blob: d99cfb6eb03a343f7ef0987b2f540327d1a02c31 [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;
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400415 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700416 struct inode *inode;
417 struct shmem_inode_info *info;
418 struct page *page;
419 unsigned long batch = sc ? sc->nr_to_scan : 128;
420 int removed = 0, split = 0;
421
422 if (list_empty(&sbinfo->shrinklist))
423 return SHRINK_STOP;
424
425 spin_lock(&sbinfo->shrinklist_lock);
426 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
427 info = list_entry(pos, struct shmem_inode_info, shrinklist);
428
429 /* pin the inode */
430 inode = igrab(&info->vfs_inode);
431
432 /* inode is about to be evicted */
433 if (!inode) {
434 list_del_init(&info->shrinklist);
435 removed++;
436 goto next;
437 }
438
439 /* Check if there's anything to gain */
440 if (round_up(inode->i_size, PAGE_SIZE) ==
441 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400442 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700443 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700444 goto next;
445 }
446
447 list_move(&info->shrinklist, &list);
448next:
449 if (!--batch)
450 break;
451 }
452 spin_unlock(&sbinfo->shrinklist_lock);
453
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400454 list_for_each_safe(pos, next, &to_remove) {
455 info = list_entry(pos, struct shmem_inode_info, shrinklist);
456 inode = &info->vfs_inode;
457 list_del_init(&info->shrinklist);
458 iput(inode);
459 }
460
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700461 list_for_each_safe(pos, next, &list) {
462 int ret;
463
464 info = list_entry(pos, struct shmem_inode_info, shrinklist);
465 inode = &info->vfs_inode;
466
467 if (nr_to_split && split >= nr_to_split) {
468 iput(inode);
469 continue;
470 }
471
472 page = find_lock_page(inode->i_mapping,
473 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
474 if (!page)
475 goto drop;
476
477 if (!PageTransHuge(page)) {
478 unlock_page(page);
479 put_page(page);
480 goto drop;
481 }
482
483 ret = split_huge_page(page);
484 unlock_page(page);
485 put_page(page);
486
487 if (ret) {
488 /* split failed: leave it on the list */
489 iput(inode);
490 continue;
491 }
492
493 split++;
494drop:
495 list_del_init(&info->shrinklist);
496 removed++;
497 iput(inode);
498 }
499
500 spin_lock(&sbinfo->shrinklist_lock);
501 list_splice_tail(&list, &sbinfo->shrinklist);
502 sbinfo->shrinklist_len -= removed;
503 spin_unlock(&sbinfo->shrinklist_lock);
504
505 return split;
506}
507
508static long shmem_unused_huge_scan(struct super_block *sb,
509 struct shrink_control *sc)
510{
511 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
512
513 if (!READ_ONCE(sbinfo->shrinklist_len))
514 return SHRINK_STOP;
515
516 return shmem_unused_huge_shrink(sbinfo, sc, 0);
517}
518
519static long shmem_unused_huge_count(struct super_block *sb,
520 struct shrink_control *sc)
521{
522 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
523 return READ_ONCE(sbinfo->shrinklist_len);
524}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700525#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700526
527#define shmem_huge SHMEM_HUGE_DENY
528
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700529static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
530 struct shrink_control *sc, unsigned long nr_to_split)
531{
532 return 0;
533}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700534#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700535
536/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700537 * Like add_to_page_cache_locked, but error if expected item has gone.
538 */
539static int shmem_add_to_page_cache(struct page *page,
540 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700541 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700542{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700543 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700544
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700545 VM_BUG_ON_PAGE(PageTail(page), page);
546 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800547 VM_BUG_ON_PAGE(!PageLocked(page), page);
548 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700549 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700550
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700551 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700552 page->mapping = mapping;
553 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700554
Hugh Dickinsb065b432012-07-11 14:02:48 -0700555 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700556 if (PageTransHuge(page)) {
557 void __rcu **results;
558 pgoff_t idx;
559 int i;
560
561 error = 0;
562 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
563 &results, &idx, index, 1) &&
564 idx < index + HPAGE_PMD_NR) {
565 error = -EEXIST;
566 }
567
568 if (!error) {
569 for (i = 0; i < HPAGE_PMD_NR; i++) {
570 error = radix_tree_insert(&mapping->page_tree,
571 index + i, page + i);
572 VM_BUG_ON(error);
573 }
574 count_vm_event(THP_FILE_ALLOC);
575 }
576 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700577 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700578 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700579 error = shmem_radix_tree_replace(mapping, index, expected,
580 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700581 }
582
Hugh Dickinsb065b432012-07-11 14:02:48 -0700583 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700584 mapping->nrpages += nr;
585 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700586 __inc_node_page_state(page, NR_SHMEM_THPS);
587 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
588 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700589 spin_unlock_irq(&mapping->tree_lock);
590 } else {
591 page->mapping = NULL;
592 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700593 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700594 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700595 return error;
596}
597
598/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700599 * Like delete_from_page_cache, but substitutes swap for page.
600 */
601static void shmem_delete_from_page_cache(struct page *page, void *radswap)
602{
603 struct address_space *mapping = page->mapping;
604 int error;
605
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700606 VM_BUG_ON_PAGE(PageCompound(page), page);
607
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700608 spin_lock_irq(&mapping->tree_lock);
609 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
610 page->mapping = NULL;
611 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700612 __dec_node_page_state(page, NR_FILE_PAGES);
613 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700614 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300615 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700616 BUG_ON(error);
617}
618
619/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700620 * Remove swap entry from radix tree, free the swap and its page cache.
621 */
622static int shmem_free_swap(struct address_space *mapping,
623 pgoff_t index, void *radswap)
624{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700625 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700626
627 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700628 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700629 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700630 if (old != radswap)
631 return -ENOENT;
632 free_swap_and_cache(radix_to_swp_entry(radswap));
633 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700634}
635
636/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800637 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800638 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800639 *
640 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
641 * as long as the inode doesn't go away and racy results are not a problem.
642 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800643unsigned long shmem_partial_swap_usage(struct address_space *mapping,
644 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800645{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800646 struct radix_tree_iter iter;
647 void **slot;
648 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800649 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800650
651 rcu_read_lock();
652
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800653 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
654 if (iter.index >= end)
655 break;
656
657 page = radix_tree_deref_slot(slot);
658
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700659 if (radix_tree_deref_retry(page)) {
660 slot = radix_tree_iter_retry(&iter);
661 continue;
662 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800663
664 if (radix_tree_exceptional_entry(page))
665 swapped++;
666
667 if (need_resched()) {
668 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700669 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800670 }
671 }
672
673 rcu_read_unlock();
674
675 return swapped << PAGE_SHIFT;
676}
677
678/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800679 * Determine (in bytes) how many of the shmem object's pages mapped by the
680 * given vma is swapped out.
681 *
682 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
683 * as long as the inode doesn't go away and racy results are not a problem.
684 */
685unsigned long shmem_swap_usage(struct vm_area_struct *vma)
686{
687 struct inode *inode = file_inode(vma->vm_file);
688 struct shmem_inode_info *info = SHMEM_I(inode);
689 struct address_space *mapping = inode->i_mapping;
690 unsigned long swapped;
691
692 /* Be careful as we don't hold info->lock */
693 swapped = READ_ONCE(info->swapped);
694
695 /*
696 * The easier cases are when the shmem object has nothing in swap, or
697 * the vma maps it whole. Then we can simply use the stats that we
698 * already track.
699 */
700 if (!swapped)
701 return 0;
702
703 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
704 return swapped << PAGE_SHIFT;
705
706 /* Here comes the more involved part */
707 return shmem_partial_swap_usage(mapping,
708 linear_page_index(vma, vma->vm_start),
709 linear_page_index(vma, vma->vm_end));
710}
711
712/*
Hugh Dickins24513262012-01-20 14:34:21 -0800713 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
714 */
715void shmem_unlock_mapping(struct address_space *mapping)
716{
717 struct pagevec pvec;
718 pgoff_t indices[PAGEVEC_SIZE];
719 pgoff_t index = 0;
720
721 pagevec_init(&pvec, 0);
722 /*
723 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
724 */
725 while (!mapping_unevictable(mapping)) {
726 /*
727 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
728 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
729 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700730 pvec.nr = find_get_entries(mapping, index,
731 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800732 if (!pvec.nr)
733 break;
734 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700735 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800736 check_move_unevictable_pages(pvec.pages, pvec.nr);
737 pagevec_release(&pvec);
738 cond_resched();
739 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700740}
741
742/*
743 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700744 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700745 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700746static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
747 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700749 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300751 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
752 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
753 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
754 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700755 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700756 pgoff_t indices[PAGEVEC_SIZE];
757 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700758 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700759 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700761 if (lend == -1)
762 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700763
764 pagevec_init(&pvec, 0);
765 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700766 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700767 pvec.nr = find_get_entries(mapping, index,
768 min(end - index, (pgoff_t)PAGEVEC_SIZE),
769 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700770 if (!pvec.nr)
771 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700772 for (i = 0; i < pagevec_count(&pvec); i++) {
773 struct page *page = pvec.pages[i];
774
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700775 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700776 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700777 break;
778
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700779 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700780 if (unfalloc)
781 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700782 nr_swaps_freed += !shmem_free_swap(mapping,
783 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700784 continue;
785 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700786
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700787 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
788
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700789 if (!trylock_page(page))
790 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700791
792 if (PageTransTail(page)) {
793 /* Middle of THP: zero out the page */
794 clear_highpage(page);
795 unlock_page(page);
796 continue;
797 } else if (PageTransHuge(page)) {
798 if (index == round_down(end, HPAGE_PMD_NR)) {
799 /*
800 * Range ends in the middle of THP:
801 * zero out the page
802 */
803 clear_highpage(page);
804 unlock_page(page);
805 continue;
806 }
807 index += HPAGE_PMD_NR - 1;
808 i += HPAGE_PMD_NR - 1;
809 }
810
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700811 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700812 VM_BUG_ON_PAGE(PageTail(page), page);
813 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800814 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700815 truncate_inode_page(mapping, page);
816 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700817 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700818 unlock_page(page);
819 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700820 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800821 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700822 cond_resched();
823 index++;
824 }
825
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700826 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700827 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700828 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700829 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300830 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700831 if (start > end) {
832 top = partial_end;
833 partial_end = 0;
834 }
835 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700836 set_page_dirty(page);
837 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300838 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700839 }
840 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700841 if (partial_end) {
842 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700843 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700844 if (page) {
845 zero_user_segment(page, 0, partial_end);
846 set_page_dirty(page);
847 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300848 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700849 }
850 }
851 if (start >= end)
852 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700853
854 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700855 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700856 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700857
858 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700859 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700860 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700861 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700862 /* If all gone or hole-punch or unfalloc, we're done */
863 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700864 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700865 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700866 index = start;
867 continue;
868 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700869 for (i = 0; i < pagevec_count(&pvec); i++) {
870 struct page *page = pvec.pages[i];
871
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700872 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700873 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700874 break;
875
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700876 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700877 if (unfalloc)
878 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700879 if (shmem_free_swap(mapping, index, page)) {
880 /* Swap was replaced by page: retry */
881 index--;
882 break;
883 }
884 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700885 continue;
886 }
887
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700888 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700889
890 if (PageTransTail(page)) {
891 /* Middle of THP: zero out the page */
892 clear_highpage(page);
893 unlock_page(page);
894 /*
895 * Partial thp truncate due 'start' in middle
896 * of THP: don't need to look on these pages
897 * again on !pvec.nr restart.
