;
2432 rcu_read_lock();
2433
David Herrmann05f65b52014-08-08 14:25:36 -07002434 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2435 page = radix_tree_deref_slot(slot);
2436 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002437 if (radix_tree_deref_retry(page)) {
2438 slot = radix_tree_iter_retry(&iter);
2439 continue;
2440 }
David Herrmann05f65b52014-08-08 14:25:36 -07002441 } else if (page_count(page) - page_mapcount(page) > 1) {
2442 spin_lock_irq(&mapping->tree_lock);
2443 radix_tree_tag_set(&mapping->page_tree, iter.index,
2444 SHMEM_TAG_PINNED);
2445 spin_unlock_irq(&mapping->tree_lock);
2446 }
2447
2448 if (need_resched()) {
2449 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002450 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002451 }
2452 }
2453 rcu_read_unlock();
2454}
2455
2456/*
2457 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2458 * via get_user_pages(), drivers might have some pending I/O without any active
2459 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2460 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2461 * them to be dropped.
2462 * The caller must guarantee that no new user will acquire writable references
2463 * to those pages to avoid races.
2464 */
David Herrmann40e041a2014-08-08 14:25:27 -07002465static int shmem_wait_for_pins(struct address_space *mapping)
2466{
David Herrmann05f65b52014-08-08 14:25:36 -07002467 struct radix_tree_iter iter;
2468 void **slot;
2469 pgoff_t start;
2470 struct page *page;
2471 int error, scan;
2472
2473 shmem_tag_pins(mapping);
2474
2475 error = 0;
2476 for (scan = 0; scan <= LAST_SCAN; scan++) {
2477 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2478 break;
2479
2480 if (!scan)
2481 lru_add_drain_all();
2482 else if (schedule_timeout_killable((HZ << scan) / 200))
2483 scan = LAST_SCAN;
2484
2485 start = 0;
2486 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002487 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2488 start, SHMEM_TAG_PINNED) {
2489
2490 page = radix_tree_deref_slot(slot);
2491 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002492 if (radix_tree_deref_retry(page)) {
2493 slot = radix_tree_iter_retry(&iter);
2494 continue;
2495 }
David Herrmann05f65b52014-08-08 14:25:36 -07002496
2497 page = NULL;
2498 }
2499
2500 if (page &&
2501 page_count(page) - page_mapcount(page) != 1) {
2502 if (scan < LAST_SCAN)
2503 goto continue_resched;
2504
2505 /*
2506 * On the last scan, we clean up all those tags
2507 * we inserted; but make a note that we still
2508 * found pages pinned.
2509 */
2510 error = -EBUSY;
2511 }
2512
2513 spin_lock_irq(&mapping->tree_lock);
2514 radix_tree_tag_clear(&mapping->page_tree,
2515 iter.index, SHMEM_TAG_PINNED);
2516 spin_unlock_irq(&mapping->tree_lock);
2517continue_resched:
2518 if (need_resched()) {
2519 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002520 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002521 }
2522 }
2523 rcu_read_unlock();
2524 }
2525
2526 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002527}
2528
2529#define F_ALL_SEALS (F_SEAL_SEAL | \
2530 F_SEAL_SHRINK | \
2531 F_SEAL_GROW | \
2532 F_SEAL_WRITE)
2533
2534int shmem_add_seals(struct file *file, unsigned int seals)
2535{
2536 struct inode *inode = file_inode(file);
2537 struct shmem_inode_info *info = SHMEM_I(inode);
2538 int error;
2539
2540 /*
2541 * SEALING
2542 * Sealing allows multiple parties to share a shmem-file but restrict
2543 * access to a specific subset of file operations. Seals can only be
2544 * added, but never removed. This way, mutually untrusted parties can
2545 * share common memory regions with a well-defined policy. A malicious
2546 * peer can thus never perform unwanted operations on a shared object.
2547 *
2548 * Seals are only supported on special shmem-files and always affect
2549 * the whole underlying inode. Once a seal is set, it may prevent some
2550 * kinds of access to the file. Currently, the following seals are
2551 * defined:
2552 * SEAL_SEAL: Prevent further seals from being set on this file
2553 * SEAL_SHRINK: Prevent the file from shrinking
2554 * SEAL_GROW: Prevent the file from growing
2555 * SEAL_WRITE: Prevent write access to the file
2556 *
2557 * As we don't require any trust relationship between two parties, we
2558 * must prevent seals from being removed. Therefore, sealing a file
2559 * only adds a given set of seals to the file, it never touches
2560 * existing seals. Furthermore, the "setting seals"-operation can be
2561 * sealed itself, which basically prevents any further seal from being
2562 * added.
2563 *
2564 * Semantics of sealing are only defined on volatile files. Only
2565 * anonymous shmem files support sealing. More importantly, seals are
2566 * never written to disk. Therefore, there's no plan to support it on
2567 * other file types.
