blob: 7a74b6dfa5d20f020e29015b4c1e1ce50fb4ef39 [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);
Cong Wange2286912017-08-10 15:24:24 -07001010 /*
1011 * _careful to defend against unlocked access to
1012 * ->shrink_list in shmem_unused_huge_shrink()
1013 */
1014 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001015 list_add_tail(&info->shrinklist,
1016 &sbinfo->shrinklist);
1017 sbinfo->shrinklist_len++;
1018 }
1019 spin_unlock(&sbinfo->shrinklist_lock);
1020 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
1023
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001024 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001025 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001026 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 return error;
1028}
1029
Al Viro1f895f72010-06-05 19:10:41 -04001030static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001033 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001035 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 shmem_unacct_size(info->flags, inode->i_size);
1037 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001038 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001039 if (!list_empty(&info->shrinklist)) {
1040 spin_lock(&sbinfo->shrinklist_lock);
1041 if (!list_empty(&info->shrinklist)) {
1042 list_del_init(&info->shrinklist);
1043 sbinfo->shrinklist_len--;
1044 }
1045 spin_unlock(&sbinfo->shrinklist_lock);
1046 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001048 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001050 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
Al Viro3ed47db2016-01-22 18:08:52 -05001052 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001053
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001054 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001055 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001056 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001057 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058}
1059
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001060/*
1061 * If swap found in inode, free it and move page from swapcache to filecache.
1062 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001063static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001064 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001066 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001067 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001068 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001069 gfp_t gfp;
1070 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001072 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001073 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001074 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001075 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001076
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001077 /*
1078 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001079 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001080 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001081 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001082 */
1083 if (shmem_swaplist.next != &info->swaplist)
1084 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001085
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001086 gfp = mapping_gfp_mask(mapping);
1087 if (shmem_should_replace_page(*pagep, gfp)) {
1088 mutex_unlock(&shmem_swaplist_mutex);
1089 error = shmem_replace_page(pagep, gfp, info, index);
1090 mutex_lock(&shmem_swaplist_mutex);
1091 /*
1092 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001093 * allocation, but the inode might have been freed while we
1094 * dropped it: although a racing shmem_evict_inode() cannot
1095 * complete without emptying the radix_tree, our page lock
1096 * on this swapcache page is not enough to prevent that -
1097 * free_swap_and_cache() of our swap entry will only
1098 * trylock_page(), removing swap from radix_tree whatever.
1099 *
1100 * We must not proceed to shmem_add_to_page_cache() if the
1101 * inode has been freed, but of course we cannot rely on
1102 * inode or mapping or info to check that. However, we can
1103 * safely check if our swap entry is still in use (and here
1104 * it can't have got reused for another page): if it's still
1105 * in use, then the inode cannot have been freed yet, and we
1106 * can safely proceed (if it's no longer in use, that tells
1107 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001108 */
1109 if (!page_swapcount(*pagep))
1110 error = -ENOENT;
1111 }
1112
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001113 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001114 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1115 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1116 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001117 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001118 if (!error)
1119 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001120 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001121 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001122 /*
1123 * Truncation and eviction use free_swap_and_cache(), which
1124 * only does trylock page: if we raced, best clean up here.
1125 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001126 delete_from_swap_cache(*pagep);
1127 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001128 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001129 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001130 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001131 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001132 swap_free(swap);
1133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001135 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136}
1137
1138/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001139 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001141int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001143 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001145 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001146 int error = 0;
1147
1148 /*
1149 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001150 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001151 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001152 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001153 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001154
1155 /*
1156 * Charge page using GFP_KERNEL while we can wait, before taking
1157 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1158 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001159 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001160 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1161 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001162 if (error)
1163 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001164 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001165 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001167 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001168 list_for_each_safe(this, next, &shmem_swaplist) {
1169 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001170 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001171 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001172 else
1173 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001174 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001175 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001176 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001177 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001179 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001180
Johannes Weiner00501b52014-08-08 14:19:20 -07001181 if (error) {
1182 if (error != -ENOMEM)
1183 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001184 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001185 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001186 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001187out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001188 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001189 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001190 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191}
1192
1193/*
1194 * Move the page from the page cache to the swap cache.