898 */
899 if (index != round_down(end, HPAGE_PMD_NR))
900 start++;
901 continue;
902 } else if (PageTransHuge(page)) {
903 if (index == round_down(end, HPAGE_PMD_NR)) {
904 /*
905 * Range ends in the middle of THP:
906 * zero out the page
907 */
908 clear_highpage(page);
909 unlock_page(page);
910 continue;
911 }
912 index += HPAGE_PMD_NR - 1;
913 i += HPAGE_PMD_NR - 1;
914 }
915
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700916 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700917 VM_BUG_ON_PAGE(PageTail(page), page);
918 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800919 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700920 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700921 } else {
922 /* Page was replaced by swap: retry */
923 unlock_page(page);
924 index--;
925 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700926 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700927 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700928 unlock_page(page);
929 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700930 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800931 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700932 index++;
933 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700934
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700935 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700936 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700938 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700939}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700941void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
942{
943 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700944 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700946EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Yu Zhao44a30222015-09-08 15:03:33 -0700948static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
949 struct kstat *stat)
950{
951 struct inode *inode = dentry->d_inode;
952 struct shmem_inode_info *info = SHMEM_I(inode);
953
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800954 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700955 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800956 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700957 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800958 }
Yu Zhao44a30222015-09-08 15:03:33 -0700959 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700960 return 0;
961}
962
Hugh Dickins94c1e622011-06-27 16:18:03 -0700963static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
David Howells75c3cfa2015-03-17 22:26:12 +0000965 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700966 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700967 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 int error;
969
Jan Kara31051c82016-05-26 16:55:18 +0200970 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200971 if (error)
972 return error;
973
Hugh Dickins94c1e622011-06-27 16:18:03 -0700974 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
975 loff_t oldsize = inode->i_size;
976 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000977
David Herrmann40e041a2014-08-08 14:25:27 -0700978 /* protected by i_mutex */
979 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
980 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
981 return -EPERM;
982
Hugh Dickins94c1e622011-06-27 16:18:03 -0700983 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700984 error = shmem_reacct_size(SHMEM_I(inode)->flags,
985 oldsize, newsize);
986 if (error)
987 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700988 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700989 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700990 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700991 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700992 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800993 if (oldsize > holebegin)
994 unmap_mapping_range(inode->i_mapping,
995 holebegin, 0, 1);
996 if (info->alloced)
997 shmem_truncate_range(inode,
998 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700999 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001000 if (oldsize > holebegin)
1001 unmap_mapping_range(inode->i_mapping,
1002 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001003
1004 /*
1005 * Part of the huge page can be beyond i_size: subject
1006 * to shrink under memory pressure.
1007 */
1008 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1009 spin_lock(&sbinfo->shrinklist_lock);
1010 if (list_empty(&info->shrinklist)) {
1011 list_add_tail(&info->shrinklist,
1012 &sbinfo->shrinklist);
1013 sbinfo->shrinklist_len++;
1014 }
1015 spin_unlock(&sbinfo->shrinklist_lock);
1016 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001017 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001020 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001021 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001022 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return error;
1024}
1025
Al Viro1f895f72010-06-05 19:10:41 -04001026static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001029 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001031 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 shmem_unacct_size(info->flags, inode->i_size);
1033 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001034 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001035 if (!list_empty(&info->shrinklist)) {
1036 spin_lock(&sbinfo->shrinklist_lock);
1037 if (!list_empty(&info->shrinklist)) {
1038 list_del_init(&info->shrinklist);
1039 sbinfo->shrinklist_len--;
1040 }
1041 spin_unlock(&sbinfo->shrinklist_lock);
1042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001044 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001046 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
Al Viro3ed47db2016-01-22 18:08:52 -05001048 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001049
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001050 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001051 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001052 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001053 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
1055
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001056/*
1057 * If swap found in inode, free it and move page from swapcache to filecache.
1058 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001059static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001060 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001062 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001063 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001064 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001065 gfp_t gfp;
1066 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001068 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001069 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001070 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001071 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001072
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001073 /*
1074 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001075 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001076 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001077 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001078 */
1079 if (shmem_swaplist.next != &info->swaplist)
1080 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001081
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001082 gfp = mapping_gfp_mask(mapping);
1083 if (shmem_should_replace_page(*pagep, gfp)) {
1084 mutex_unlock(&shmem_swaplist_mutex);
1085 error = shmem_replace_page(pagep, gfp, info, index);
1086 mutex_lock(&shmem_swaplist_mutex);
1087 /*
1088 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001089 * allocation, but the inode might have been freed while we
1090 * dropped it: although a racing shmem_evict_inode() cannot
1091 * complete without emptying the radix_tree, our page lock
1092 * on this swapcache page is not enough to prevent that -
1093 * free_swap_and_cache() of our swap entry will only
1094 * trylock_page(), removing swap from radix_tree whatever.
1095 *
1096 * We must not proceed to shmem_add_to_page_cache() if the
1097 * inode has been freed, but of course we cannot rely on
1098 * inode or mapping or info to check that. However, we can
1099 * safely check if our swap entry is still in use (and here
1100 * it can't have got reused for another page): if it's still
1101 * in use, then the inode cannot have been freed yet, and we
1102 * can safely proceed (if it's no longer in use, that tells
1103 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001104 */
1105 if (!page_swapcount(*pagep))
1106 error = -ENOENT;
1107 }
1108
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001109 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001110 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1111 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1112 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001113 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001114 if (!error)
1115 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001116 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001117 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001118 /*
1119 * Truncation and eviction use free_swap_and_cache(), which
1120 * only does trylock page: if we raced, best clean up here.
1121 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001122 delete_from_swap_cache(*pagep);
1123 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001124 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001125 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001126 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001127 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001128 swap_free(swap);
1129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001131 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132}
1133
1134/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001135 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001137int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001139 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001141 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001142 int error = 0;
1143
1144 /*
1145 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001146 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001147 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001148 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001149 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001150
1151 /*
1152 * Charge page using GFP_KERNEL while we can wait, before taking
1153 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1154 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001155 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001156 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1157 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001158 if (error)
1159 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001160 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001161 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001163 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001164 list_for_each_safe(this, next, &shmem_swaplist) {
1165 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001166 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001167 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001168 else
1169 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001170 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001171 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001172 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001173 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001175 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001176
Johannes Weiner00501b52014-08-08 14:19:20 -07001177 if (error) {
1178 if (error != -ENOMEM)
1179 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001180 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001181 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001182 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001183out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001184 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001185 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001186 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
1189/*
1190 * Move the page from the page cache to the swap cache.
1191 */
1192static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1193{
1194 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001197 swp_entry_t swap;
1198 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001200 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 mapping = page->mapping;
1203 index = page->index;
1204 inode = mapping->host;
1205 info = SHMEM_I(inode);
1206 if (info->flags & VM_LOCKED)
1207 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001208 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 goto redirty;
1210
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001211 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001212 * Our capabilities prevent regular writeback or sync from ever calling
1213 * shmem_writepage; but a stacking filesystem might use ->writepage of
1214 * its underlying filesystem, in which case tmpfs should write out to
1215 * swap only in response to memory pressure, and not for the writeback
1216 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001217 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001218 if (!wbc->for_reclaim) {
1219 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1220 goto redirty;
1221 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001222
1223 /*
1224 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1225 * value into swapfile.c, the only way we can correctly account for a
1226 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001227 *
1228 * That's okay for a page already fallocated earlier, but if we have
1229 * not yet completed the fallocation, then (a) we want to keep track
1230 * of this page in case we have to undo it, and (b) it may not be a
1231 * good idea to continue anyway, once we're pushing into swap. So
1232 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001233 */
1234 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001235 if (inode->i_private) {
1236 struct shmem_falloc *shmem_falloc;
1237 spin_lock(&inode->i_lock);
1238 shmem_falloc = inode->i_private;
1239 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001240 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001241 index >= shmem_falloc->start &&
1242 index < shmem_falloc->next)
1243 shmem_falloc->nr_unswapped++;
1244 else
1245 shmem_falloc = NULL;
1246 spin_unlock(&inode->i_lock);
1247 if (shmem_falloc)
1248 goto redirty;
1249 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001250 clear_highpage(page);
1251 flush_dcache_page(page);
1252 SetPageUptodate(page);
1253 }
1254
Hugh Dickins48f170f2011-07-25 17:12:37 -07001255 swap = get_swap_page();
1256 if (!swap.val)
1257 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001258
Vladimir Davydov37e84352016-01-20 15:02:56 -08001259 if (mem_cgroup_try_charge_swap(page, swap))
1260 goto free_swap;
1261
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001262 /*
1263 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001264 * if it's not already there. Do it now before the page is
1265 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001266 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001267 * we've incremented swapped, because shmem_unuse_inode() will
1268 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001269 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001270 mutex_lock(&shmem_swaplist_mutex);
1271 if (list_empty(&info->swaplist))
1272 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001273
Hugh Dickins48f170f2011-07-25 17:12:37 -07001274 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001275 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001276 shmem_recalc_inode(inode);
1277 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001278 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001279
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001280 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001281 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1282
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001283 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001284 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001285 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return 0;
1287 }
1288
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001289 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001290free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001291 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292redirty:
1293 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001294 if (wbc->for_reclaim)
1295 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1296 unlock_page(page);
1297 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
1299
Hugh Dickins75edd342016-05-19 17:12:44 -07001300#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001301static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001302{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001303 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001304
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001305 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001306 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001307
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001308 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001309
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001310 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001311}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001312
1313static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1314{
1315 struct mempolicy *mpol = NULL;
1316 if (sbinfo->mpol) {
1317 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1318 mpol = sbinfo->mpol;
1319 mpol_get(mpol);
1320 spin_unlock(&sbinfo->stat_lock);
1321 }
1322 return mpol;
1323}
Hugh Dickins75edd342016-05-19 17:12:44 -07001324#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1325static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1326{
1327}
1328static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1329{
1330 return NULL;
1331}
1332#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1333#ifndef CONFIG_NUMA
1334#define vm_policy vm_private_data
1335#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001336
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001337static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1338 struct shmem_inode_info *info, pgoff_t index)
1339{
1340 /* Create a pseudo vma that just contains the policy */
1341 vma->vm_start = 0;
1342 /* Bias interleave by inode number to distribute better across nodes */
1343 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1344 vma->vm_ops = NULL;
1345 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1346}
1347
1348static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1349{
1350 /* Drop reference taken by mpol_shared_policy_lookup() */
1351 mpol_cond_put(vma->vm_policy);
1352}
1353
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001354static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1355 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001358 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001360 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001361 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001362 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001363
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001364 return page;
1365}
Mel Gorman18a2f372012-12-05 14:01:41 -08001366
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001367static struct page *shmem_alloc_hugepage(gfp_t gfp,
1368 struct shmem_inode_info *info, pgoff_t index)
1369{
1370 struct vm_area_struct pvma;
1371 struct inode *inode = &info->vfs_inode;
1372 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001373 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001374 void __rcu **results;
1375 struct page *page;
1376
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001377 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001378 return NULL;
1379
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001380 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001381 rcu_read_lock();
1382 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1383 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1384 rcu_read_unlock();
1385 return NULL;
1386 }
1387 rcu_read_unlock();
1388
1389 shmem_pseudo_vma_init(&pvma, info, hindex);
1390 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1391 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1392 shmem_pseudo_vma_destroy(&pvma);
1393 if (page)
1394 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001395 return page;
1396}
1397
1398static struct page *shmem_alloc_page(gfp_t gfp,
1399 struct shmem_inode_info *info, pgoff_t index)
1400{
1401 struct vm_area_struct pvma;
1402 struct page *page;
1403
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001404 shmem_pseudo_vma_init(&pvma, info, index);
1405 page = alloc_page_vma(gfp, &pvma, 0);
1406 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001407
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001408 return page;
1409}
1410
1411static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1412 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1413 pgoff_t index, bool huge)
1414{
1415 struct page *page;
1416 int nr;
1417 int err = -ENOSPC;
1418
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001419 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001420 huge = false;
1421 nr = huge ? HPAGE_PMD_NR : 1;
1422
1423 if (shmem_acct_block(info->flags, nr))
1424 goto failed;
1425 if (sbinfo->max_blocks) {
1426 if (percpu_counter_compare(&sbinfo->used_blocks,
1427 sbinfo->max_blocks - nr) > 0)
1428 goto unacct;
1429 percpu_counter_add(&sbinfo->used_blocks, nr);
1430 }
1431
1432 if (huge)
1433 page = shmem_alloc_hugepage(gfp, info, index);
1434 else
1435 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001436 if (page) {
1437 __SetPageLocked(page);
1438 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001439 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001440 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001441
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001442 err = -ENOMEM;
1443 if (sbinfo->max_blocks)
1444 percpu_counter_add(&sbinfo->used_blocks, -nr);
1445unacct:
1446 shmem_unacct_blocks(info->flags, nr);
1447failed:
1448 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001450
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001452 * When a page is moved from swapcache to shmem filecache (either by the
1453 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1454 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1455 * ignorance of the mapping it belongs to. If that mapping has special
1456 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1457 * we may need to copy to a suitable page before moving to filecache.