2568 */
2569
2570 if (file->f_op != &shmem_file_operations)
2571 return -EINVAL;
2572 if (!(file->f_mode & FMODE_WRITE))
2573 return -EPERM;
2574 if (seals & ~(unsigned int)F_ALL_SEALS)
2575 return -EINVAL;
2576
Al Viro59551022016-01-22 15:40:57 -05002577 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002578
2579 if (info->seals & F_SEAL_SEAL) {
2580 error = -EPERM;
2581 goto unlock;
2582 }
2583
2584 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2585 error = mapping_deny_writable(file->f_mapping);
2586 if (error)
2587 goto unlock;
2588
2589 error = shmem_wait_for_pins(file->f_mapping);
2590 if (error) {
2591 mapping_allow_writable(file->f_mapping);
2592 goto unlock;
2593 }
2594 }
2595
2596 info->seals |= seals;
2597 error = 0;
2598
2599unlock:
Al Viro59551022016-01-22 15:40:57 -05002600 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002601 return error;
2602}
2603EXPORT_SYMBOL_GPL(shmem_add_seals);
2604
2605int shmem_get_seals(struct file *file)
2606{
2607 if (file->f_op != &shmem_file_operations)
2608 return -EINVAL;
2609
2610 return SHMEM_I(file_inode(file))->seals;
2611}
2612EXPORT_SYMBOL_GPL(shmem_get_seals);
2613
2614long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2615{
2616 long error;
2617
2618 switch (cmd) {
2619 case F_ADD_SEALS:
2620 /* disallow upper 32bit */
2621 if (arg > UINT_MAX)
2622 return -EINVAL;
2623
2624 error = shmem_add_seals(file, arg);
2625 break;
2626 case F_GET_SEALS:
2627 error = shmem_get_seals(file);
2628 break;
2629 default:
2630 error = -EINVAL;
2631 break;
2632 }
2633
2634 return error;
2635}
2636
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002637static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2638 loff_t len)
2639{
Al Viro496ad9a2013-01-23 17:07:38 -05002640 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002641 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002642 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002643 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002644 pgoff_t start, index, end;
2645 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002646
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002647 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2648 return -EOPNOTSUPP;
2649
Al Viro59551022016-01-22 15:40:57 -05002650 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002651
2652 if (mode & FALLOC_FL_PUNCH_HOLE) {
2653 struct address_space *mapping = file->f_mapping;
2654 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2655 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002656 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002657
David Herrmann40e041a2014-08-08 14:25:27 -07002658 /* protected by i_mutex */
2659 if (info->seals & F_SEAL_WRITE) {
2660 error = -EPERM;
2661 goto out;
2662 }
2663
Hugh Dickins8e205f72014-07-23 14:00:10 -07002664 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002665 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2666 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2667 spin_lock(&inode->i_lock);
2668 inode->i_private = &shmem_falloc;
2669 spin_unlock(&inode->i_lock);
2670
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002671 if ((u64)unmap_end > (u64)unmap_start)
2672 unmap_mapping_range(mapping, unmap_start,
2673 1 + unmap_end - unmap_start, 0);
2674 shmem_truncate_range(inode, offset, offset + len - 1);
2675 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002676
2677 spin_lock(&inode->i_lock);
2678 inode->i_private = NULL;
2679 wake_up_all(&shmem_falloc_waitq);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002680 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002681 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002682 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002683 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002684 }
2685
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002686 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2687 error = inode_newsize_ok(inode, offset + len);
2688 if (error)
2689 goto out;
2690
David Herrmann40e041a2014-08-08 14:25:27 -07002691 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2692 error = -EPERM;
2693 goto out;
2694 }
2695
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002696 start = offset >> PAGE_SHIFT;
2697 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002698 /* Try to avoid a swapstorm if len is impossible to satisfy */
2699 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2700 error = -ENOSPC;
2701 goto out;
2702 }
2703
Hugh Dickins8e205f72014-07-23 14:00:10 -07002704 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002705 shmem_falloc.start = start;
2706 shmem_falloc.next = start;
2707 shmem_falloc.nr_falloced = 0;
2708 shmem_falloc.nr_unswapped = 0;
2709 spin_lock(&inode->i_lock);
2710 inode->i_private = &shmem_falloc;
2711 spin_unlock(&inode->i_lock);
2712
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002713 for (index = start; index < end; index++) {
2714 struct page *page;
2715
2716 /*
2717 * Good, the fallocate(2) manpage permits EINTR: we may have
2718 * been interrupted because we are using up too much memory.
2719 */
2720 if (signal_pending(current))
2721 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002722 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2723 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002724 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002725 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002726 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002727 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002728 if (index > start) {
2729 shmem_undo_range(inode,
2730 (loff_t)start << PAGE_SHIFT,
2731 ((loff_t)index << PAGE_SHIFT) - 1, true);
2732 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002733 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002734 }
2735
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002736 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002737 * Inform shmem_writepage() how far we have reached.
2738 * No need for lock or barrier: we have the page lock.
2739 */
2740 shmem_falloc.next++;
2741 if (!PageUptodate(page))
2742 shmem_falloc.nr_falloced++;
2743
2744 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002745 * If !PageUptodate, leave it that way so that freeable pages
2746 * can be recognized if we need to rollback on error later.
2747 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002748 * than free the pages we are allocating (and SGP_CACHE pages
2749 * might still be clean: we now need to mark those dirty too).
2750 */
2751 set_page_dirty(page);
2752 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002753 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002754 cond_resched();
2755 }
2756
2757 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2758 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002759 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002760undone:
2761 spin_lock(&inode->i_lock);
2762 inode->i_private = NULL;
2763 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002764out:
Al Viro59551022016-01-22 15:40:57 -05002765 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002766 return error;
2767}
2768
David Howells726c3342006-06-23 02:02:58 -07002769static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770{
David Howells726c3342006-06-23 02:02:58 -07002771 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772
2773 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002774 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002776 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002778 buf->f_bavail =
2779 buf->f_bfree = sbinfo->max_blocks -
2780 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002781 }
2782 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 buf->f_files = sbinfo->max_inodes;
2784 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 }
2786 /* else leave those fields 0 like simple_statfs */
2787 return 0;
2788}
2789
2790/*
2791 * File creation. Allocate an inode, and we're done..