1195 */
1196static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1197{
1198 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001201 swp_entry_t swap;
1202 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001204 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 mapping = page->mapping;
1207 index = page->index;
1208 inode = mapping->host;
1209 info = SHMEM_I(inode);
1210 if (info->flags & VM_LOCKED)
1211 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001212 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 goto redirty;
1214
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001215 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001216 * Our capabilities prevent regular writeback or sync from ever calling
1217 * shmem_writepage; but a stacking filesystem might use ->writepage of
1218 * its underlying filesystem, in which case tmpfs should write out to
1219 * swap only in response to memory pressure, and not for the writeback
1220 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001221 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001222 if (!wbc->for_reclaim) {
1223 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1224 goto redirty;
1225 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001226
1227 /*
1228 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1229 * value into swapfile.c, the only way we can correctly account for a
1230 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001231 *
1232 * That's okay for a page already fallocated earlier, but if we have
1233 * not yet completed the fallocation, then (a) we want to keep track
1234 * of this page in case we have to undo it, and (b) it may not be a
1235 * good idea to continue anyway, once we're pushing into swap. So
1236 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001237 */
1238 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001239 if (inode->i_private) {
1240 struct shmem_falloc *shmem_falloc;
1241 spin_lock(&inode->i_lock);
1242 shmem_falloc = inode->i_private;
1243 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001244 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001245 index >= shmem_falloc->start &&
1246 index < shmem_falloc->next)
1247 shmem_falloc->nr_unswapped++;
1248 else
1249 shmem_falloc = NULL;
1250 spin_unlock(&inode->i_lock);
1251 if (shmem_falloc)
1252 goto redirty;
1253 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001254 clear_highpage(page);
1255 flush_dcache_page(page);
1256 SetPageUptodate(page);
1257 }
1258
Hugh Dickins48f170f2011-07-25 17:12:37 -07001259 swap = get_swap_page();
1260 if (!swap.val)
1261 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001262
Vladimir Davydov37e84352016-01-20 15:02:56 -08001263 if (mem_cgroup_try_charge_swap(page, swap))
1264 goto free_swap;
1265
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001266 /*
1267 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001268 * if it's not already there. Do it now before the page is
1269 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001270 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001271 * we've incremented swapped, because shmem_unuse_inode() will
1272 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001273 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001274 mutex_lock(&shmem_swaplist_mutex);
1275 if (list_empty(&info->swaplist))
1276 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001277
Hugh Dickins48f170f2011-07-25 17:12:37 -07001278 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001279 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001280 shmem_recalc_inode(inode);
1281 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001282 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001283
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001284 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001285 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1286
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001287 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001288 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001289 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 return 0;
1291 }
1292
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001293 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001294free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001295 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296redirty:
1297 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001298 if (wbc->for_reclaim)
1299 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1300 unlock_page(page);
1301 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Hugh Dickins75edd342016-05-19 17:12:44 -07001304#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001305static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001306{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001307 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001308
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001309 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001310 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001311
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001312 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001313
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001314 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001315}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001316
1317static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1318{
1319 struct mempolicy *mpol = NULL;
1320 if (sbinfo->mpol) {
1321 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1322 mpol = sbinfo->mpol;
1323 mpol_get(mpol);
1324 spin_unlock(&sbinfo->stat_lock);
1325 }
1326 return mpol;
1327}
Hugh Dickins75edd342016-05-19 17:12:44 -07001328#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1329static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1330{
1331}
1332static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1333{
1334 return NULL;
1335}
1336#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1337#ifndef CONFIG_NUMA
1338#define vm_policy vm_private_data
1339#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001340
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001341static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1342 struct shmem_inode_info *info, pgoff_t index)
1343{
1344 /* Create a pseudo vma that just contains the policy */
1345 vma->vm_start = 0;
1346 /* Bias interleave by inode number to distribute better across nodes */
1347 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1348 vma->vm_ops = NULL;
1349 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1350}
1351
1352static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1353{
1354 /* Drop reference taken by mpol_shared_policy_lookup() */
1355 mpol_cond_put(vma->vm_policy);
1356}
1357
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001358static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1359 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001362 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001364 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001365 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001366 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001367
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001368 return page;
1369}
Mel Gorman18a2f372012-12-05 14:01:41 -08001370
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001371static struct page *shmem_alloc_hugepage(gfp_t gfp,
1372 struct shmem_inode_info *info, pgoff_t index)
1373{
1374 struct vm_area_struct pvma;
1375 struct inode *inode = &info->vfs_inode;
1376 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001377 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001378 void __rcu **results;
1379 struct page *page;
1380
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001381 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001382 return NULL;
1383
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001384 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001385 rcu_read_lock();
1386 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1387 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1388 rcu_read_unlock();
1389 return NULL;
1390 }
1391 rcu_read_unlock();
1392
1393 shmem_pseudo_vma_init(&pvma, info, hindex);
1394 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1395 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1396 shmem_pseudo_vma_destroy(&pvma);
1397 if (page)
1398 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001399 return page;
1400}
1401
1402static struct page *shmem_alloc_page(gfp_t gfp,
1403 struct shmem_inode_info *info, pgoff_t index)
1404{
1405 struct vm_area_struct pvma;
1406 struct page *page;
1407
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001408 shmem_pseudo_vma_init(&pvma, info, index);
1409 page = alloc_page_vma(gfp, &pvma, 0);
1410 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001411
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001412 return page;
1413}
1414
1415static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1416 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1417 pgoff_t index, bool huge)
1418{
1419 struct page *page;
1420 int nr;
1421 int err = -ENOSPC;
1422
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001423 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001424 huge = false;
1425 nr = huge ? HPAGE_PMD_NR : 1;
1426
1427 if (shmem_acct_block(info->flags, nr))
1428 goto failed;
1429 if (sbinfo->max_blocks) {
1430 if (percpu_counter_compare(&sbinfo->used_blocks,
1431 sbinfo->max_blocks - nr) > 0)
1432 goto unacct;
1433 percpu_counter_add(&sbinfo->used_blocks, nr);
1434 }
1435
1436 if (huge)
1437 page = shmem_alloc_hugepage(gfp, info, index);
1438 else
1439 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001440 if (page) {
1441 __SetPageLocked(page);
1442 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001443 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001444 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001445
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001446 err = -ENOMEM;
1447 if (sbinfo->max_blocks)
1448 percpu_counter_add(&sbinfo->used_blocks, -nr);
1449unacct:
1450 shmem_unacct_blocks(info->flags, nr);
1451failed:
1452 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001456 * When a page is moved from swapcache to shmem filecache (either by the
1457 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1458 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1459 * ignorance of the mapping it belongs to. If that mapping has special
1460 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1461 * we may need to copy to a suitable page before moving to filecache.
1462 *
1463 * In a future release, this may well be extended to respect cpuset and
1464 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1465 * but for now it is a simple matter of zone.
1466 */
1467static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1468{
1469 return page_zonenum(page) > gfp_zone(gfp);
1470}
1471
1472static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1473 struct shmem_inode_info *info, pgoff_t index)
1474{
1475 struct page *oldpage, *newpage;
1476 struct address_space *swap_mapping;
1477 pgoff_t swap_index;
1478 int error;
1479
1480 oldpage = *pagep;
1481 swap_index = page_private(oldpage);
1482 swap_mapping = page_mapping(oldpage);
1483
1484 /*
1485 * We have arrived here because our zones are constrained, so don't
1486 * limit chance of success by further cpuset and node constraints.