1458 *
1459 * In a future release, this may well be extended to respect cpuset and
1460 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1461 * but for now it is a simple matter of zone.
1462 */
1463static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1464{
1465 return page_zonenum(page) > gfp_zone(gfp);
1466}
1467
1468static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1469 struct shmem_inode_info *info, pgoff_t index)
1470{
1471 struct page *oldpage, *newpage;
1472 struct address_space *swap_mapping;
1473 pgoff_t swap_index;
1474 int error;
1475
1476 oldpage = *pagep;
1477 swap_index = page_private(oldpage);
1478 swap_mapping = page_mapping(oldpage);
1479
1480 /*
1481 * We have arrived here because our zones are constrained, so don't
1482 * limit chance of success by further cpuset and node constraints.
1483 */
1484 gfp &= ~GFP_CONSTRAINT_MASK;
1485 newpage = shmem_alloc_page(gfp, info, index);
1486 if (!newpage)
1487 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001488
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001489 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001490 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001491 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001492
Hugh Dickins9956edf2016-11-10 10:46:11 -08001493 __SetPageLocked(newpage);
1494 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001495 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001496 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001497 SetPageSwapCache(newpage);
1498
1499 /*
1500 * Our caller will very soon move newpage out of swapcache, but it's
1501 * a nice clean interface for us to replace oldpage by newpage there.
1502 */
1503 spin_lock_irq(&swap_mapping->tree_lock);
1504 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1505 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001506 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001507 __inc_node_page_state(newpage, NR_FILE_PAGES);
1508 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001509 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001510 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001511
Hugh Dickins0142ef62012-06-07 14:21:09 -07001512 if (unlikely(error)) {
1513 /*
1514 * Is this possible? I think not, now that our callers check
1515 * both PageSwapCache and page_private after getting page lock;
1516 * but be defensive. Reverse old to newpage for clear and free.
1517 */
1518 oldpage = newpage;
1519 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001520 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001521 lru_cache_add_anon(newpage);
1522 *pagep = newpage;
1523 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001524
1525 ClearPageSwapCache(oldpage);
1526 set_page_private(oldpage, 0);
1527
1528 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001529 put_page(oldpage);
1530 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001531 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001532}
1533
1534/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001535 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 *
1537 * If we allocate a new one we do not mark it dirty. That's up to the
1538 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001539 * entry since a page cannot live in both the swap and page cache.
1540 *
1541 * fault_mm and fault_type are only supplied by shmem_fault:
1542 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001544static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001545 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1546 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
1548 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001549 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001551 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001552 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001553 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001555 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001556 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001558 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001559 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001561 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001563 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1564 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001566 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001567 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001568 if (radix_tree_exceptional_entry(page)) {
1569 swap = radix_to_swp_entry(page);
1570 page = NULL;
1571 }
1572
Hugh Dickins75edd342016-05-19 17:12:44 -07001573 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001574 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001575 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001576 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001577 }
1578
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001579 if (page && sgp == SGP_WRITE)
1580 mark_page_accessed(page);
1581
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001582 /* fallocated page? */
1583 if (page && !PageUptodate(page)) {
1584 if (sgp != SGP_READ)
1585 goto clear;
1586 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001587 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001588 page = NULL;
1589 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001590 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001591 *pagep = page;
1592 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001596 * Fast cache lookup did not find it:
1597 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001599 info = SHMEM_I(inode);
1600 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001601 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 if (swap.val) {
1604 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001605 page = lookup_swap_cache(swap);
1606 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001607 /* Or update major stats only when swapin succeeds?? */
1608 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001609 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001610 count_vm_event(PGMAJFAULT);
1611 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1612 }
1613 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001614 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001615 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001616 error = -ENOMEM;
1617 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 }
1620
1621 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001622 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001623 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001624 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001625 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001626 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001627 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001628 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001630 goto failed;
1631 }
1632 wait_on_page_writeback(page);
1633
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001634 if (shmem_should_replace_page(page, gfp)) {
1635 error = shmem_replace_page(&page, gfp, info, index);
1636 if (error)
1637 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001640 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001641 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001642 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001643 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001644 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001645 /*
1646 * We already confirmed swap under page lock, and make
1647 * no memory allocation here, so usually no possibility
1648 * of error; but free_swap_and_cache() only trylocks a
1649 * page, so it is just possible that the entry has been
1650 * truncated or holepunched since swap was confirmed.
1651 * shmem_undo_range() will have done some of the
1652 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001653 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001654 * Reset swap.val? No, leave it so "failed" goes back to
1655 * "repeat": reading a hole and writing should succeed.
1656 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001657 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001658 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001659 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001660 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001661 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001662 if (error)
1663 goto failed;
1664
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001665 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001666
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001667 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001668 info->swapped--;
1669 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001670 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001671
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001672 if (sgp == SGP_WRITE)
1673 mark_page_accessed(page);
1674
Hugh Dickins27ab7002011-07-25 17:12:36 -07001675 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001676 set_page_dirty(page);
1677 swap_free(swap);
1678
Hugh Dickins54af60422011-08-03 16:21:24 -07001679 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001680 /* shmem_symlink() */
1681 if (mapping->a_ops != &shmem_aops)
1682 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001683 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001684 goto alloc_nohuge;
1685 if (shmem_huge == SHMEM_HUGE_FORCE)
1686 goto alloc_huge;
1687 switch (sbinfo->huge) {
1688 loff_t i_size;
1689 pgoff_t off;
1690 case SHMEM_HUGE_NEVER:
1691 goto alloc_nohuge;
1692 case SHMEM_HUGE_WITHIN_SIZE:
1693 off = round_up(index, HPAGE_PMD_NR);
1694 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1695 if (i_size >= HPAGE_PMD_SIZE &&
1696 i_size >> PAGE_SHIFT >= off)
1697 goto alloc_huge;
1698 /* fallthrough */
1699 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001700 if (sgp_huge == SGP_HUGE)
1701 goto alloc_huge;
1702 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001703 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001704 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001706alloc_huge:
1707 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1708 index, true);
1709 if (IS_ERR(page)) {
1710alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1711 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001712 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001713 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001714 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001715 error = PTR_ERR(page);
1716 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001717 if (error != -ENOSPC)
1718 goto failed;
1719 /*
1720 * Try to reclaim some spece by splitting a huge page
1721 * beyond i_size on the filesystem.
1722 */
1723 while (retry--) {
1724 int ret;
1725 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1726 if (ret == SHRINK_STOP)
1727 break;
1728 if (ret)
1729 goto alloc_nohuge;
1730 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001731 goto failed;
1732 }
1733
1734 if (PageTransHuge(page))
1735 hindex = round_down(index, HPAGE_PMD_NR);
1736 else
1737 hindex = index;
1738
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001739 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001740 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001741
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001742 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001743 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001744 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001745 goto unacct;
1746 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1747 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001748 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001749 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001750 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001751 radix_tree_preload_end();
1752 }
1753 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001754 mem_cgroup_cancel_charge(page, memcg,
1755 PageTransHuge(page));
1756 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001757 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001758 mem_cgroup_commit_charge(page, memcg, false,
1759 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001760 lru_cache_add_anon(page);
1761
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001762 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001763 info->alloced += 1 << compound_order(page);
1764 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001765 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001766 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001767 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001768
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001769 if (PageTransHuge(page) &&
1770 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1771 hindex + HPAGE_PMD_NR - 1) {
1772 /*
1773 * Part of the huge page is beyond i_size: subject
1774 * to shrink under memory pressure.
1775 */
1776 spin_lock(&sbinfo->shrinklist_lock);
1777 if (list_empty(&info->shrinklist)) {
1778 list_add_tail(&info->shrinklist,
1779 &sbinfo->shrinklist);
1780 sbinfo->shrinklist_len++;
1781 }
1782 spin_unlock(&sbinfo->shrinklist_lock);
1783 }
1784
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001785 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001786 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1787 */
1788 if (sgp == SGP_FALLOC)
1789 sgp = SGP_WRITE;
1790clear:
1791 /*
1792 * Let SGP_WRITE caller clear ends if write does not fill page;
1793 * but SGP_FALLOC on a page fallocated earlier must initialize
1794 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001795 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001796 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1797 struct page *head = compound_head(page);
1798 int i;
1799
1800 for (i = 0; i < (1 << compound_order(head)); i++) {
1801 clear_highpage(head + i);
1802 flush_dcache_page(head + i);
1803 }
1804 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001805 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001806 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001807
Hugh Dickins54af60422011-08-03 16:21:24 -07001808 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001809 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001810 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001811 if (alloced) {
1812 ClearPageDirty(page);
1813 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001814 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001815 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001816 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001817 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001818 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001819 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001820 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001821 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001822 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001823
Nick Piggind0217ac2007-07-19 01:47:03 -07001824 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001825 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001827unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001828 if (sbinfo->max_blocks)
1829 percpu_counter_sub(&sbinfo->used_blocks,
1830 1 << compound_order(page));
1831 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1832
1833 if (PageTransHuge(page)) {
1834 unlock_page(page);
1835 put_page(page);
1836 goto alloc_nohuge;
1837 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001838failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001839 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001840 error = -EEXIST;
1841unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001842 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001843 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001844 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001845 }
1846 if (error == -ENOSPC && !once++) {
1847 info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001848 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001849 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001850 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001851 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001852 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001853 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001854 goto repeat;
1855 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001858/*
1859 * This is like autoremove_wake_function, but it removes the wait queue
1860 * entry unconditionally - even if something else had already woken the
1861 * target.