2792 */
2793static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002794shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002796 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 int error = -ENOSPC;
2798
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002799 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002801 error = simple_acl_create(dir, inode);
2802 if (error)
2803 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002804 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002805 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002806 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002807 if (error && error != -EOPNOTSUPP)
2808 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002809
Al Viro718deb62009-12-16 19:35:36 -05002810 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002812 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 d_instantiate(dentry, inode);
2814 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 }
2816 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002817out_iput:
2818 iput(inode);
2819 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820}
2821
Al Viro60545d02013-06-07 01:20:27 -04002822static int
2823shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2824{
2825 struct inode *inode;
2826 int error = -ENOSPC;
2827
2828 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2829 if (inode) {
2830 error = security_inode_init_security(inode, dir,
2831 NULL,
2832 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002833 if (error && error != -EOPNOTSUPP)
2834 goto out_iput;
2835 error = simple_acl_create(dir, inode);
2836 if (error)
2837 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002838 d_tmpfile(dentry, inode);
2839 }
2840 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002841out_iput:
2842 iput(inode);
2843 return error;
Al Viro60545d02013-06-07 01:20:27 -04002844}
2845
Al Viro18bb1db2011-07-26 01:41:39 -04002846static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847{
2848 int error;
2849
2850 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2851 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002852 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 return 0;
2854}
2855
Al Viro4acdaf22011-07-26 01:42:34 -04002856static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002857 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858{
2859 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2860}
2861
2862/*
2863 * Link a file..
2864 */
2865static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2866{
David Howells75c3cfa2015-03-17 22:26:12 +00002867 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002868 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869
2870 /*
2871 * No ordinary (disk based) filesystem counts links as inodes;
2872 * but each new link needs a new dentry, pinning lowmem, and
2873 * tmpfs dentries cannot be pruned until they are unlinked.
2874 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002875 ret = shmem_reserve_inode(inode->i_sb);
2876 if (ret)
2877 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878
2879 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002880 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002881 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002882 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 dget(dentry); /* Extra pinning count for the created dentry */
2884 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002885out:
2886 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887}
2888
2889static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2890{
David Howells75c3cfa2015-03-17 22:26:12 +00002891 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002893 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2894 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
2896 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002897 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002898 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 dput(dentry); /* Undo the count from "create" - this does all the work */
2900 return 0;
2901}
2902
2903static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2904{
2905 if (!simple_empty(dentry))
2906 return -ENOTEMPTY;
2907
David Howells75c3cfa2015-03-17 22:26:12 +00002908 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002909 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 return shmem_unlink(dir, dentry);
2911}
2912
Miklos Szeredi37456772014-07-23 15:15:34 +02002913static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2914{
David Howellse36cb0b2015-01-29 12:02:35 +00002915 bool old_is_dir = d_is_dir(old_dentry);
2916 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002917
2918 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2919 if (old_is_dir) {
2920 drop_nlink(old_dir);
2921 inc_nlink(new_dir);
2922 } else {
2923 drop_nlink(new_dir);
2924 inc_nlink(old_dir);
2925 }
2926 }
2927 old_dir->i_ctime = old_dir->i_mtime =
2928 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002929 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002930 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02002931
2932 return 0;
2933}
2934
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002935static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2936{
2937 struct dentry *whiteout;
2938 int error;
2939
2940 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2941 if (!whiteout)
2942 return -ENOMEM;
2943
2944 error = shmem_mknod(old_dir, whiteout,
2945 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2946 dput(whiteout);
2947 if (error)
2948 return error;
2949
2950 /*
2951 * Cheat and hash the whiteout while the old dentry is still in
2952 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2953 *
2954 * d_lookup() will consistently find one of them at this point,
2955 * not sure which one, but that isn't even important.
2956 */
2957 d_rehash(whiteout);
2958 return 0;
2959}
2960
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961/*
2962 * The VFS layer already does all the dentry stuff for rename,
2963 * we just have to decrement the usage count for the target if
2964 * it exists so that the VFS layer correctly free's it when it
2965 * gets overwritten.