1487 */
1488 gfp &= ~GFP_CONSTRAINT_MASK;
1489 newpage = shmem_alloc_page(gfp, info, index);
1490 if (!newpage)
1491 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001492
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001493 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001494 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001495 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001496
Hugh Dickins9956edf2016-11-10 10:46:11 -08001497 __SetPageLocked(newpage);
1498 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001499 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001500 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001501 SetPageSwapCache(newpage);
1502
1503 /*
1504 * Our caller will very soon move newpage out of swapcache, but it's
1505 * a nice clean interface for us to replace oldpage by newpage there.
1506 */
1507 spin_lock_irq(&swap_mapping->tree_lock);
1508 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1509 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001510 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001511 __inc_node_page_state(newpage, NR_FILE_PAGES);
1512 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001513 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001514 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001515
Hugh Dickins0142ef62012-06-07 14:21:09 -07001516 if (unlikely(error)) {
1517 /*
1518 * Is this possible? I think not, now that our callers check
1519 * both PageSwapCache and page_private after getting page lock;
1520 * but be defensive. Reverse old to newpage for clear and free.
1521 */
1522 oldpage = newpage;
1523 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001524 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001525 lru_cache_add_anon(newpage);
1526 *pagep = newpage;
1527 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001528
1529 ClearPageSwapCache(oldpage);
1530 set_page_private(oldpage, 0);
1531
1532 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001533 put_page(oldpage);
1534 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001535 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001536}
1537
1538/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001539 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 *
1541 * If we allocate a new one we do not mark it dirty. That's up to the
1542 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001543 * entry since a page cannot live in both the swap and page cache.
1544 *
1545 * fault_mm and fault_type are only supplied by shmem_fault:
1546 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001548static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001549 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1550 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
1552 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001553 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001555 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001556 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001557 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001559 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001560 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001562 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001563 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001565 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001567 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1568 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001570 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001571 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001572 if (radix_tree_exceptional_entry(page)) {
1573 swap = radix_to_swp_entry(page);
1574 page = NULL;
1575 }
1576
Hugh Dickins75edd342016-05-19 17:12:44 -07001577 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001578 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001579 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001580 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001581 }
1582
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001583 if (page && sgp == SGP_WRITE)
1584 mark_page_accessed(page);
1585
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001586 /* fallocated page? */
1587 if (page && !PageUptodate(page)) {
1588 if (sgp != SGP_READ)
1589 goto clear;
1590 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001591 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001592 page = NULL;
1593 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001594 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001595 *pagep = page;
1596 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
1599 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001600 * Fast cache lookup did not find it:
1601 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001603 info = SHMEM_I(inode);
1604 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001605 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 if (swap.val) {
1608 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001609 page = lookup_swap_cache(swap);
1610 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001611 /* Or update major stats only when swapin succeeds?? */
1612 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001613 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001614 count_vm_event(PGMAJFAULT);
1615 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1616 }
1617 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001618 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001619 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001620 error = -ENOMEM;
1621 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 }
1624
1625 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001626 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001627 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001628 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001629 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001630 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001631 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001632 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001634 goto failed;
1635 }
1636 wait_on_page_writeback(page);
1637
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001638 if (shmem_should_replace_page(page, gfp)) {
1639 error = shmem_replace_page(&page, gfp, info, index);
1640 if (error)
1641 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 }
1643
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001644 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001645 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001646 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001647 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001648 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001649 /*
1650 * We already confirmed swap under page lock, and make
1651 * no memory allocation here, so usually no possibility
1652 * of error; but free_swap_and_cache() only trylocks a
1653 * page, so it is just possible that the entry has been
1654 * truncated or holepunched since swap was confirmed.
1655 * shmem_undo_range() will have done some of the
1656 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001657 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001658 * Reset swap.val? No, leave it so "failed" goes back to
1659 * "repeat": reading a hole and writing should succeed.
1660 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001661 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001662 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001663 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001664 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001665 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001666 if (error)
1667 goto failed;
1668
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001669 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001670
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001671 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001672 info->swapped--;
1673 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001674 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001675
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001676 if (sgp == SGP_WRITE)
1677 mark_page_accessed(page);
1678
Hugh Dickins27ab7002011-07-25 17:12:36 -07001679 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001680 set_page_dirty(page);
1681 swap_free(swap);
1682
Hugh Dickins54af60422011-08-03 16:21:24 -07001683 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001684 /* shmem_symlink() */
1685 if (mapping->a_ops != &shmem_aops)
1686 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001687 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001688 goto alloc_nohuge;
1689 if (shmem_huge == SHMEM_HUGE_FORCE)
1690 goto alloc_huge;
1691 switch (sbinfo->huge) {
1692 loff_t i_size;
1693 pgoff_t off;
1694 case SHMEM_HUGE_NEVER:
1695 goto alloc_nohuge;
1696 case SHMEM_HUGE_WITHIN_SIZE:
1697 off = round_up(index, HPAGE_PMD_NR);
1698 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1699 if (i_size >= HPAGE_PMD_SIZE &&
1700 i_size >> PAGE_SHIFT >= off)
1701 goto alloc_huge;
1702 /* fallthrough */
1703 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001704 if (sgp_huge == SGP_HUGE)
1705 goto alloc_huge;
1706 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001707 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001710alloc_huge:
1711 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1712 index, true);
1713 if (IS_ERR(page)) {
1714alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1715 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001716 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001717 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001718 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001719 error = PTR_ERR(page);
1720 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001721 if (error != -ENOSPC)
1722 goto failed;
1723 /*
1724 * Try to reclaim some spece by splitting a huge page
1725 * beyond i_size on the filesystem.