1862 */
1863static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1864{
1865 int ret = default_wake_function(wait, mode, sync, key);
1866 list_del_init(&wait->task_list);
1867 return ret;
1868}
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1871{
Al Viro496ad9a2013-01-23 17:07:38 -05001872 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001873 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001874 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001876 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001878 /*
1879 * Trinity finds that probing a hole which tmpfs is punching can
1880 * prevent the hole-punch from ever completing: which in turn
1881 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001882 * faulting pages into the hole while it's being punched. Although
1883 * shmem_undo_range() does remove the additions, it may be unable to
1884 * keep up, as each new page needs its own unmap_mapping_range() call,
1885 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1886 *
1887 * It does not matter if we sometimes reach this check just before the
1888 * hole-punch begins, so that one fault then races with the punch:
1889 * we just need to make racing faults a rare case.
1890 *
1891 * The implementation below would be much simpler if we just used a
1892 * standard mutex or completion: but we cannot take i_mutex in fault,
1893 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001894 */
1895 if (unlikely(inode->i_private)) {
1896 struct shmem_falloc *shmem_falloc;
1897
1898 spin_lock(&inode->i_lock);
1899 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001900 if (shmem_falloc &&
1901 shmem_falloc->waitq &&
1902 vmf->pgoff >= shmem_falloc->start &&
1903 vmf->pgoff < shmem_falloc->next) {
1904 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001905 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001906
1907 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001908 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1909 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001910 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001911 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001912 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001913 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001914
1915 shmem_falloc_waitq = shmem_falloc->waitq;
1916 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1917 TASK_UNINTERRUPTIBLE);
1918 spin_unlock(&inode->i_lock);
1919 schedule();
1920
1921 /*
1922 * shmem_falloc_waitq points into the shmem_fallocate()
1923 * stack of the hole-punching task: shmem_falloc_waitq
1924 * is usually invalid by the time we reach here, but
1925 * finish_wait() does not dereference it in that case;
1926 * though i_lock needed lest racing with wake_up_all().
1927 */
1928 spin_lock(&inode->i_lock);
1929 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1930 spin_unlock(&inode->i_lock);
1931 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001932 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001933 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001934 }
1935
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001936 sgp = SGP_CACHE;
1937 if (vma->vm_flags & VM_HUGEPAGE)
1938 sgp = SGP_HUGE;
1939 else if (vma->vm_flags & VM_NOHUGEPAGE)
1940 sgp = SGP_NOHUGE;
1941
1942 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001943 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (error)
1945 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001946 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947}
1948
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001949unsigned long shmem_get_unmapped_area(struct file *file,
1950 unsigned long uaddr, unsigned long len,
1951 unsigned long pgoff, unsigned long flags)
1952{
1953 unsigned long (*get_area)(struct file *,
1954 unsigned long, unsigned long, unsigned long, unsigned long);
1955 unsigned long addr;
1956 unsigned long offset;
1957 unsigned long inflated_len;
1958 unsigned long inflated_addr;
1959 unsigned long inflated_offset;
1960
1961 if (len > TASK_SIZE)
1962 return -ENOMEM;
1963
1964 get_area = current->mm->get_unmapped_area;
1965 addr = get_area(file, uaddr, len, pgoff, flags);
1966
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001967 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001968 return addr;
1969 if (IS_ERR_VALUE(addr))
1970 return addr;
1971 if (addr & ~PAGE_MASK)
1972 return addr;
1973 if (addr > TASK_SIZE - len)
1974 return addr;
1975
1976 if (shmem_huge == SHMEM_HUGE_DENY)
1977 return addr;
1978 if (len < HPAGE_PMD_SIZE)
1979 return addr;
1980 if (flags & MAP_FIXED)
1981 return addr;
1982 /*
1983 * Our priority is to support MAP_SHARED mapped hugely;
1984 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1985 * But if caller specified an address hint, respect that as before.
1986 */
1987 if (uaddr)
1988 return addr;
1989
1990 if (shmem_huge != SHMEM_HUGE_FORCE) {
1991 struct super_block *sb;
1992
1993 if (file) {
1994 VM_BUG_ON(file->f_op != &shmem_file_operations);
1995 sb = file_inode(file)->i_sb;
1996 } else {
1997 /*
1998 * Called directly from mm/mmap.c, or drivers/char/mem.c
1999 * for "/dev/zero", to create a shared anonymous object.
2000 */
2001 if (IS_ERR(shm_mnt))
2002 return addr;
2003 sb = shm_mnt->mnt_sb;
2004 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002005 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002006 return addr;
2007 }
2008
2009 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2010 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2011 return addr;
2012 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2013 return addr;
2014
2015 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2016 if (inflated_len > TASK_SIZE)
2017 return addr;
2018 if (inflated_len < len)
2019 return addr;
2020
2021 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2022 if (IS_ERR_VALUE(inflated_addr))
2023 return addr;
2024 if (inflated_addr & ~PAGE_MASK)
2025 return addr;
2026
2027 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2028 inflated_addr += offset - inflated_offset;
2029 if (inflated_offset > offset)
2030 inflated_addr += HPAGE_PMD_SIZE;
2031
2032 if (inflated_addr > TASK_SIZE - len)
2033 return addr;
2034 return inflated_addr;
2035}
2036
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002038static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
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 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042}
2043
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002044static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2045 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046{
Al Viro496ad9a2013-01-23 17:07:38 -05002047 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002048 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002050 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2051 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052}
2053#endif
2054
2055int shmem_lock(struct file *file, int lock, struct user_struct *user)
2056{
Al Viro496ad9a2013-01-23 17:07:38 -05002057 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 struct shmem_inode_info *info = SHMEM_I(inode);
2059 int retval = -ENOMEM;
2060
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002061 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (lock && !(info->flags & VM_LOCKED)) {
2063 if (!user_shm_lock(inode->i_size, user))
2064 goto out_nomem;
2065 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002066 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 }
2068 if (!lock && (info->flags & VM_LOCKED) && user) {
2069 user_shm_unlock(inode->i_size, user);
2070 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002071 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002076 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 return retval;
2078}
2079
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002080static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081{
2082 file_accessed(file);
2083 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002084 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002085 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2086 (vma->vm_end & HPAGE_PMD_MASK)) {
2087 khugepaged_enter(vma, vma->vm_flags);
2088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 return 0;
2090}
2091
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002092static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002093 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094{
2095 struct inode *inode;
2096 struct shmem_inode_info *info;
2097 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2098
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002099 if (shmem_reserve_inode(sb))
2100 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
2102 inode = new_inode(sb);
2103 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002104 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002105 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002107 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002108 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 info = SHMEM_I(inode);
2110 memset(info, 0, (char *)inode - (char *)info);
2111 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002112 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002113 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002114 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002116 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002117 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 switch (mode & S_IFMT) {
2120 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002121 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 init_special_inode(inode, mode, dev);
2123 break;
2124 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002125 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 inode->i_op = &shmem_inode_operations;
2127 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002128 mpol_shared_policy_init(&info->policy,
2129 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 break;
2131 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002132 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 /* Some things misbehave if size == 0 on a directory */
2134 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2135 inode->i_op = &shmem_dir_inode_operations;
2136 inode->i_fop = &simple_dir_operations;
2137 break;
2138 case S_IFLNK:
2139 /*
2140 * Must not load anything in the rbtree,
2141 * mpol_free_shared_policy will not be called.
2142 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002143 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 break;
2145 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002146 } else
2147 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 return inode;
2149}
2150
Johannes Weiner0cd61442014-04-03 14:47:46 -07002151bool shmem_mapping(struct address_space *mapping)
2152{
Sasha Levinf0774d82015-02-23 05:38:00 -05002153 if (!mapping->host)
2154 return false;
2155
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002156 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002157}
2158
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002160static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002161static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002163#ifdef CONFIG_TMPFS_XATTR
2164static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2165#else
2166#define shmem_initxattrs NULL
2167#endif
2168
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002170shmem_write_begin(struct file *file, struct address_space *mapping,
2171 loff_t pos, unsigned len, unsigned flags,
2172 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173{
Nick Piggin800d15a2007-10-16 01:25:03 -07002174 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002175 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002176 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002177
2178 /* i_mutex is held by caller */
2179 if (unlikely(info->seals)) {
2180 if (info->seals & F_SEAL_WRITE)
2181 return -EPERM;
2182 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2183 return -EPERM;
2184 }
2185
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002186 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002187}
2188
2189static int
2190shmem_write_end(struct file *file, struct address_space *mapping,
2191 loff_t pos, unsigned len, unsigned copied,
2192 struct page *page, void *fsdata)
2193{
2194 struct inode *inode = mapping->host;
2195
Hugh Dickinsd3602442008-02-04 22:28:44 -08002196 if (pos + copied > inode->i_size)
2197 i_size_write(inode, pos + copied);
2198
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002199 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002200 struct page *head = compound_head(page);
2201 if (PageTransCompound(page)) {
2202 int i;
2203
2204 for (i = 0; i < HPAGE_PMD_NR; i++) {
2205 if (head + i == page)
2206 continue;
2207 clear_highpage(head + i);
2208 flush_dcache_page(head + i);
2209 }
2210 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002211 if (copied < PAGE_SIZE) {
2212 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002213 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002214 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002215 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002216 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002217 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002218 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002219 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002220 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002221
Nick Piggin800d15a2007-10-16 01:25:03 -07002222 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223}
2224
Al Viro2ba5bbe2014-04-02 20:00:02 -04002225static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226{
Al Viro6e58e792014-02-03 17:07:03 -05002227 struct file *file = iocb->ki_filp;
2228 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002230 pgoff_t index;
2231 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002232 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002233 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002234 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002235 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002236
2237 /*
2238 * Might this read be for a stacking filesystem? Then when reading
2239 * holes of a sparse file, we actually need to allocate those pages,
2240 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2241 */
Al Viro777eda22014-12-17 04:46:46 -05002242 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002243 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002245 index = *ppos >> PAGE_SHIFT;
2246 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 for (;;) {
2249 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002250 pgoff_t end_index;
2251 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 loff_t i_size = i_size_read(inode);
2253
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002254 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (index > end_index)
2256 break;
2257 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002258 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 if (nr <= offset)
2260 break;
2261 }
2262
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002263 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002264 if (error) {
2265 if (error == -EINVAL)
2266 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 break;
2268 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002269 if (page) {
2270 if (sgp == SGP_CACHE)
2271 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002272 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
2275 /*
2276 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002277 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002279 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002281 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002283 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 if (nr <= offset) {
2285 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002286 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 break;
2288 }
2289 }
2290 nr -= offset;
2291
2292 if (page) {
2293 /*
2294 * If users can be writing to this page using arbitrary
2295 * virtual addresses, take care about potential aliasing
2296 * before reading the page on the kernel side.
2297 */
2298 if (mapping_writably_mapped(mapping))
2299 flush_dcache_page(page);
2300 /*
2301 * Mark the page accessed if we read the beginning.