2966 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002967static 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 -07002968{
David Howells75c3cfa2015-03-17 22:26:12 +00002969 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 int they_are_dirs = S_ISDIR(inode->i_mode);
2971
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002972 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002973 return -EINVAL;
2974
Miklos Szeredi37456772014-07-23 15:15:34 +02002975 if (flags & RENAME_EXCHANGE)
2976 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2977
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 if (!simple_empty(new_dentry))
2979 return -ENOTEMPTY;
2980
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002981 if (flags & RENAME_WHITEOUT) {
2982 int error;
2983
2984 error = shmem_whiteout(old_dir, old_dentry);
2985 if (error)
2986 return error;
2987 }
2988
David Howells75c3cfa2015-03-17 22:26:12 +00002989 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002991 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002992 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002993 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002996 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002997 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 }
2999
3000 old_dir->i_size -= BOGO_DIRENT_SIZE;
3001 new_dir->i_size += BOGO_DIRENT_SIZE;
3002 old_dir->i_ctime = old_dir->i_mtime =
3003 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003004 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 return 0;
3006}
3007
3008static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3009{
3010 int error;
3011 int len;
3012 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003013 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 struct shmem_inode_info *info;
3015
3016 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003017 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 return -ENAMETOOLONG;
3019
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003020 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 if (!inode)
3022 return -ENOSPC;
3023
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003024 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003025 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003026 if (error) {
3027 if (error != -EOPNOTSUPP) {
3028 iput(inode);
3029 return error;
3030 }
3031 error = 0;
3032 }
3033
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 info = SHMEM_I(inode);
3035 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003036 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003037 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3038 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003039 iput(inode);
3040 return -ENOMEM;
3041 }
3042 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003044 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003045 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 if (error) {
3047 iput(inode);
3048 return error;
3049 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003050 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003052 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003053 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003055 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003056 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003059 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 d_instantiate(dentry, inode);
3061 dget(dentry);
3062 return 0;
3063}
3064
Al Virofceef392015-12-29 15:58:39 -05003065static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066{
Al Virofceef392015-12-29 15:58:39 -05003067 mark_page_accessed(arg);
3068 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069}
3070
Al Viro6b255392015-11-17 10:20:54 -05003071static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003072 struct inode *inode,
3073 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003076 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003077 if (!dentry) {
3078 page = find_get_page(inode->i_mapping, 0);
3079 if (!page)
3080 return ERR_PTR(-ECHILD);
3081 if (!PageUptodate(page)) {
3082 put_page(page);
3083 return ERR_PTR(-ECHILD);
3084 }
3085 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003086 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003087 if (error)
3088 return ERR_PTR(error);
3089 unlock_page(page);
3090 }
Al Virofceef392015-12-29 15:58:39 -05003091 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003092 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093}
3094
Eric Parisb09e0fa2011-05-24 17:12:39 -07003095#ifdef CONFIG_TMPFS_XATTR
3096/*
3097 * Superblocks without xattr inode operations may get some security.* xattr
3098 * support from the LSM "for free". As soon as we have any other xattrs
3099 * like ACLs, we also need to implement the security.* handlers at
3100 * filesystem level, though.
3101 */
3102
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003103/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003104 * Callback for security_inode_init_security() for acquiring xattrs.
3105 */
3106static int shmem_initxattrs(struct inode *inode,
3107 const struct xattr *xattr_array,
3108 void *fs_info)
3109{
3110 struct shmem_inode_info *info = SHMEM_I(inode);
3111 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003112 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003113 size_t len;
3114
3115 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003116 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003117 if (!new_xattr)
3118 return -ENOMEM;
3119
3120 len = strlen(xattr->name) + 1;
3121 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3122 GFP_KERNEL);
3123 if (!new_xattr->name) {
3124 kfree(new_xattr);
3125 return -ENOMEM;
3126 }
3127
3128 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3129 XATTR_SECURITY_PREFIX_LEN);
3130 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3131 xattr->name, len);
3132
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003133 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003134 }
3135
3136 return 0;
3137}
3138
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003139static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003140 struct dentry *unused, struct inode *inode,
3141 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003142{
Al Virob2968212016-04-10 20:48:24 -04003143 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003144
3145 name = xattr_full_name(handler, name);
3146 return simple_xattr_get(&info->xattrs, name, buffer, size);
3147}
3148
3149static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003150 struct dentry *unused, struct inode *inode,
3151 const char *name, const void *value,
3152 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003153{
Al Viro59301222016-05-27 10:19:30 -04003154 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003155
3156 name = xattr_full_name(handler, name);
3157 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3158}
3159
3160static const struct xattr_handler shmem_security_xattr_handler = {
3161 .prefix = XATTR_SECURITY_PREFIX,
3162 .get = shmem_xattr_handler_get,
3163 .set = shmem_xattr_handler_set,
3164};
3165
3166static const struct xattr_handler shmem_trusted_xattr_handler = {
3167 .prefix = XATTR_TRUSTED_PREFIX,
3168 .get = shmem_xattr_handler_get,
3169 .set = shmem_xattr_handler_set,
3170};
3171
Eric Parisb09e0fa2011-05-24 17:12:39 -07003172static const struct xattr_handler *shmem_xattr_handlers[] = {
3173#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003174 &posix_acl_access_xattr_handler,
3175 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003176#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003177 &shmem_security_xattr_handler,
3178 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003179 NULL
3180};
3181
Eric Parisb09e0fa2011-05-24 17:12:39 -07003182static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3183{
David Howells75c3cfa2015-03-17 22:26:12 +00003184 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003185 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003186}
3187#endif /* CONFIG_TMPFS_XATTR */
3188
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003189static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003191 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003192#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003193 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003194#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195};
3196
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003197static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003199 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003200#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003201 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003202#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003203};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003204