1726 */
1727 while (retry--) {
1728 int ret;
1729 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1730 if (ret == SHRINK_STOP)
1731 break;
1732 if (ret)
1733 goto alloc_nohuge;
1734 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001735 goto failed;
1736 }
1737
1738 if (PageTransHuge(page))
1739 hindex = round_down(index, HPAGE_PMD_NR);
1740 else
1741 hindex = index;
1742
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001743 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001744 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001745
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001746 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001747 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001748 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001749 goto unacct;
1750 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1751 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001752 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001753 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001754 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001755 radix_tree_preload_end();
1756 }
1757 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001758 mem_cgroup_cancel_charge(page, memcg,
1759 PageTransHuge(page));
1760 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001761 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001762 mem_cgroup_commit_charge(page, memcg, false,
1763 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001764 lru_cache_add_anon(page);
1765
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001766 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001767 info->alloced += 1 << compound_order(page);
1768 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001769 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001770 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001771 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001772
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001773 if (PageTransHuge(page) &&
1774 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1775 hindex + HPAGE_PMD_NR - 1) {
1776 /*
1777 * Part of the huge page is beyond i_size: subject
1778 * to shrink under memory pressure.
1779 */
1780 spin_lock(&sbinfo->shrinklist_lock);
Cong Wange2286912017-08-10 15:24:24 -07001781 /*
1782 * _careful to defend against unlocked access to
1783 * ->shrink_list in shmem_unused_huge_shrink()
1784 */
1785 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001786 list_add_tail(&info->shrinklist,
1787 &sbinfo->shrinklist);
1788 sbinfo->shrinklist_len++;
1789 }
1790 spin_unlock(&sbinfo->shrinklist_lock);
1791 }
1792
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001793 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001794 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1795 */
1796 if (sgp == SGP_FALLOC)
1797 sgp = SGP_WRITE;
1798clear:
1799 /*
1800 * Let SGP_WRITE caller clear ends if write does not fill page;
1801 * but SGP_FALLOC on a page fallocated earlier must initialize
1802 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001803 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001804 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1805 struct page *head = compound_head(page);
1806 int i;
1807
1808 for (i = 0; i < (1 << compound_order(head)); i++) {
1809 clear_highpage(head + i);
1810 flush_dcache_page(head + i);
1811 }
1812 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001813 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001814 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001815
Hugh Dickins54af60422011-08-03 16:21:24 -07001816 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001817 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001818 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001819 if (alloced) {
1820 ClearPageDirty(page);
1821 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001822 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001823 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001824 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001825 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001826 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001827 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001828 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001829 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001830 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001831
Nick Piggind0217ac2007-07-19 01:47:03 -07001832 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001833 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001835unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001836 if (sbinfo->max_blocks)
1837 percpu_counter_sub(&sbinfo->used_blocks,
1838 1 << compound_order(page));
1839 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1840
1841 if (PageTransHuge(page)) {
1842 unlock_page(page);
1843 put_page(page);
1844 goto alloc_nohuge;
1845 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001846failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001847 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001848 error = -EEXIST;
1849unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001850 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001851 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001852 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001853 }
1854 if (error == -ENOSPC && !once++) {
1855 info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001856 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001857 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001858 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001859 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001860 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001861 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001862 goto repeat;
1863 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864}
1865
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001866/*
1867 * This is like autoremove_wake_function, but it removes the wait queue
1868 * entry unconditionally - even if something else had already woken the
1869 * target.
1870 */
1871static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1872{
1873 int ret = default_wake_function(wait, mode, sync, key);
1874 list_del_init(&wait->task_list);
1875 return ret;
1876}
1877
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1879{
Al Viro496ad9a2013-01-23 17:07:38 -05001880 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001881 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001882 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001884 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001886 /*
1887 * Trinity finds that probing a hole which tmpfs is punching can
1888 * prevent the hole-punch from ever completing: which in turn
1889 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001890 * faulting pages into the hole while it's being punched. Although
1891 * shmem_undo_range() does remove the additions, it may be unable to
1892 * keep up, as each new page needs its own unmap_mapping_range() call,
1893 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1894 *
1895 * It does not matter if we sometimes reach this check just before the
1896 * hole-punch begins, so that one fault then races with the punch:
1897 * we just need to make racing faults a rare case.
1898 *
1899 * The implementation below would be much simpler if we just used a
1900 * standard mutex or completion: but we cannot take i_mutex in fault,
1901 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001902 */
1903 if (unlikely(inode->i_private)) {
1904 struct shmem_falloc *shmem_falloc;
1905
1906 spin_lock(&inode->i_lock);
1907 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001908 if (shmem_falloc &&
1909 shmem_falloc->waitq &&
1910 vmf->pgoff >= shmem_falloc->start &&
1911 vmf->pgoff < shmem_falloc->next) {
1912 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001913 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001914
1915 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001916 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1917 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001918 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001919 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001920 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001921 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001922
1923 shmem_falloc_waitq = shmem_falloc->waitq;
1924 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1925 TASK_UNINTERRUPTIBLE);
1926 spin_unlock(&inode->i_lock);
1927 schedule();
1928
1929 /*
1930 * shmem_falloc_waitq points into the shmem_fallocate()
1931 * stack of the hole-punching task: shmem_falloc_waitq
1932 * is usually invalid by the time we reach here, but
1933 * finish_wait() does not dereference it in that case;
1934 * though i_lock needed lest racing with wake_up_all().