2302 */
2303 if (!offset)
2304 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002305 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002307 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309
2310 /*
2311 * Ok, we have the page, and it's up-to-date, so
2312 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002314 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002315 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002317 index += offset >> PAGE_SHIFT;
2318 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002320 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002321 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 break;
Al Viro6e58e792014-02-03 17:07:03 -05002323 if (ret < nr) {
2324 error = -EFAULT;
2325 break;
2326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 cond_resched();
2328 }
2329
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002330 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002331 file_accessed(file);
2332 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333}
2334
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002335/*
2336 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2337 */
2338static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002339 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002340{
2341 struct page *page;
2342 struct pagevec pvec;
2343 pgoff_t indices[PAGEVEC_SIZE];
2344 bool done = false;
2345 int i;
2346
2347 pagevec_init(&pvec, 0);
2348 pvec.nr = 1; /* start small: we may be there already */
2349 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002350 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002351 pvec.nr, pvec.pages, indices);
2352 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002353 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002354 index = end;
2355 break;
2356 }
2357 for (i = 0; i < pvec.nr; i++, index++) {
2358 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002359 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002360 done = true;
2361 break;
2362 }
2363 index = indices[i];
2364 }
2365 page = pvec.pages[i];
2366 if (page && !radix_tree_exceptional_entry(page)) {
2367 if (!PageUptodate(page))
2368 page = NULL;
2369 }
2370 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002371 (page && whence == SEEK_DATA) ||
2372 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002373 done = true;
2374 break;
2375 }
2376 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002377 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002378 pagevec_release(&pvec);
2379 pvec.nr = PAGEVEC_SIZE;
2380 cond_resched();
2381 }
2382 return index;
2383}
2384
Andrew Morton965c8e52012-12-17 15:59:39 -08002385static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002386{
2387 struct address_space *mapping = file->f_mapping;
2388 struct inode *inode = mapping->host;
2389 pgoff_t start, end;
2390 loff_t new_offset;
2391
Andrew Morton965c8e52012-12-17 15:59:39 -08002392 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2393 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002394 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002395 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002396 /* We're holding i_mutex so we can access i_size directly */
2397
2398 if (offset < 0)
2399 offset = -EINVAL;
2400 else if (offset >= inode->i_size)
2401 offset = -ENXIO;
2402 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002403 start = offset >> PAGE_SHIFT;
2404 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002405 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002406 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002407 if (new_offset > offset) {
2408 if (new_offset < inode->i_size)
2409 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002410 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002411 offset = -ENXIO;
2412 else
2413 offset = inode->i_size;
2414 }
2415 }
2416
Hugh Dickins387aae62013-08-04 11:30:25 -07002417 if (offset >= 0)
2418 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002419 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002420 return offset;
2421}
2422
David Herrmann05f65b52014-08-08 14:25:36 -07002423/*
2424 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2425 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2426 */
2427#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2428#define LAST_SCAN 4 /* about 150ms max */
2429
2430static void shmem_tag_pins(struct address_space *mapping)
2431{
2432 struct radix_tree_iter iter;
2433 void **slot;
2434 pgoff_t start;
2435 struct page *page;
2436
2437 lru_add_drain();
2438 start = 0;
2439 rcu_read_lock();
2440
David Herrmann05f65b52014-08-08 14:25:36 -07002441 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2442 page = radix_tree_deref_slot(slot);
2443 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002444 if (radix_tree_deref_retry(page)) {
2445 slot = radix_tree_iter_retry(&iter);
2446 continue;
2447 }
David Herrmann05f65b52014-08-08 14:25:36 -07002448 } else if (page_count(page) - page_mapcount(page) > 1) {
2449 spin_lock_irq(&mapping->tree_lock);
2450 radix_tree_tag_set(&mapping->page_tree, iter.index,
2451 SHMEM_TAG_PINNED);
2452 spin_unlock_irq(&mapping->tree_lock);
2453 }
2454
2455 if (need_resched()) {
2456 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002457 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002458 }
2459 }
2460 rcu_read_unlock();
2461}
2462
2463/*
2464 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2465 * via get_user_pages(), drivers might have some pending I/O without any active
2466 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2467 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2468 * them to be dropped.
2469 * The caller must guarantee that no new user will acquire writable references
2470 * to those pages to avoid races.
2471 */
David Herrmann40e041a2014-08-08 14:25:27 -07002472static int shmem_wait_for_pins(struct address_space *mapping)
2473{
David Herrmann05f65b52014-08-08 14:25:36 -07002474 struct radix_tree_iter iter;
2475 void **slot;
2476 pgoff_t start;
2477 struct page *page;
2478 int error, scan;
2479
2480 shmem_tag_pins(mapping);
2481
2482 error = 0;
2483 for (scan = 0; scan <= LAST_SCAN; scan++) {
2484 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2485 break;
2486
2487 if (!scan)
2488 lru_add_drain_all();
2489 else if (schedule_timeout_killable((HZ << scan) / 200))
2490 scan = LAST_SCAN;
2491
2492 start = 0;
2493 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002494 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2495 start, SHMEM_TAG_PINNED) {
2496
2497 page = radix_tree_deref_slot(slot);
2498 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002499 if (radix_tree_deref_retry(page)) {
2500 slot = radix_tree_iter_retry(&iter);
2501 continue;
2502 }
David Herrmann05f65b52014-08-08 14:25:36 -07002503
2504 page = NULL;
2505 }
2506
2507 if (page &&
2508 page_count(page) - page_mapcount(page) != 1) {
2509 if (scan < LAST_SCAN)
2510 goto continue_resched;
2511
2512 /*
2513 * On the last scan, we clean up all those tags
2514 * we inserted; but make a note that we still
2515 * found pages pinned.
2516 */
2517 error = -EBUSY;
2518 }
2519
2520 spin_lock_irq(&mapping->tree_lock);
2521 radix_tree_tag_clear(&mapping->page_tree,
2522 iter.index, SHMEM_TAG_PINNED);
2523 spin_unlock_irq(&mapping->tree_lock);
2524continue_resched:
2525 if (need_resched()) {
2526 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002527 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002528 }
2529 }
2530 rcu_read_unlock();
2531 }
2532
2533 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002534}
2535
2536#define F_ALL_SEALS (F_SEAL_SEAL | \
2537 F_SEAL_SHRINK | \
2538 F_SEAL_GROW | \
2539 F_SEAL_WRITE)
2540
2541int shmem_add_seals(struct file *file, unsigned int seals)
2542{
2543 struct inode *inode = file_inode(file);
2544 struct shmem_inode_info *info = SHMEM_I(inode);
2545 int error;
2546
2547 /*
2548 * SEALING
2549 * Sealing allows multiple parties to share a shmem-file but restrict
2550 * access to a specific subset of file operations. Seals can only be
2551 * added, but never removed. This way, mutually untrusted parties can
2552 * share common memory regions with a well-defined policy. A malicious
2553 * peer can thus never perform unwanted operations on a shared object.
2554 *
2555 * Seals are only supported on special shmem-files and always affect
2556 * the whole underlying inode. Once a seal is set, it may prevent some
2557 * kinds of access to the file. Currently, the following seals are
2558 * defined:
2559 * SEAL_SEAL: Prevent further seals from being set on this file
2560 * SEAL_SHRINK: Prevent the file from shrinking
2561 * SEAL_GROW: Prevent the file from growing
2562 * SEAL_WRITE: Prevent write access to the file
2563 *
2564 * As we don't require any trust relationship between two parties, we
2565 * must prevent seals from being removed. Therefore, sealing a file
2566 * only adds a given set of seals to the file, it never touches
2567 * existing seals. Furthermore, the "setting seals"-operation can be
2568 * sealed itself, which basically prevents any further seal from being
2569 * added.
2570 *
2571 * Semantics of sealing are only defined on volatile files. Only
2572 * anonymous shmem files support sealing. More importantly, seals are
2573 * never written to disk. Therefore, there's no plan to support it on
2574 * other file types.
2575 */
2576
2577 if (file->f_op != &shmem_file_operations)
2578 return -EINVAL;
2579 if (!(file->f_mode & FMODE_WRITE))
2580 return -EPERM;
2581 if (seals & ~(unsigned int)F_ALL_SEALS)
2582 return -EINVAL;
2583
Al Viro59551022016-01-22 15:40:57 -05002584 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002585
2586 if (info->seals & F_SEAL_SEAL) {
2587 error = -EPERM;
2588 goto unlock;
2589 }
2590
2591 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2592 error = mapping_deny_writable(file->f_mapping);
2593 if (error)
2594 goto unlock;
2595
2596 error = shmem_wait_for_pins(file->f_mapping);
2597 if (error) {
2598 mapping_allow_writable(file->f_mapping);
2599 goto unlock;
2600 }
2601 }
2602
2603 info->seals |= seals;
2604 error = 0;
2605
2606unlock:
Al Viro59551022016-01-22 15:40:57 -05002607 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002608 return error;
2609}
2610EXPORT_SYMBOL_GPL(shmem_add_seals);
2611
2612int shmem_get_seals(struct file *file)
2613{
2614 if (file->f_op != &shmem_file_operations)
2615 return -EINVAL;
2616
2617 return SHMEM_I(file_inode(file))->seals;
2618}
2619EXPORT_SYMBOL_GPL(shmem_get_seals);
2620
2621long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2622{
2623 long error;
2624
2625 switch (cmd) {
2626 case F_ADD_SEALS:
2627 /* disallow upper 32bit */
2628 if (arg > UINT_MAX)
2629 return -EINVAL;
2630
2631 error = shmem_add_seals(file, arg);
2632 break;
2633 case F_GET_SEALS:
2634 error = shmem_get_seals(file);
2635 break;
2636 default:
2637 error = -EINVAL;
2638 break;
2639 }
2640
2641 return error;
2642}
2643
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002644static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2645 loff_t len)
2646{
Al Viro496ad9a2013-01-23 17:07:38 -05002647 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002648 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002649 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002650 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002651 pgoff_t start, index, end;
2652 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002653
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002654 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2655 return -EOPNOTSUPP;
2656
Al Viro59551022016-01-22 15:40:57 -05002657 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002658
2659 if (mode & FALLOC_FL_PUNCH_HOLE) {
2660 struct address_space *mapping = file->f_mapping;
2661 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2662 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002663 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002664
David Herrmann40e041a2014-08-08 14:25:27 -07002665 /* protected by i_mutex */
2666 if (info->seals & F_SEAL_WRITE) {
2667 error = -EPERM;
2668 goto out;
2669 }
2670
Hugh Dickins8e205f72014-07-23 14:00:10 -07002671 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002672 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2673 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2674 spin_lock(&inode->i_lock);
2675 inode->i_private = &shmem_falloc;
2676 spin_unlock(&inode->i_lock);
2677
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002678 if ((u64)unmap_end > (u64)unmap_start)
2679 unmap_mapping_range(mapping, unmap_start,
2680 1 + unmap_end - unmap_start, 0);
2681 shmem_truncate_range(inode, offset, offset + len - 1);
2682 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002683
2684 spin_lock(&inode->i_lock);
2685 inode->i_private = NULL;
2686 wake_up_all(&shmem_falloc_waitq);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002687 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002688 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002689 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002690 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002691 }
2692
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002693 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2694 error = inode_newsize_ok(inode, offset + len);
2695 if (error)
2696 goto out;
2697
David Herrmann40e041a2014-08-08 14:25:27 -07002698 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2699 error = -EPERM;
2700 goto out;
2701 }
2702
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002703 start = offset >> PAGE_SHIFT;
2704 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002705 /* Try to avoid a swapstorm if len is impossible to satisfy */
2706 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2707 error = -ENOSPC;
2708 goto out;
2709 }
2710
Hugh Dickins8e205f72014-07-23 14:00:10 -07002711 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002712 shmem_falloc.start = start;
2713 shmem_falloc.next = start;
2714 shmem_falloc.nr_falloced = 0;
2715 shmem_falloc.nr_unswapped = 0;
2716 spin_lock(&inode->i_lock);
2717 inode->i_private = &shmem_falloc;
2718 spin_unlock(&inode->i_lock);
2719
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002720 for (index = start; index < end; index++) {
2721 struct page *page;
2722
2723 /*
2724 * Good, the fallocate(2) manpage permits EINTR: we may have
2725 * been interrupted because we are using up too much memory.