David M. Grimes91828a42006-10-17 00:09:45 -07003205static struct dentry *shmem_get_parent(struct dentry *child)
3206{
3207 return ERR_PTR(-ESTALE);
3208}
3209
3210static int shmem_match(struct inode *ino, void *vfh)
3211{
3212 __u32 *fh = vfh;
3213 __u64 inum = fh[2];
3214 inum = (inum << 32) | fh[1];
3215 return ino->i_ino == inum && fh[0] == ino->i_generation;
3216}
3217
Christoph Hellwig480b1162007-10-21 16:42:13 -07003218static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3219 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003220{
David M. Grimes91828a42006-10-17 00:09:45 -07003221 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003222 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003223 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003224
Christoph Hellwig480b1162007-10-21 16:42:13 -07003225 if (fh_len < 3)
3226 return NULL;
3227
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003228 inum = fid->raw[2];
3229 inum = (inum << 32) | fid->raw[1];
3230
Christoph Hellwig480b1162007-10-21 16:42:13 -07003231 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3232 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003233 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003234 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003235 iput(inode);
3236 }
3237
Christoph Hellwig480b1162007-10-21 16:42:13 -07003238 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003239}
3240
Al Virob0b03822012-04-02 14:34:06 -04003241static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3242 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003243{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303244 if (*len < 3) {
3245 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003246 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303247 }
David M. Grimes91828a42006-10-17 00:09:45 -07003248
Al Viro1d3382cb2010-10-23 15:19:20 -04003249 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003250 /* Unfortunately insert_inode_hash is not idempotent,
3251 * so as we hash inodes here rather than at creation
3252 * time, we need a lock to ensure we only try
3253 * to do it once
3254 */
3255 static DEFINE_SPINLOCK(lock);
3256 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003257 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003258 __insert_inode_hash(inode,
3259 inode->i_ino + inode->i_generation);
3260 spin_unlock(&lock);
3261 }
3262
3263 fh[0] = inode->i_generation;
3264 fh[1] = inode->i_ino;
3265 fh[2] = ((__u64)inode->i_ino) >> 32;
3266
3267 *len = 3;
3268 return 1;
3269}
3270
Christoph Hellwig39655162007-10-21 16:42:17 -07003271static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003272 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003273 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003274 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003275};
3276
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003277static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3278 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279{
3280 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003281 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003282 uid_t uid;
3283 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003285 while (options != NULL) {
3286 this_char = options;
3287 for (;;) {
3288 /*
3289 * NUL-terminate this option: unfortunately,
3290 * mount options form a comma-separated list,
3291 * but mpol's nodelist may also contain commas.
3292 */
3293 options = strchr(options, ',');
3294 if (options == NULL)
3295 break;
3296 options++;
3297 if (!isdigit(*options)) {
3298 options[-1] = '\0';
3299 break;
3300 }
3301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 if (!*this_char)
3303 continue;
3304 if ((value = strchr(this_char,'=')) != NULL) {
3305 *value++ = 0;
3306 } else {
Joe Perches11705322016-03-17 14:19:50 -07003307 pr_err("tmpfs: No value for mount option '%s'\n",
3308 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003309 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 }
3311
3312 if (!strcmp(this_char,"size")) {
3313 unsigned long long size;
3314 size = memparse(value,&rest);
3315 if (*rest == '%') {
3316 size <<= PAGE_SHIFT;
3317 size *= totalram_pages;
3318 do_div(size, 100);
3319 rest++;
3320 }
3321 if (*rest)
3322 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003323 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003324 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003326 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 if (*rest)
3328 goto bad_val;
3329 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003330 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331 if (*rest)
3332 goto bad_val;
3333 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003334 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003336 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 if (*rest)
3338 goto bad_val;
3339 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003340 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003342 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 if (*rest)
3344 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003345 sbinfo->uid = make_kuid(current_user_ns(), uid);
3346 if (!uid_valid(sbinfo->uid))
3347 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003349 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003351 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 if (*rest)
3353 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003354 sbinfo->gid = make_kgid(current_user_ns(), gid);
3355 if (!gid_valid(sbinfo->gid))
3356 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003357#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003358 } else if (!strcmp(this_char, "huge")) {
3359 int huge;
3360 huge = shmem_parse_huge(value);
3361 if (huge < 0)
3362 goto bad_val;
3363 if (!has_transparent_hugepage() &&
3364 huge != SHMEM_HUGE_NEVER)
3365 goto bad_val;
3366 sbinfo->huge = huge;
3367#endif
3368#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003369 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003370 mpol_put(mpol);
3371 mpol = NULL;
3372 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003373 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003374#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 } else {
Joe Perches11705322016-03-17 14:19:50 -07003376 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003377 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 }
3379 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003380 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 return 0;
3382
3383bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003384 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003386error:
3387 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 return 1;
3389
3390}
3391
3392static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3393{
3394 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003395 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003396 unsigned long inodes;
3397 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398
Greg Thelen5f001102013-02-22 16:36:01 -08003399 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003400 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003401 return error;
3402
3403 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003404 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003405 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003406 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003407 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003408 goto out;
3409 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003410 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003411 * but we must separately disallow unlimited->limited, because
3412 * in that case we have no record of how much is already in use.
3413 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003414 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003415 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003416 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003417 goto out;
3418
3419 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003420 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003421 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003422 sbinfo->max_inodes = config.max_inodes;
3423 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003424
Greg Thelen5f001102013-02-22 16:36:01 -08003425 /*
3426 * Preserve previous mempolicy unless mpol remount option was specified.