1935 */
1936 spin_lock(&inode->i_lock);
1937 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1938 spin_unlock(&inode->i_lock);
1939 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001940 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001941 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001942 }
1943
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001944 sgp = SGP_CACHE;
1945 if (vma->vm_flags & VM_HUGEPAGE)
1946 sgp = SGP_HUGE;
1947 else if (vma->vm_flags & VM_NOHUGEPAGE)
1948 sgp = SGP_NOHUGE;
1949
1950 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001951 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 if (error)
1953 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001954 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955}
1956
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001957unsigned long shmem_get_unmapped_area(struct file *file,
1958 unsigned long uaddr, unsigned long len,
1959 unsigned long pgoff, unsigned long flags)
1960{
1961 unsigned long (*get_area)(struct file *,
1962 unsigned long, unsigned long, unsigned long, unsigned long);
1963 unsigned long addr;
1964 unsigned long offset;
1965 unsigned long inflated_len;
1966 unsigned long inflated_addr;
1967 unsigned long inflated_offset;
1968
1969 if (len > TASK_SIZE)
1970 return -ENOMEM;
1971
1972 get_area = current->mm->get_unmapped_area;
1973 addr = get_area(file, uaddr, len, pgoff, flags);
1974
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001975 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001976 return addr;
1977 if (IS_ERR_VALUE(addr))
1978 return addr;
1979 if (addr & ~PAGE_MASK)
1980 return addr;
1981 if (addr > TASK_SIZE - len)
1982 return addr;
1983
1984 if (shmem_huge == SHMEM_HUGE_DENY)
1985 return addr;
1986 if (len < HPAGE_PMD_SIZE)
1987 return addr;
1988 if (flags & MAP_FIXED)
1989 return addr;
1990 /*
1991 * Our priority is to support MAP_SHARED mapped hugely;
1992 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1993 * But if caller specified an address hint, respect that as before.
1994 */
1995 if (uaddr)
1996 return addr;
1997
1998 if (shmem_huge != SHMEM_HUGE_FORCE) {
1999 struct super_block *sb;
2000
2001 if (file) {
2002 VM_BUG_ON(file->f_op != &shmem_file_operations);
2003 sb = file_inode(file)->i_sb;
2004 } else {
2005 /*
2006 * Called directly from mm/mmap.c, or drivers/char/mem.c
2007 * for "/dev/zero", to create a shared anonymous object.
2008 */
2009 if (IS_ERR(shm_mnt))
2010 return addr;
2011 sb = shm_mnt->mnt_sb;
2012 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002013 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002014 return addr;
2015 }
2016
2017 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2018 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2019 return addr;
2020 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2021 return addr;
2022
2023 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2024 if (inflated_len > TASK_SIZE)
2025 return addr;
2026 if (inflated_len < len)
2027 return addr;
2028
2029 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2030 if (IS_ERR_VALUE(inflated_addr))
2031 return addr;
2032 if (inflated_addr & ~PAGE_MASK)
2033 return addr;
2034
2035 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2036 inflated_addr += offset - inflated_offset;
2037 if (inflated_offset > offset)
2038 inflated_addr += HPAGE_PMD_SIZE;
2039
2040 if (inflated_addr > TASK_SIZE - len)
2041 return addr;
2042 return inflated_addr;
2043}
2044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002046static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047{
Al Viro496ad9a2013-01-23 17:07:38 -05002048 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002049 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050}
2051
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002052static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2053 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
Al Viro496ad9a2013-01-23 17:07:38 -05002055 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002056 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002058 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2059 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060}
2061#endif
2062
2063int shmem_lock(struct file *file, int lock, struct user_struct *user)
2064{
Al Viro496ad9a2013-01-23 17:07:38 -05002065 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 struct shmem_inode_info *info = SHMEM_I(inode);
2067 int retval = -ENOMEM;
2068
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002069 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 if (lock && !(info->flags & VM_LOCKED)) {
2071 if (!user_shm_lock(inode->i_size, user))
2072 goto out_nomem;
2073 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002074 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 }
2076 if (!lock && (info->flags & VM_LOCKED) && user) {
2077 user_shm_unlock(inode->i_size, user);
2078 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002079 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 }
2081 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002082
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002084 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return retval;
2086}
2087
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002088static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089{
2090 file_accessed(file);
2091 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002092 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002093 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2094 (vma->vm_end & HPAGE_PMD_MASK)) {
2095 khugepaged_enter(vma, vma->vm_flags);
2096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 return 0;
2098}
2099
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002100static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002101 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102{
2103 struct inode *inode;
2104 struct shmem_inode_info *info;
2105 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2106
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002107 if (shmem_reserve_inode(sb))
2108 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
2110 inode = new_inode(sb);
2111 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002112 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002113 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002115 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002116 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 info = SHMEM_I(inode);
2118 memset(info, 0, (char *)inode - (char *)info);
2119 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002120 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002121 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002122 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002124 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002125 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
2127 switch (mode & S_IFMT) {
2128 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002129 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 init_special_inode(inode, mode, dev);
2131 break;
2132 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002133 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 inode->i_op = &shmem_inode_operations;
2135 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002136 mpol_shared_policy_init(&info->policy,
2137 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 break;
2139 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002140 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 /* Some things misbehave if size == 0 on a directory */
2142 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2143 inode->i_op = &shmem_dir_inode_operations;
2144 inode->i_fop = &simple_dir_operations;
2145 break;
2146 case S_IFLNK:
2147 /*
2148 * Must not load anything in the rbtree,
2149 * mpol_free_shared_policy will not be called.