2726 */
2727 if (signal_pending(current))
2728 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002729 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2730 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002731 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002732 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002733 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002734 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002735 if (index > start) {
2736 shmem_undo_range(inode,
2737 (loff_t)start << PAGE_SHIFT,
2738 ((loff_t)index << PAGE_SHIFT) - 1, true);
2739 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002740 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002741 }
2742
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002743 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002744 * Inform shmem_writepage() how far we have reached.
2745 * No need for lock or barrier: we have the page lock.
2746 */
2747 shmem_falloc.next++;
2748 if (!PageUptodate(page))
2749 shmem_falloc.nr_falloced++;
2750
2751 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002752 * If !PageUptodate, leave it that way so that freeable pages
2753 * can be recognized if we need to rollback on error later.
2754 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002755 * than free the pages we are allocating (and SGP_CACHE pages
2756 * might still be clean: we now need to mark those dirty too).
2757 */
2758 set_page_dirty(page);
2759 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002760 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002761 cond_resched();
2762 }
2763
2764 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2765 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002766 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002767undone:
2768 spin_lock(&inode->i_lock);
2769 inode->i_private = NULL;
2770 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002771out:
Al Viro59551022016-01-22 15:40:57 -05002772 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002773 return error;
2774}
2775
David Howells726c3342006-06-23 02:02:58 -07002776static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777{
David Howells726c3342006-06-23 02:02:58 -07002778 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
2780 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002781 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002783 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002785 buf->f_bavail =
2786 buf->f_bfree = sbinfo->max_blocks -
2787 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002788 }
2789 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 buf->f_files = sbinfo->max_inodes;
2791 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 }
2793 /* else leave those fields 0 like simple_statfs */
2794 return 0;
2795}
2796
2797/*
2798 * File creation. Allocate an inode, and we're done..
2799 */
2800static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002801shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002803 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 int error = -ENOSPC;
2805
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002806 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002808 error = simple_acl_create(dir, inode);
2809 if (error)
2810 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002811 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002812 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002813 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002814 if (error && error != -EOPNOTSUPP)
2815 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002816
Al Viro718deb62009-12-16 19:35:36 -05002817 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002819 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 d_instantiate(dentry, inode);
2821 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 }
2823 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002824out_iput:
2825 iput(inode);
2826 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827}
2828
Al Viro60545d02013-06-07 01:20:27 -04002829static int
2830shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2831{
2832 struct inode *inode;
2833 int error = -ENOSPC;
2834
2835 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2836 if (inode) {
2837 error = security_inode_init_security(inode, dir,
2838 NULL,
2839 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002840 if (error && error != -EOPNOTSUPP)
2841 goto out_iput;
2842 error = simple_acl_create(dir, inode);
2843 if (error)
2844 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002845 d_tmpfile(dentry, inode);
2846 }
2847 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002848out_iput:
2849 iput(inode);
2850 return error;
Al Viro60545d02013-06-07 01:20:27 -04002851}
2852
Al Viro18bb1db2011-07-26 01:41:39 -04002853static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854{
2855 int error;
2856
2857 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2858 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002859 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 return 0;
2861}
2862
Al Viro4acdaf22011-07-26 01:42:34 -04002863static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002864 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865{
2866 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2867}
2868
2869/*
2870 * Link a file..
2871 */
2872static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2873{
David Howells75c3cfa2015-03-17 22:26:12 +00002874 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002875 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876
2877 /*
2878 * No ordinary (disk based) filesystem counts links as inodes;
2879 * but each new link needs a new dentry, pinning lowmem, and
2880 * tmpfs dentries cannot be pruned until they are unlinked.
2881 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002882 ret = shmem_reserve_inode(inode->i_sb);
2883 if (ret)
2884 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885
2886 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002887 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002888 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002889 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 dget(dentry); /* Extra pinning count for the created dentry */
2891 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002892out:
2893 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894}
2895
2896static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2897{
David Howells75c3cfa2015-03-17 22:26:12 +00002898 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002900 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2901 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
2903 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002904 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002905 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 dput(dentry); /* Undo the count from "create" - this does all the work */
2907 return 0;
2908}
2909
2910static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2911{
2912 if (!simple_empty(dentry))
2913 return -ENOTEMPTY;
2914
David Howells75c3cfa2015-03-17 22:26:12 +00002915 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002916 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 return shmem_unlink(dir, dentry);
2918}
2919
Miklos Szeredi37456772014-07-23 15:15:34 +02002920static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2921{
David Howellse36cb0b2015-01-29 12:02:35 +00002922 bool old_is_dir = d_is_dir(old_dentry);
2923 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002924
2925 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2926 if (old_is_dir) {
2927 drop_nlink(old_dir);
2928 inc_nlink(new_dir);
2929 } else {
2930 drop_nlink(new_dir);
2931 inc_nlink(old_dir);
2932 }
2933 }
2934 old_dir->i_ctime = old_dir->i_mtime =
2935 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002936 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002937 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02002938
2939 return 0;
2940}
2941
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002942static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2943{
2944 struct dentry *whiteout;
2945 int error;
2946
2947 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2948 if (!whiteout)
2949 return -ENOMEM;
2950
2951 error = shmem_mknod(old_dir, whiteout,
2952 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2953 dput(whiteout);
2954 if (error)
2955 return error;
2956
2957 /*
2958 * Cheat and hash the whiteout while the old dentry is still in
2959 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2960 *
2961 * d_lookup() will consistently find one of them at this point,
2962 * not sure which one, but that isn't even important.
2963 */
2964 d_rehash(whiteout);
2965 return 0;
2966}
2967
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968/*
2969 * The VFS layer already does all the dentry stuff for rename,
2970 * we just have to decrement the usage count for the target if
2971 * it exists so that the VFS layer correctly free's it when it
2972 * gets overwritten.
2973 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002974static 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 -07002975{
David Howells75c3cfa2015-03-17 22:26:12 +00002976 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 int they_are_dirs = S_ISDIR(inode->i_mode);
2978
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002979 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002980 return -EINVAL;
2981
Miklos Szeredi37456772014-07-23 15:15:34 +02002982 if (flags & RENAME_EXCHANGE)
2983 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2984
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 if (!simple_empty(new_dentry))
2986 return -ENOTEMPTY;
2987
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002988 if (flags & RENAME_WHITEOUT) {
2989 int error;
2990
2991 error = shmem_whiteout(old_dir, old_dentry);
2992 if (error)
2993 return error;
2994 }
2995
David Howells75c3cfa2015-03-17 22:26:12 +00002996 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002998 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002999 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003000 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003003 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003004 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 }
3006
3007 old_dir->i_size -= BOGO_DIRENT_SIZE;
3008 new_dir->i_size += BOGO_DIRENT_SIZE;
3009 old_dir->i_ctime = old_dir->i_mtime =
3010 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003011 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 return 0;
3013}
3014
3015static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3016{
3017 int error;
3018 int len;
3019 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003020 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 struct shmem_inode_info *info;
3022
3023 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003024 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 return -ENAMETOOLONG;
3026
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003027 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 if (!inode)
3029 return -ENOSPC;
3030
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003031 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003032 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003033 if (error) {
3034 if (error != -EOPNOTSUPP) {
3035 iput(inode);
3036 return error;
3037 }
3038 error = 0;
3039 }
3040
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 info = SHMEM_I(inode);
3042 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003043 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003044 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3045 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003046 iput(inode);
3047 return -ENOMEM;
3048 }
3049 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003051 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003052 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 if (error) {
3054 iput(inode);
3055 return error;
3056 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003057 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003059 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003060 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003062 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003063 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003066 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 d_instantiate(dentry, inode);
3068 dget(dentry);
3069 return 0;
3070}
3071
Al Virofceef392015-12-29 15:58:39 -05003072static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073{
Al Virofceef392015-12-29 15:58:39 -05003074 mark_page_accessed(arg);
3075 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076}
3077
Al Viro6b255392015-11-17 10:20:54 -05003078static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003079 struct inode *inode,
3080 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003083 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003084 if (!dentry) {
3085 page = find_get_page(inode->i_mapping, 0);
3086 if (!page)
3087 return ERR_PTR(-ECHILD);
3088 if (!PageUptodate(page)) {
3089 put_page(page);
3090 return ERR_PTR(-ECHILD);
3091 }
3092 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003093 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003094 if (error)
3095 return ERR_PTR(error);
3096 unlock_page(page);
3097 }
Al Virofceef392015-12-29 15:58:39 -05003098 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003099 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100}
3101
Eric Parisb09e0fa2011-05-24 17:12:39 -07003102#ifdef CONFIG_TMPFS_XATTR
3103/*
3104 * Superblocks without xattr inode operations may get some security.* xattr
3105 * support from the LSM "for free". As soon as we have any other xattrs
3106 * like ACLs, we also need to implement the security.* handlers at
3107 * filesystem level, though.
3108 */
3109
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003110/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003111 * Callback for security_inode_init_security() for acquiring xattrs.