3427 */
3428 if (config.mpol) {
3429 mpol_put(sbinfo->mpol);
3430 sbinfo->mpol = config.mpol; /* transfers initial ref */
3431 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003432out:
3433 spin_unlock(&sbinfo->stat_lock);
3434 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003436
Al Viro34c80b12011-12-08 21:32:45 -05003437static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003438{
Al Viro34c80b12011-12-08 21:32:45 -05003439 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003440
3441 if (sbinfo->max_blocks != shmem_default_max_blocks())
3442 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003443 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003444 if (sbinfo->max_inodes != shmem_default_max_inodes())
3445 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3446 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003447 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003448 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3449 seq_printf(seq, ",uid=%u",
3450 from_kuid_munged(&init_user_ns, sbinfo->uid));
3451 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3452 seq_printf(seq, ",gid=%u",
3453 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003454#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003455 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3456 if (sbinfo->huge)
3457 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3458#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003459 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003460 return 0;
3461}
David Herrmann9183df22014-08-08 14:25:29 -07003462
3463#define MFD_NAME_PREFIX "memfd:"
3464#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3465#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3466
3467#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3468
3469SYSCALL_DEFINE2(memfd_create,
3470 const char __user *, uname,
3471 unsigned int, flags)
3472{
3473 struct shmem_inode_info *info;
3474 struct file *file;
3475 int fd, error;
3476 char *name;
3477 long len;
3478
3479 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3480 return -EINVAL;
3481
3482 /* length includes terminating zero */
3483 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3484 if (len <= 0)
3485 return -EFAULT;
3486 if (len > MFD_NAME_MAX_LEN + 1)
3487 return -EINVAL;
3488
3489 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3490 if (!name)
3491 return -ENOMEM;
3492
3493 strcpy(name, MFD_NAME_PREFIX);
3494 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3495 error = -EFAULT;
3496 goto err_name;
3497 }
3498
3499 /* terminating-zero may have changed after strnlen_user() returned */
3500 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3501 error = -EFAULT;
3502 goto err_name;
3503 }
3504
3505 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3506 if (fd < 0) {
3507 error = fd;
3508 goto err_name;
3509 }
3510
3511 file = shmem_file_setup(name, 0, VM_NORESERVE);
3512 if (IS_ERR(file)) {
3513 error = PTR_ERR(file);
3514 goto err_fd;
3515 }
3516 info = SHMEM_I(file_inode(file));
3517 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3518 file->f_flags |= O_RDWR | O_LARGEFILE;
3519 if (flags & MFD_ALLOW_SEALING)
3520 info->seals &= ~F_SEAL_SEAL;
3521
3522 fd_install(fd, file);
3523 kfree(name);
3524 return fd;
3525
3526err_fd:
3527 put_unused_fd(fd);
3528err_name:
3529 kfree(name);
3530 return error;
3531}
3532
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003533#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534
3535static void shmem_put_super(struct super_block *sb)
3536{
Hugh Dickins602586a2010-08-17 15:23:56 -07003537 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3538
3539 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003540 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003541 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 sb->s_fs_info = NULL;
3543}
3544
Kay Sievers2b2af542009-04-30 15:23:42 +02003545int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546{
3547 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003548 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003549 int err = -ENOMEM;
3550
3551 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003552 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003553 L1_CACHE_BYTES), GFP_KERNEL);
3554 if (!sbinfo)
3555 return -ENOMEM;
3556
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003557 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003558 sbinfo->uid = current_fsuid();
3559 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003560 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003562#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 /*
3564 * Per default we only allow half of the physical ram per
3565 * tmpfs instance, limiting inodes to one per page of lowmem;
3566 * but the internal instance is left unlimited.
3567 */
Al Viroca4e0512013-08-31 12:57:10 -04003568 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003569 sbinfo->max_blocks = shmem_default_max_blocks();
3570 sbinfo->max_inodes = shmem_default_max_inodes();
3571 if (shmem_parse_options(data, sbinfo, false)) {
3572 err = -EINVAL;
3573 goto failed;
3574 }
Al Viroca4e0512013-08-31 12:57:10 -04003575 } else {
3576 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577 }
David M. Grimes91828a42006-10-17 00:09:45 -07003578 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003579 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580#else
3581 sb->s_flags |= MS_NOUSER;
3582#endif
3583
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003584 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003585 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003586 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003587 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003588 spin_lock_init(&sbinfo->shrinklist_lock);
3589 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003590
Hugh Dickins285b2c42011-08-03 16:21:20 -07003591 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003592 sb->s_blocksize = PAGE_SIZE;
3593 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 sb->s_magic = TMPFS_MAGIC;
3595 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003596 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003597#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003598 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003599#endif
3600#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003601 sb->s_flags |= MS_POSIXACL;
3602#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003603
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003604 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 if (!inode)
3606 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003607 inode->i_uid = sbinfo->uid;
3608 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003609 sb->s_root = d_make_root(inode);
3610 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003611 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 return 0;
3613
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614failed:
3615 shmem_put_super(sb);
3616 return err;
3617}
3618
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003619static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620
3621static struct inode *shmem_alloc_inode(struct super_block *sb)
3622{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003623 struct shmem_inode_info *info;
3624 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3625 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003627 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628}
3629
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003630static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003631{
3632 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003633 if (S_ISLNK(inode->i_mode))
3634 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003635 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3636}
3637
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638static void shmem_destroy_inode(struct inode *inode)
3639{
Al Viro09208d12011-07-26 03:15:03 -04003640 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003642 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643}
3644
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003645static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003647 struct shmem_inode_info *info = foo;
3648 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649}
3650
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003651static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652{
3653 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3654 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003655 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 return 0;
3657}
3658
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003659static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003661 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662}
3663
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003664static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003666 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003668 .write_begin = shmem_write_begin,