2150 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002151 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 break;
2153 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002154 } else
2155 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 return inode;
2157}
2158
Johannes Weiner0cd61442014-04-03 14:47:46 -07002159bool shmem_mapping(struct address_space *mapping)
2160{
Sasha Levinf0774d82015-02-23 05:38:00 -05002161 if (!mapping->host)
2162 return false;
2163
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002164 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002165}
2166
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002168static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002169static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002171#ifdef CONFIG_TMPFS_XATTR
2172static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2173#else
2174#define shmem_initxattrs NULL
2175#endif
2176
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002178shmem_write_begin(struct file *file, struct address_space *mapping,
2179 loff_t pos, unsigned len, unsigned flags,
2180 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181{
Nick Piggin800d15a2007-10-16 01:25:03 -07002182 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002183 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002184 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002185
2186 /* i_mutex is held by caller */
2187 if (unlikely(info->seals)) {
2188 if (info->seals & F_SEAL_WRITE)
2189 return -EPERM;
2190 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2191 return -EPERM;
2192 }
2193
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002194 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002195}
2196
2197static int
2198shmem_write_end(struct file *file, struct address_space *mapping,
2199 loff_t pos, unsigned len, unsigned copied,
2200 struct page *page, void *fsdata)
2201{
2202 struct inode *inode = mapping->host;
2203
Hugh Dickinsd3602442008-02-04 22:28:44 -08002204 if (pos + copied > inode->i_size)
2205 i_size_write(inode, pos + copied);
2206
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002207 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002208 struct page *head = compound_head(page);
2209 if (PageTransCompound(page)) {
2210 int i;
2211
2212 for (i = 0; i < HPAGE_PMD_NR; i++) {
2213 if (head + i == page)
2214 continue;
2215 clear_highpage(head + i);
2216 flush_dcache_page(head + i);
2217 }
2218 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002219 if (copied < PAGE_SIZE) {
2220 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002221 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002222 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002223 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002224 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002225 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002226 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002227 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002228 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002229
Nick Piggin800d15a2007-10-16 01:25:03 -07002230 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231}
2232
Al Viro2ba5bbe2014-04-02 20:00:02 -04002233static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234{
Al Viro6e58e792014-02-03 17:07:03 -05002235 struct file *file = iocb->ki_filp;
2236 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002238 pgoff_t index;
2239 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002240 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002241 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002242 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002243 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002244
2245 /*
2246 * Might this read be for a stacking filesystem? Then when reading
2247 * holes of a sparse file, we actually need to allocate those pages,
2248 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2249 */
Al Viro777eda22014-12-17 04:46:46 -05002250 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002251 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002253 index = *ppos >> PAGE_SHIFT;
2254 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
2256 for (;;) {
2257 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002258 pgoff_t end_index;
2259 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 loff_t i_size = i_size_read(inode);
2261
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002262 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 if (index > end_index)
2264 break;
2265 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002266 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 if (nr <= offset)
2268 break;
2269 }
2270
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002271 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002272 if (error) {
2273 if (error == -EINVAL)
2274 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 break;
2276 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002277 if (page) {
2278 if (sgp == SGP_CACHE)
2279 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002280 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
2283 /*
2284 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002285 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002287 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002289 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002291 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 if (nr <= offset) {
2293 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002294 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 break;
2296 }
2297 }
2298 nr -= offset;
2299
2300 if (page) {
2301 /*
2302 * If users can be writing to this page using arbitrary
2303 * virtual addresses, take care about potential aliasing
2304 * before reading the page on the kernel side.
2305 */
2306 if (mapping_writably_mapped(mapping))
2307 flush_dcache_page(page);
2308 /*
2309 * Mark the page accessed if we read the beginning.
2310 */
2311 if (!offset)
2312 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002313 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002315 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
2318 /*
2319 * Ok, we have the page, and it's up-to-date, so
2320 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002322 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002323 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002325 index += offset >> PAGE_SHIFT;
2326 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002328 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002329 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 break;
Al Viro6e58e792014-02-03 17:07:03 -05002331 if (ret < nr) {
2332 error = -EFAULT;
2333 break;
2334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 cond_resched();
2336 }
2337
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002338 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002339 file_accessed(file);
2340 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341}
2342
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002343/*
2344 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2345 */
2346static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002347 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002348{
2349 struct page *page;
2350 struct pagevec pvec;
2351 pgoff_t indices[PAGEVEC_SIZE];
2352 bool done = false;
2353 int i;
2354
2355 pagevec_init(&pvec, 0);
2356 pvec.nr = 1; /* start small: we may be there already */
2357 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002358 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002359 pvec.nr, pvec.pages, indices);
2360 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002361 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002362 index = end;
2363 break;
2364 }
2365 for (i = 0; i < pvec.nr; i++, index++) {
2366 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002367 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002368 done = true;
2369 break;
2370 }
2371 index = indices[i];
2372 }
2373 page = pvec.pages[i];
2374 if (page && !radix_tree_exceptional_entry(page)) {
2375 if (!PageUptodate(page))
2376 page = NULL;
2377 }
2378 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002379 (page && whence == SEEK_DATA) ||
2380 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002381 done = true;
2382 break;
2383 }
2384 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002385 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002386 pagevec_release(&pvec);
2387 pvec.nr = PAGEVEC_SIZE;
2388 cond_resched();
2389 }
2390 return index;
2391}
2392
Andrew Morton965c8e52012-12-17 15:59:39 -08002393static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002394{
2395 struct address_space *mapping = file->f_mapping;
2396 struct inode *inode = mapping->host;
2397 pgoff_t start, end;
2398 loff_t new_offset;
2399
Andrew Morton965c8e52012-12-17 15:59:39 -08002400 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2401 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002402 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002403 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002404 /* We're holding i_mutex so we can access i_size directly */
2405
2406 if (offset < 0)
2407 offset = -EINVAL;
2408 else if (offset >= inode->i_size)
2409 offset = -ENXIO;
2410 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002411 start = offset >> PAGE_SHIFT;
2412 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002413 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002414 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002415 if (new_offset > offset) {
2416 if (new_offset < inode->i_size)
2417 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002418 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002419 offset = -ENXIO;
2420 else
2421 offset = inode->i_size;
2422 }
2423 }
2424
Hugh Dickins387aae62013-08-04 11:30:25 -07002425 if (offset >= 0)
2426 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002427 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002428 return offset;
2429}
2430
David Herrmann05f65b52014-08-08 14:25:36 -07002431/*
2432 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2433 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2434 */
2435#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2436#define LAST_SCAN 4 /* about 150ms max */
2437
2438static void shmem_tag_pins(struct address_space *mapping)
2439{
2440 struct radix_tree_iter iter;
2441 void **slot;
2442 pgoff_t start;
2443 struct page *page;
2444
2445 lru_add_drain();
2446 start = 0;
2447 rcu_read_lock();
2448
David Herrmann05f65b52014-08-08 14:25:36 -07002449 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2450 page = radix_tree_deref_slot(slot);
2451 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002452 if (radix_tree_deref_retry(page)) {
2453 slot = radix_tree_iter_retry(&iter);
2454 continue;
2455 }
David Herrmann05f65b52014-08-08 14:25:36 -07002456 } else if (page_count(page) - page_mapcount(page) > 1) {
2457 spin_lock_irq(&mapping->tree_lock);
2458 radix_tree_tag_set(&mapping->page_tree, iter.index,
2459 SHMEM_TAG_PINNED);
2460 spin_unlock_irq(&mapping->tree_lock);
2461 }
2462
2463 if (need_resched()) {
2464 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002465 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002466 }
2467 }
2468 rcu_read_unlock();
2469}
2470
2471/*
2472 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2473 * via get_user_pages(), drivers might have some pending I/O without any active
2474 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2475 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2476 * them to be dropped.