3112 */
3113static int shmem_initxattrs(struct inode *inode,
3114 const struct xattr *xattr_array,
3115 void *fs_info)
3116{
3117 struct shmem_inode_info *info = SHMEM_I(inode);
3118 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003119 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003120 size_t len;
3121
3122 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003123 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003124 if (!new_xattr)
3125 return -ENOMEM;
3126
3127 len = strlen(xattr->name) + 1;
3128 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3129 GFP_KERNEL);
3130 if (!new_xattr->name) {
3131 kfree(new_xattr);
3132 return -ENOMEM;
3133 }
3134
3135 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3136 XATTR_SECURITY_PREFIX_LEN);
3137 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3138 xattr->name, len);
3139
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003140 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003141 }
3142
3143 return 0;
3144}
3145
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003146static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003147 struct dentry *unused, struct inode *inode,
3148 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003149{
Al Virob2968212016-04-10 20:48:24 -04003150 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003151
3152 name = xattr_full_name(handler, name);
3153 return simple_xattr_get(&info->xattrs, name, buffer, size);
3154}
3155
3156static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003157 struct dentry *unused, struct inode *inode,
3158 const char *name, const void *value,
3159 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003160{
Al Viro59301222016-05-27 10:19:30 -04003161 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003162
3163 name = xattr_full_name(handler, name);
3164 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3165}
3166
3167static const struct xattr_handler shmem_security_xattr_handler = {
3168 .prefix = XATTR_SECURITY_PREFIX,
3169 .get = shmem_xattr_handler_get,
3170 .set = shmem_xattr_handler_set,
3171};
3172
3173static const struct xattr_handler shmem_trusted_xattr_handler = {
3174 .prefix = XATTR_TRUSTED_PREFIX,
3175 .get = shmem_xattr_handler_get,
3176 .set = shmem_xattr_handler_set,
3177};
3178
Eric Parisb09e0fa2011-05-24 17:12:39 -07003179static const struct xattr_handler *shmem_xattr_handlers[] = {
3180#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003181 &posix_acl_access_xattr_handler,
3182 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003183#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003184 &shmem_security_xattr_handler,
3185 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003186 NULL
3187};
3188
Eric Parisb09e0fa2011-05-24 17:12:39 -07003189static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3190{
David Howells75c3cfa2015-03-17 22:26:12 +00003191 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003192 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003193}
3194#endif /* CONFIG_TMPFS_XATTR */
3195
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003196static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003198 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003199#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003200 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003201#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202};
3203
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003204static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003206 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003207#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003208 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003209#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003210};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003211
David M. Grimes91828a42006-10-17 00:09:45 -07003212static struct dentry *shmem_get_parent(struct dentry *child)
3213{
3214 return ERR_PTR(-ESTALE);
3215}
3216
3217static int shmem_match(struct inode *ino, void *vfh)
3218{
3219 __u32 *fh = vfh;
3220 __u64 inum = fh[2];
3221 inum = (inum << 32) | fh[1];
3222 return ino->i_ino == inum && fh[0] == ino->i_generation;
3223}
3224
Christoph Hellwig480b1162007-10-21 16:42:13 -07003225static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3226 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003227{
David M. Grimes91828a42006-10-17 00:09:45 -07003228 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003229 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003230 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003231
Christoph Hellwig480b1162007-10-21 16:42:13 -07003232 if (fh_len < 3)
3233 return NULL;
3234
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003235 inum = fid->raw[2];
3236 inum = (inum << 32) | fid->raw[1];
3237
Christoph Hellwig480b1162007-10-21 16:42:13 -07003238 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3239 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003240 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003241 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003242 iput(inode);
3243 }
3244
Christoph Hellwig480b1162007-10-21 16:42:13 -07003245 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003246}
3247
Al Virob0b03822012-04-02 14:34:06 -04003248static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3249 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003250{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303251 if (*len < 3) {
3252 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003253 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303254 }
David M. Grimes91828a42006-10-17 00:09:45 -07003255
Al Viro1d3382cb2010-10-23 15:19:20 -04003256 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003257 /* Unfortunately insert_inode_hash is not idempotent,
3258 * so as we hash inodes here rather than at creation
3259 * time, we need a lock to ensure we only try
3260 * to do it once
3261 */
3262 static DEFINE_SPINLOCK(lock);
3263 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003264 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003265 __insert_inode_hash(inode,
3266 inode->i_ino + inode->i_generation);
3267 spin_unlock(&lock);
3268 }
3269
3270 fh[0] = inode->i_generation;
3271 fh[1] = inode->i_ino;
3272 fh[2] = ((__u64)inode->i_ino) >> 32;
3273
3274 *len = 3;
3275 return 1;
3276}
3277
Christoph Hellwig39655162007-10-21 16:42:17 -07003278static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003279 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003280 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003281 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003282};
3283
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003284static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3285 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286{
3287 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003288 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003289 uid_t uid;
3290 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003292 while (options != NULL) {
3293 this_char = options;
3294 for (;;) {
3295 /*
3296 * NUL-terminate this option: unfortunately,
3297 * mount options form a comma-separated list,
3298 * but mpol's nodelist may also contain commas.
3299 */
3300 options = strchr(options, ',');
3301 if (options == NULL)
3302 break;
3303 options++;
3304 if (!isdigit(*options)) {
3305 options[-1] = '\0';
3306 break;
3307 }
3308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 if (!*this_char)
3310 continue;
3311 if ((value = strchr(this_char,'=')) != NULL) {
3312 *value++ = 0;
3313 } else {
Joe Perches11705322016-03-17 14:19:50 -07003314 pr_err("tmpfs: No value for mount option '%s'\n",
3315 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003316 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 }
3318
3319 if (!strcmp(this_char,"size")) {
3320 unsigned long long size;
3321 size = memparse(value,&rest);
3322 if (*rest == '%') {
3323 size <<= PAGE_SHIFT;
3324 size *= totalram_pages;
3325 do_div(size, 100);
3326 rest++;
3327 }
3328 if (*rest)
3329 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003330 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003331 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003333 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 if (*rest)
3335 goto bad_val;
3336 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003337 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 if (*rest)
3339 goto bad_val;
3340 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003341 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003343 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344 if (*rest)
3345 goto bad_val;
3346 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003347 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003349 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 if (*rest)
3351 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003352 sbinfo->uid = make_kuid(current_user_ns(), uid);
3353 if (!uid_valid(sbinfo->uid))
3354 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003356 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003358 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 if (*rest)
3360 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003361 sbinfo->gid = make_kgid(current_user_ns(), gid);
3362 if (!gid_valid(sbinfo->gid))
3363 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003364#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003365 } else if (!strcmp(this_char, "huge")) {
3366 int huge;
3367 huge = shmem_parse_huge(value);
3368 if (huge < 0)
3369 goto bad_val;
3370 if (!has_transparent_hugepage() &&
3371 huge != SHMEM_HUGE_NEVER)
3372 goto bad_val;
3373 sbinfo->huge = huge;
3374#endif
3375#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003376 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003377 mpol_put(mpol);
3378 mpol = NULL;
3379 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003380 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003381#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 } else {
Joe Perches11705322016-03-17 14:19:50 -07003383 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003384 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 }
3386 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003387 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 return 0;
3389
3390bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003391 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003393error:
3394 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 return 1;
3396
3397}
3398
3399static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3400{
3401 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003402 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003403 unsigned long inodes;
3404 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405
Greg Thelen5f001102013-02-22 16:36:01 -08003406 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003407 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003408 return error;
3409
3410 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003411 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003412 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003413 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003414 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003415 goto out;
3416 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003417 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003418 * but we must separately disallow unlimited->limited, because
3419 * in that case we have no record of how much is already in use.
3420 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003421 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003422 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003423 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003424 goto out;
3425
3426 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003427 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003428 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003429 sbinfo->max_inodes = config.max_inodes;
3430 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003431
Greg Thelen5f001102013-02-22 16:36:01 -08003432 /*
3433 * Preserve previous mempolicy unless mpol remount option was specified.
3434 */
3435 if (config.mpol) {
3436 mpol_put(sbinfo->mpol);
3437 sbinfo->mpol = config.mpol; /* transfers initial ref */
3438 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003439out:
3440 spin_unlock(&sbinfo->stat_lock);
3441 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003443
Al Viro34c80b12011-12-08 21:32:45 -05003444static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003445{
Al Viro34c80b12011-12-08 21:32:45 -05003446 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003447
3448 if (sbinfo->max_blocks != shmem_default_max_blocks())
3449 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003450 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003451 if (sbinfo->max_inodes != shmem_default_max_inodes())
3452 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3453 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003454 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003455 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3456 seq_printf(seq, ",uid=%u",
3457 from_kuid_munged(&init_user_ns, sbinfo->uid));
3458 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3459 seq_printf(seq, ",gid=%u",
3460 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003461#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003462 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3463 if (sbinfo->huge)
3464 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3465#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003466 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003467 return 0;
3468}
David Herrmann9183df22014-08-08 14:25:29 -07003469
3470#define MFD_NAME_PREFIX "memfd:"
3471#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3472#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3473
3474#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3475
3476SYSCALL_DEFINE2(memfd_create,
3477 const char __user *, uname,
3478 unsigned int, flags)
3479{
3480 struct shmem_inode_info *info;
3481 struct file *file;
3482 int fd, error;
3483 char *name;
3484 long len;
3485
3486 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3487 return -EINVAL;
3488
3489 /* length includes terminating zero */
3490 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3491 if (len <= 0)
3492 return -EFAULT;
3493 if (len > MFD_NAME_MAX_LEN + 1)
3494 return -EINVAL;
3495
3496 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3497 if (!name)
3498 return -ENOMEM;
3499
3500 strcpy(name, MFD_NAME_PREFIX);
3501 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3502 error = -EFAULT;
3503 goto err_name;
3504 }
3505
3506 /* terminating-zero may have changed after strnlen_user() returned */
3507 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3508 error = -EFAULT;
3509 goto err_name;
3510 }
3511
3512 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3513 if (fd < 0) {
3514 error = fd;
3515 goto err_name;
3516 }
3517
3518 file = shmem_file_setup(name, 0, VM_NORESERVE);
3519 if (IS_ERR(file)) {
3520 error = PTR_ERR(file);
3521 goto err_fd;
3522 }
3523 info = SHMEM_I(file_inode(file));
3524 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3525 file->f_flags |= O_RDWR | O_LARGEFILE;
3526 if (flags & MFD_ALLOW_SEALING)
3527 info->seals &= ~F_SEAL_SEAL;
3528
3529 fd_install(fd, file);
3530 kfree(name);
3531 return fd;
3532
3533err_fd:
3534 put_unused_fd(fd);
3535err_name:
3536 kfree(name);
3537 return error;
3538}
3539
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003540#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541
3542static void shmem_put_super(struct super_block *sb)
3543{
Hugh Dickins602586a2010-08-17 15:23:56 -07003544 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3545
3546 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003547 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003548 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549 sb->s_fs_info = NULL;
3550}
3551
Kay Sievers2b2af542009-04-30 15:23:42 +02003552int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553{
3554 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003555 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003556 int err = -ENOMEM;
3557
3558 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003559 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003560 L1_CACHE_BYTES), GFP_KERNEL);
3561 if (!sbinfo)
3562 return -ENOMEM;
3563
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003564 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003565 sbinfo->uid = current_fsuid();
3566 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003567 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003569#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 /*
3571 * Per default we only allow half of the physical ram per
3572 * tmpfs instance, limiting inodes to one per page of lowmem;
3573 * but the internal instance is left unlimited.