3669 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003671#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003672 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003673#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003674 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675};
3676
Helge Deller15ad7cd2006-12-06 20:40:36 -08003677static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003679 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003681 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003682 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003683 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003684 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003685 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003686 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003687 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688#endif
3689};
3690
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003691static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003692 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003693 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003694#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003695 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003696 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003697#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698};
3699
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003700static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701#ifdef CONFIG_TMPFS
3702 .create = shmem_create,
3703 .lookup = simple_lookup,
3704 .link = shmem_link,
3705 .unlink = shmem_unlink,
3706 .symlink = shmem_symlink,
3707 .mkdir = shmem_mkdir,
3708 .rmdir = shmem_rmdir,
3709 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003710 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003711 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003713#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003714 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003715#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003716#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003717 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003718 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003719#endif
3720};
3721
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003722static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003723#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003724 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003725#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003726#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003727 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003728 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003729#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730};
3731
Hugh Dickins759b9772007-03-05 00:30:28 -08003732static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733 .alloc_inode = shmem_alloc_inode,
3734 .destroy_inode = shmem_destroy_inode,
3735#ifdef CONFIG_TMPFS
3736 .statfs = shmem_statfs,
3737 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003738 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739#endif
Al Viro1f895f72010-06-05 19:10:41 -04003740 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 .drop_inode = generic_delete_inode,
3742 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003743#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3744 .nr_cached_objects = shmem_unused_huge_count,
3745 .free_cached_objects = shmem_unused_huge_scan,
3746#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747};
3748
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003749static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003750 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003751 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752#ifdef CONFIG_NUMA
3753 .set_policy = shmem_set_policy,
3754 .get_policy = shmem_get_policy,
3755#endif
3756};
3757
Al Viro3c26ff62010-07-25 11:46:36 +04003758static struct dentry *shmem_mount(struct file_system_type *fs_type,
3759 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760{
Al Viro3c26ff62010-07-25 11:46:36 +04003761 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762}
3763
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003764static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 .owner = THIS_MODULE,
3766 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003767 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003769 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003772int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773{
3774 int error;
3775
Rob Landley16203a72013-09-11 14:26:12 -07003776 /* If rootfs called this, don't re-init */
3777 if (shmem_inode_cachep)
3778 return 0;
3779
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003780 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 if (error)
3782 goto out3;
3783
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003784 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003786 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 goto out2;
3788 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003789
Al Viroca4e0512013-08-31 12:57:10 -04003790 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 if (IS_ERR(shm_mnt)) {
3792 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003793 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 goto out1;
3795 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003796
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003797#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003798 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3799 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3800 else
3801 shmem_huge = 0; /* just in case it was patched */
3802#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 return 0;
3804
3805out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003806 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003808 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809out3:
3810 shm_mnt = ERR_PTR(error);
3811 return error;
3812}
Matt Mackall853ac432009-01-06 14:40:20 -08003813
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003814#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003815static ssize_t shmem_enabled_show(struct kobject *kobj,
3816 struct kobj_attribute *attr, char *buf)
3817{
3818 int values[] = {
3819 SHMEM_HUGE_ALWAYS,
3820 SHMEM_HUGE_WITHIN_SIZE,
3821 SHMEM_HUGE_ADVISE,
3822 SHMEM_HUGE_NEVER,
3823 SHMEM_HUGE_DENY,
3824 SHMEM_HUGE_FORCE,
3825 };
3826 int i, count;
3827
3828 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3829 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3830
3831 count += sprintf(buf + count, fmt,
3832 shmem_format_huge(values[i]));
3833 }
3834 buf[count - 1] = '\n';
3835 return count;
3836}
3837
3838static ssize_t shmem_enabled_store(struct kobject *kobj,
3839 struct kobj_attribute *attr, const char *buf, size_t count)
3840{
3841 char tmp[16];
3842 int huge;
3843
3844 if (count + 1 > sizeof(tmp))
3845 return -EINVAL;
3846 memcpy(tmp, buf, count);
3847 tmp[count] = '\0';
3848 if (count && tmp[count - 1] == '\n')
3849 tmp[count - 1] = '\0';
3850
3851 huge = shmem_parse_huge(tmp);
3852 if (huge == -EINVAL)
3853 return -EINVAL;
3854 if (!has_transparent_hugepage() &&
3855 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3856 return -EINVAL;
3857
3858 shmem_huge = huge;
3859 if (shmem_huge < SHMEM_HUGE_DENY)
3860 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3861 return count;
3862}
3863
3864struct kobj_attribute shmem_enabled_attr =
3865 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003866#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003867
Arnd Bergmann3b337192016-08-10 16:27:44 -07003868#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003869bool shmem_huge_enabled(struct vm_area_struct *vma)
3870{
3871 struct inode *inode = file_inode(vma->vm_file);
3872 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3873 loff_t i_size;
3874 pgoff_t off;
3875
3876 if (shmem_huge == SHMEM_HUGE_FORCE)
3877 return true;
3878 if (shmem_huge == SHMEM_HUGE_DENY)
3879 return false;
3880 switch (sbinfo->huge) {
3881 case SHMEM_HUGE_NEVER:
3882 return false;
3883 case SHMEM_HUGE_ALWAYS:
3884 return true;
3885 case SHMEM_HUGE_WITHIN_SIZE:
3886 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3887 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3888 if (i_size >= HPAGE_PMD_SIZE &&
3889 i_size >> PAGE_SHIFT >= off)
3890 return true;
3891 case SHMEM_HUGE_ADVISE:
3892 /* TODO: implement fadvise() hints */
3893 return (vma->vm_flags & VM_HUGEPAGE);
3894 default:
3895 VM_BUG_ON(1);
3896 return false;
3897 }
3898}
Arnd Bergmann3b337192016-08-10 16:27:44 -07003899#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003900
Matt Mackall853ac432009-01-06 14:40:20 -08003901#else /* !CONFIG_SHMEM */
3902
3903/*
3904 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3905 *
3906 * This is intended for small system where the benefits of the full
3907 * shmem code (swap-backed and resource-limited) are outweighed by
3908 * their complexity. On systems without swap this code should be
3909 * effectively equivalent, but much lighter weight.