2477 * The caller must guarantee that no new user will acquire writable references
2478 * to those pages to avoid races.
2479 */
David Herrmann40e041a2014-08-08 14:25:27 -07002480static int shmem_wait_for_pins(struct address_space *mapping)
2481{
David Herrmann05f65b52014-08-08 14:25:36 -07002482 struct radix_tree_iter iter;
2483 void **slot;
2484 pgoff_t start;
2485 struct page *page;
2486 int error, scan;
2487
2488 shmem_tag_pins(mapping);
2489
2490 error = 0;
2491 for (scan = 0; scan <= LAST_SCAN; scan++) {
2492 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2493 break;
2494
2495 if (!scan)
2496 lru_add_drain_all();
2497 else if (schedule_timeout_killable((HZ << scan) / 200))
2498 scan = LAST_SCAN;
2499
2500 start = 0;
2501 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002502 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2503 start, SHMEM_TAG_PINNED) {
2504
2505 page = radix_tree_deref_slot(slot);
2506 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002507 if (radix_tree_deref_retry(page)) {
2508 slot = radix_tree_iter_retry(&iter);
2509 continue;
2510 }
David Herrmann05f65b52014-08-08 14:25:36 -07002511
2512 page = NULL;
2513 }
2514
2515 if (page &&
2516 page_count(page) - page_mapcount(page) != 1) {
2517 if (scan < LAST_SCAN)
2518 goto continue_resched;
2519
2520 /*
2521 * On the last scan, we clean up all those tags
2522 * we inserted; but make a note that we still
2523 * found pages pinned.
2524 */
2525 error = -EBUSY;
2526 }
2527
2528 spin_lock_irq(&mapping->tree_lock);
2529 radix_tree_tag_clear(&mapping->page_tree,
2530 iter.index, SHMEM_TAG_PINNED);
2531 spin_unlock_irq(&mapping->tree_lock);
2532continue_resched:
2533 if (need_resched()) {
2534 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002535 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002536 }
2537 }
2538 rcu_read_unlock();
2539 }
2540
2541 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002542}
2543
2544#define F_ALL_SEALS (F_SEAL_SEAL | \
2545 F_SEAL_SHRINK | \
2546 F_SEAL_GROW | \
2547 F_SEAL_WRITE)
2548
2549int shmem_add_seals(struct file *file, unsigned int seals)
2550{
2551 struct inode *inode = file_inode(file);
2552 struct shmem_inode_info *info = SHMEM_I(inode);
2553 int error;
2554
2555 /*
2556 * SEALING
2557 * Sealing allows multiple parties to share a shmem-file but restrict
2558 * access to a specific subset of file operations. Seals can only be
2559 * added, but never removed. This way, mutually untrusted parties can
2560 * share common memory regions with a well-defined policy. A malicious
2561 * peer can thus never perform unwanted operations on a shared object.
2562 *
2563 * Seals are only supported on special shmem-files and always affect
2564 * the whole underlying inode. Once a seal is set, it may prevent some
2565 * kinds of access to the file. Currently, the following seals are
2566 * defined:
2567 * SEAL_SEAL: Prevent further seals from being set on this file
2568 * SEAL_SHRINK: Prevent the file from shrinking
2569 * SEAL_GROW: Prevent the file from growing
2570 * SEAL_WRITE: Prevent write access to the file
2571 *
2572 * As we don't require any trust relationship between two parties, we
2573 * must prevent seals from being removed. Therefore, sealing a file
2574 * only adds a given set of seals to the file, it never touches
2575 * existing seals. Furthermore, the "setting seals"-operation can be
2576 * sealed itself, which basically prevents any further seal from being
2577 * added.
2578 *
2579 * Semantics of sealing are only defined on volatile files. Only
2580 * anonymous shmem files support sealing. More importantly, seals are
2581 * never written to disk. Therefore, there's no plan to support it on
2582 * other file types.