3574 */
Al Viroca4e0512013-08-31 12:57:10 -04003575 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003576 sbinfo->max_blocks = shmem_default_max_blocks();
3577 sbinfo->max_inodes = shmem_default_max_inodes();
3578 if (shmem_parse_options(data, sbinfo, false)) {
3579 err = -EINVAL;
3580 goto failed;
3581 }
Al Viroca4e0512013-08-31 12:57:10 -04003582 } else {
3583 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 }
David M. Grimes91828a42006-10-17 00:09:45 -07003585 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003586 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587#else
3588 sb->s_flags |= MS_NOUSER;
3589#endif
3590
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003591 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003592 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003593 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003594 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003595 spin_lock_init(&sbinfo->shrinklist_lock);
3596 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003597
Hugh Dickins285b2c42011-08-03 16:21:20 -07003598 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003599 sb->s_blocksize = PAGE_SIZE;
3600 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 sb->s_magic = TMPFS_MAGIC;
3602 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003603 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003604#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003605 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003606#endif
3607#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003608 sb->s_flags |= MS_POSIXACL;
3609#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003610
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003611 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 if (!inode)
3613 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003614 inode->i_uid = sbinfo->uid;
3615 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003616 sb->s_root = d_make_root(inode);
3617 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003618 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 return 0;
3620
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621failed:
3622 shmem_put_super(sb);
3623 return err;
3624}
3625
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003626static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627
3628static struct inode *shmem_alloc_inode(struct super_block *sb)
3629{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003630 struct shmem_inode_info *info;
3631 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3632 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003634 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635}
3636
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003637static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003638{
3639 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003640 if (S_ISLNK(inode->i_mode))
3641 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003642 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3643}
3644
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645static void shmem_destroy_inode(struct inode *inode)
3646{
Al Viro09208d12011-07-26 03:15:03 -04003647 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003649 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650}
3651
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003652static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003654 struct shmem_inode_info *info = foo;
3655 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656}
3657
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003658static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659{
3660 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3661 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003662 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 return 0;
3664}
3665
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003666static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003668 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669}
3670
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003671static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003673 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003675 .write_begin = shmem_write_begin,
3676 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003678#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003679 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003680#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003681 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682};
3683
Helge Deller15ad7cd2006-12-06 20:40:36 -08003684static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003686 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003688 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003689 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003690 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003691 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003692 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003693 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003694 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695#endif
3696};
3697
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003698static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003699 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003700 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003701#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003702 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003703 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003704#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705};
3706
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003707static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708#ifdef CONFIG_TMPFS
3709 .create = shmem_create,
3710 .lookup = simple_lookup,
3711 .link = shmem_link,
3712 .unlink = shmem_unlink,
3713 .symlink = shmem_symlink,
3714 .mkdir = shmem_mkdir,
3715 .rmdir = shmem_rmdir,
3716 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003717 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003718 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003720#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003721 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003722#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003723#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003724 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003725 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003726#endif
3727};
3728
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003729static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003730#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003731 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003732#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003733#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003734 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003735 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003736#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737};
3738
Hugh Dickins759b9772007-03-05 00:30:28 -08003739static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 .alloc_inode = shmem_alloc_inode,
3741 .destroy_inode = shmem_destroy_inode,
3742#ifdef CONFIG_TMPFS
3743 .statfs = shmem_statfs,
3744 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003745 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746#endif
Al Viro1f895f72010-06-05 19:10:41 -04003747 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748 .drop_inode = generic_delete_inode,
3749 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003750#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3751 .nr_cached_objects = shmem_unused_huge_count,
3752 .free_cached_objects = shmem_unused_huge_scan,
3753#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754};
3755
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003756static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003757 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003758 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759#ifdef CONFIG_NUMA
3760 .set_policy = shmem_set_policy,
3761 .get_policy = shmem_get_policy,
3762#endif
3763};
3764
Al Viro3c26ff62010-07-25 11:46:36 +04003765static struct dentry *shmem_mount(struct file_system_type *fs_type,
3766 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767{
Al Viro3c26ff62010-07-25 11:46:36 +04003768 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769}
3770
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003771static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772 .owner = THIS_MODULE,
3773 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003774 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003776 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003779int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780{
3781 int error;
3782
Rob Landley16203a72013-09-11 14:26:12 -07003783 /* If rootfs called this, don't re-init */
3784 if (shmem_inode_cachep)
3785 return 0;
3786
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003787 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 if (error)
3789 goto out3;
3790
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003791 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003793 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 goto out2;
3795 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003796
Al Viroca4e0512013-08-31 12:57:10 -04003797 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 if (IS_ERR(shm_mnt)) {
3799 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003800 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 goto out1;
3802 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003803
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003804#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003805 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3806 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3807 else
3808 shmem_huge = 0; /* just in case it was patched */
3809#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810 return 0;
3811
3812out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003813 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003815 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816out3:
3817 shm_mnt = ERR_PTR(error);
3818 return error;
3819}
Matt Mackall853ac432009-01-06 14:40:20 -08003820
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003821#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003822static ssize_t shmem_enabled_show(struct kobject *kobj,
3823 struct kobj_attribute *attr, char *buf)
3824{
3825 int values[] = {
3826 SHMEM_HUGE_ALWAYS,
3827 SHMEM_HUGE_WITHIN_SIZE,
3828 SHMEM_HUGE_ADVISE,
3829 SHMEM_HUGE_NEVER,
3830 SHMEM_HUGE_DENY,
3831 SHMEM_HUGE_FORCE,
3832 };
3833 int i, count;
3834
3835 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3836 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3837
3838 count += sprintf(buf + count, fmt,
3839 shmem_format_huge(values[i]));
3840 }
3841 buf[count - 1] = '\n';
3842 return count;
3843}
3844
3845static ssize_t shmem_enabled_store(struct kobject *kobj,
3846 struct kobj_attribute *attr, const char *buf, size_t count)
3847{
3848 char tmp[16];
3849 int huge;
3850
3851 if (count + 1 > sizeof(tmp))
3852 return -EINVAL;
3853 memcpy(tmp, buf, count);
3854 tmp[count] = '\0';
3855 if (count && tmp[count - 1] == '\n')
3856 tmp[count - 1] = '\0';
3857
3858 huge = shmem_parse_huge(tmp);
3859 if (huge == -EINVAL)
3860 return -EINVAL;
3861 if (!has_transparent_hugepage() &&
3862 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3863 return -EINVAL;
3864
3865 shmem_huge = huge;
3866 if (shmem_huge < SHMEM_HUGE_DENY)
3867 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3868 return count;
3869}
3870
3871struct kobj_attribute shmem_enabled_attr =
3872 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003873#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003874
Arnd Bergmann3b337192016-08-10 16:27:44 -07003875#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003876bool shmem_huge_enabled(struct vm_area_struct *vma)
3877{
3878 struct inode *inode = file_inode(vma->vm_file);
3879 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3880 loff_t i_size;
3881 pgoff_t off;
3882
3883 if (shmem_huge == SHMEM_HUGE_FORCE)
3884 return true;
3885 if (shmem_huge == SHMEM_HUGE_DENY)
3886 return false;
3887 switch (sbinfo->huge) {
3888 case SHMEM_HUGE_NEVER:
3889 return false;
3890 case SHMEM_HUGE_ALWAYS:
3891 return true;
3892 case SHMEM_HUGE_WITHIN_SIZE:
3893 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3894 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3895 if (i_size >= HPAGE_PMD_SIZE &&
3896 i_size >> PAGE_SHIFT >= off)
3897 return true;
3898 case SHMEM_HUGE_ADVISE:
3899 /* TODO: implement fadvise() hints */
3900 return (vma->vm_flags & VM_HUGEPAGE);
3901 default:
3902 VM_BUG_ON(1);
3903 return false;
3904 }
3905}
Arnd Bergmann3b337192016-08-10 16:27:44 -07003906#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003907
Matt Mackall853ac432009-01-06 14:40:20 -08003908#else /* !CONFIG_SHMEM */
3909
3910/*
3911 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3912 *
3913 * This is intended for small system where the benefits of the full
3914 * shmem code (swap-backed and resource-limited) are outweighed by
3915 * their complexity. On systems without swap this code should be
3916 * effectively equivalent, but much lighter weight.
3917 */
3918
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003919static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003920 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003921 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003922 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003923 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003924};
3925
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003926int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003927{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003928 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003929
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003930 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003931 BUG_ON(IS_ERR(shm_mnt));
3932
3933 return 0;
3934}
3935
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003936int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003937{
3938 return 0;
3939}
3940
Hugh Dickins3f96b792009-09-21 17:03:37 -07003941int shmem_lock(struct file *file, int lock, struct user_struct *user)
3942{
3943 return 0;
3944}
3945
Hugh Dickins24513262012-01-20 14:34:21 -08003946void shmem_unlock_mapping(struct address_space *mapping)
3947{
3948}
3949
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003950#ifdef CONFIG_MMU
3951unsigned long shmem_get_unmapped_area(struct file *file,
3952 unsigned long addr, unsigned long len,
3953 unsigned long pgoff, unsigned long flags)
3954{
3955 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3956}
3957#endif
3958
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003959void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003960{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003961 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003962}
3963EXPORT_SYMBOL_GPL(shmem_truncate_range);
3964
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003965#define shmem_vm_ops generic_file_vm_ops
3966#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003967#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003968#define shmem_acct_size(flags, size) 0
3969#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003970
3971#endif /* CONFIG_SHMEM */
3972
3973/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974
Rasmus Villemoes19938e32016-10-07 17:01:01 -07003975static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003976 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003977};
3978
Eric Parisc7277092013-12-02 11:24:19 +00003979static struct file *__shmem_file_setup(const char *name, loff_t size,
3980 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981{
Al Viro6b4d0b22013-02-14 21:37:26 -05003982 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003984 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003985 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986 struct qstr this;
3987
3988 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003989 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990
Hugh Dickins285b2c42011-08-03 16:21:20 -07003991 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 return ERR_PTR(-EINVAL);
3993
3994 if (shmem_acct_size(flags, size))
3995 return ERR_PTR(-ENOMEM);
3996
Al Viro6b4d0b22013-02-14 21:37:26 -05003997 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998 this.name = name;
3999 this.len = strlen(name);
4000 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004001 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004002 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004003 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004004 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004006 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007
Al Viro6b4d0b22013-02-14 21:37:26 -05004008 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004009 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004011 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012
Eric Parisc7277092013-12-02 11:24:19 +00004013 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004014 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004016 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004017 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4018 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004019 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004020
Al Viro6b4d0b22013-02-14 21:37:26 -05004021 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004022 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004023 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004024 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004025
Al Viro6b4d0b22013-02-14 21:37:26 -05004026 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004027
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028put_memory:
4029 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004030put_path:
4031 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004032 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033}
Eric Parisc7277092013-12-02 11:24:19 +00004034
4035/**
4036 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4037 * kernel internal. There will be NO LSM permission checks against the
4038 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004039 * higher layer. The users are the big_key and shm implementations. LSM
4040 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004041 * @name: name for dentry (to be seen in /proc/<pid>/maps
4042 * @size: size to be set for the file
4043 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4044 */
4045struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4046{
4047 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4048}
4049
4050/**
4051 * shmem_file_setup - get an unlinked file living in tmpfs
4052 * @name: name for dentry (to be seen in /proc/<pid>/maps
4053 * @size: size to be set for the file
4054 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4055 */
4056struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4057{
4058 return __shmem_file_setup(name, size, flags, 0);
4059}
Keith Packard395e0dd2008-06-20 00:08:06 -07004060EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061
Randy Dunlap46711812008-03-19 17:00:41 -07004062/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4065 */
4066int shmem_zero_setup(struct vm_area_struct *vma)
4067{
4068 struct file *file;
4069 loff_t size = vma->vm_end - vma->vm_start;
4070
Hugh Dickins66fc1302015-06-14 09:48:09 -07004071 /*
4072 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4073 * between XFS directory reading and selinux: since this file is only
4074 * accessible to the user through its mapping, use S_PRIVATE flag to
4075 * bypass file security, in the same way as shmem_kernel_file_setup().
4076 */
4077 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078 if (IS_ERR(file))
4079 return PTR_ERR(file);
4080
4081 if (vma->vm_file)
4082 fput(vma->vm_file);
4083 vma->vm_file = file;
4084 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004085
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004086 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004087 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4088 (vma->vm_end & HPAGE_PMD_MASK)) {
4089 khugepaged_enter(vma, vma->vm_flags);
4090 }
4091
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092 return 0;
4093}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004094
4095/**
4096 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4097 * @mapping: the page's address_space
4098 * @index: the page index
4099 * @gfp: the page allocator flags to use if allocating
4100 *
4101 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4102 * with any new page allocations done using the specified allocation flags.
4103 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4104 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4105 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4106 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004107 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4108 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004109 */
4110struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4111 pgoff_t index, gfp_t gfp)
4112{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004113#ifdef CONFIG_SHMEM
4114 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004115 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004116 int error;
4117
4118 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004119 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4120 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004121 if (error)
4122 page = ERR_PTR(error);
4123 else
4124 unlock_page(page);
4125 return page;
4126#else
4127 /*
4128 * The tiny !SHMEM case uses ramfs without swap
4129 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004130 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004131#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004132}
4133EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);