3910 */
3911
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003912static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003913 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003914 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003915 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003916 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003917};
3918
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003919int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003920{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003921 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003922
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003923 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003924 BUG_ON(IS_ERR(shm_mnt));
3925
3926 return 0;
3927}
3928
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003929int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003930{
3931 return 0;
3932}
3933
Hugh Dickins3f96b792009-09-21 17:03:37 -07003934int shmem_lock(struct file *file, int lock, struct user_struct *user)
3935{
3936 return 0;
3937}
3938
Hugh Dickins24513262012-01-20 14:34:21 -08003939void shmem_unlock_mapping(struct address_space *mapping)
3940{
3941}
3942
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003943#ifdef CONFIG_MMU
3944unsigned long shmem_get_unmapped_area(struct file *file,
3945 unsigned long addr, unsigned long len,
3946 unsigned long pgoff, unsigned long flags)
3947{
3948 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3949}
3950#endif
3951
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003952void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003953{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003954 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003955}
3956EXPORT_SYMBOL_GPL(shmem_truncate_range);
3957
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003958#define shmem_vm_ops generic_file_vm_ops
3959#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003960#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003961#define shmem_acct_size(flags, size) 0
3962#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003963
3964#endif /* CONFIG_SHMEM */
3965
3966/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967
Rasmus Villemoes19938e32016-10-07 17:01:01 -07003968static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003969 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003970};
3971
Eric Parisc7277092013-12-02 11:24:19 +00003972static struct file *__shmem_file_setup(const char *name, loff_t size,
3973 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974{
Al Viro6b4d0b22013-02-14 21:37:26 -05003975 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003977 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003978 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 struct qstr this;
3980
3981 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003982 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983
Hugh Dickins285b2c42011-08-03 16:21:20 -07003984 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 return ERR_PTR(-EINVAL);
3986
3987 if (shmem_acct_size(flags, size))
3988 return ERR_PTR(-ENOMEM);
3989
Al Viro6b4d0b22013-02-14 21:37:26 -05003990 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 this.name = name;
3992 this.len = strlen(name);
3993 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003994 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003995 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003996 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003997 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003999 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000
Al Viro6b4d0b22013-02-14 21:37:26 -05004001 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004002 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004004 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005
Eric Parisc7277092013-12-02 11:24:19 +00004006 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004007 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004009 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004010 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4011 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004012 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004013
Al Viro6b4d0b22013-02-14 21:37:26 -05004014 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004015 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004016 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004017 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004018
Al Viro6b4d0b22013-02-14 21:37:26 -05004019 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021put_memory:
4022 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004023put_path:
4024 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004025 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026}
Eric Parisc7277092013-12-02 11:24:19 +00004027
4028/**
4029 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4030 * kernel internal. There will be NO LSM permission checks against the
4031 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004032 * higher layer. The users are the big_key and shm implementations. LSM
4033 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004034 * @name: name for dentry (to be seen in /proc/<pid>/maps
4035 * @size: size to be set for the file
4036 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4037 */
4038struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4039{
4040 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4041}
4042
4043/**
4044 * shmem_file_setup - get an unlinked file living in tmpfs
4045 * @name: name for dentry (to be seen in /proc/<pid>/maps
4046 * @size: size to be set for the file
4047 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4048 */
4049struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4050{
4051 return __shmem_file_setup(name, size, flags, 0);
4052}
Keith Packard395e0dd2008-06-20 00:08:06 -07004053EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054
Randy Dunlap46711812008-03-19 17:00:41 -07004055/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004057 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4058 */
4059int shmem_zero_setup(struct vm_area_struct *vma)
4060{
4061 struct file *file;
4062 loff_t size = vma->vm_end - vma->vm_start;
4063
Hugh Dickins66fc1302015-06-14 09:48:09 -07004064 /*
4065 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4066 * between XFS directory reading and selinux: since this file is only
4067 * accessible to the user through its mapping, use S_PRIVATE flag to
4068 * bypass file security, in the same way as shmem_kernel_file_setup().
4069 */
4070 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 if (IS_ERR(file))
4072 return PTR_ERR(file);
4073
4074 if (vma->vm_file)
4075 fput(vma->vm_file);
4076 vma->vm_file = file;
4077 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004078
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004079 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004080 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4081 (vma->vm_end & HPAGE_PMD_MASK)) {
4082 khugepaged_enter(vma, vma->vm_flags);
4083 }
4084
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085 return 0;
4086}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004087
4088/**
4089 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4090 * @mapping: the page's address_space
4091 * @index: the page index
4092 * @gfp: the page allocator flags to use if allocating
4093 *
4094 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4095 * with any new page allocations done using the specified allocation flags.
4096 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4097 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4098 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4099 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004100 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4101 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004102 */
4103struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4104 pgoff_t index, gfp_t gfp)
4105{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004106#ifdef CONFIG_SHMEM
4107 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004108 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004109 int error;
4110
4111 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004112 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4113 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004114 if (error)
4115 page = ERR_PTR(error);
4116 else
4117 unlock_page(page);
4118 return page;
4119#else
4120 /*
4121 * The tiny !SHMEM case uses ramfs without swap
4122 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004123 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004124#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004125}
4126EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);