2583 */
2584
2585 if (file->f_op != &shmem_file_operations)
2586 return -EINVAL;
2587 if (!(file->f_mode & FMODE_WRITE))
2588 return -EPERM;
2589 if (seals & ~(unsigned int)F_ALL_SEALS)
2590 return -EINVAL;
2591
Al Viro59551022016-01-22 15:40:57 -05002592 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002593
2594 if (info->seals & F_SEAL_SEAL) {
2595 error = -EPERM;
2596 goto unlock;
2597 }
2598
2599 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2600 error = mapping_deny_writable(file->f_mapping);
2601 if (error)
2602 goto unlock;
2603
2604 error = shmem_wait_for_pins(file->f_mapping);
2605 if (error) {
2606 mapping_allow_writable(file->f_mapping);
2607 goto unlock;
2608 }
2609 }
2610
2611 info->seals |= seals;
2612 error = 0;
2613
2614unlock:
Al Viro59551022016-01-22 15:40:57 -05002615 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002616 return error;
2617}
2618EXPORT_SYMBOL_GPL(shmem_add_seals);
2619
2620int shmem_get_seals(struct file *file)
2621{
2622 if (file->f_op != &shmem_file_operations)
2623 return -EINVAL;
2624
2625 return SHMEM_I(file_inode(file))->seals;
2626}
2627EXPORT_SYMBOL_GPL(shmem_get_seals);
2628
2629long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2630{
2631 long error;
2632
2633 switch (cmd) {
2634 case F_ADD_SEALS:
2635 /* disallow upper 32bit */
2636 if (arg > UINT_MAX)
2637 return -EINVAL;
2638
2639 error = shmem_add_seals(file, arg);
2640 break;
2641 case F_GET_SEALS:
2642 error = shmem_get_seals(file);
2643 break;
2644 default:
2645 error = -EINVAL;
2646 break;
2647 }
2648
2649 return error;
2650}
2651
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002652static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2653 loff_t len)
2654{
Al Viro496ad9a2013-01-23 17:07:38 -05002655 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002656 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002657 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002658 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002659 pgoff_t start, index, end;
2660 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002661
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002662 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2663 return -EOPNOTSUPP;
2664
Al Viro59551022016-01-22 15:40:57 -05002665 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002666
2667 if (mode & FALLOC_FL_PUNCH_HOLE) {
2668 struct address_space *mapping = file->f_mapping;
2669 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2670 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002671 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002672
David Herrmann40e041a2014-08-08 14:25:27 -07002673 /* protected by i_mutex */
2674 if (info->seals & F_SEAL_WRITE) {
2675 error = -EPERM;
2676 goto out;
2677 }
2678
Hugh Dickins8e205f72014-07-23 14:00:10 -07002679 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002680 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2681 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2682 spin_lock(&inode->i_lock);
2683 inode->i_private = &shmem_falloc;
2684 spin_unlock(&inode->i_lock);
2685
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002686 if ((u64)unmap_end > (u64)unmap_start)
2687 unmap_mapping_range(mapping, unmap_start,
2688 1 + unmap_end - unmap_start, 0);
2689 shmem_truncate_range(inode, offset, offset + len - 1);
2690 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002691
2692 spin_lock(&inode->i_lock);
2693 inode->i_private = NULL;
2694 wake_up_all(&shmem_falloc_waitq);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002695 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002696 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002697 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002698 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002699 }
2700
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002701 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2702 error = inode_newsize_ok(inode, offset + len);
2703 if (error)
2704 goto out;
2705
David Herrmann40e041a2014-08-08 14:25:27 -07002706 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2707 error = -EPERM;
2708 goto out;
2709 }
2710
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002711 start = offset >> PAGE_SHIFT;
2712 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002713 /* Try to avoid a swapstorm if len is impossible to satisfy */
2714 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2715 error = -ENOSPC;
2716 goto out;
2717 }
2718
Hugh Dickins8e205f72014-07-23 14:00:10 -07002719 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002720 shmem_falloc.start = start;
2721 shmem_falloc.next = start;
2722 shmem_falloc.nr_falloced = 0;
2723 shmem_falloc.nr_unswapped = 0;
2724 spin_lock(&inode->i_lock);
2725 inode->i_private = &shmem_falloc;
2726 spin_unlock(&inode->i_lock);
2727
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002728 for (index = start; index < end; index++) {
2729 struct page *page;
2730
2731 /*
2732 * Good, the fallocate(2) manpage permits EINTR: we may have
2733 * been interrupted because we are using up too much memory.
2734 */
2735 if (signal_pending(current))
2736 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002737 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2738 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002739 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002740 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002741 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002742 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002743 if (index > start) {
2744 shmem_undo_range(inode,
2745 (loff_t)start << PAGE_SHIFT,
2746 ((loff_t)index << PAGE_SHIFT) - 1, true);
2747 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002748 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002749 }
2750
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002751 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002752 * Inform shmem_writepage() how far we have reached.
2753 * No need for lock or barrier: we have the page lock.
2754 */
2755 shmem_falloc.next++;
2756 if (!PageUptodate(page))
2757 shmem_falloc.nr_falloced++;
2758
2759 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002760 * If !PageUptodate, leave it that way so that freeable pages
2761 * can be recognized if we need to rollback on error later.
2762 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002763 * than free the pages we are allocating (and SGP_CACHE pages
2764 * might still be clean: we now need to mark those dirty too).
2765 */
2766 set_page_dirty(page);
2767 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002768 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002769 cond_resched();
2770 }
2771
2772 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2773 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002774 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002775undone:
2776 spin_lock(&inode->i_lock);
2777 inode->i_private = NULL;
2778 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002779out:
Al Viro59551022016-01-22 15:40:57 -05002780 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002781 return error;
2782}
2783
David Howells726c3342006-06-23 02:02:58 -07002784static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785{
David Howells726c3342006-06-23 02:02:58 -07002786 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
2788 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002789 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002791 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002793 buf->f_bavail =
2794 buf->f_bfree = sbinfo->max_blocks -
2795 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002796 }
2797 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 buf->f_files = sbinfo->max_inodes;
2799 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 }
2801 /* else leave those fields 0 like simple_statfs */
2802 return 0;
2803}
2804
2805/*
2806 * File creation. Allocate an inode, and we're done..
2807 */
2808static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002809shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002811 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 int error = -ENOSPC;
2813
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002814 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002816 error = simple_acl_create(dir, inode);
2817 if (error)
2818 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002819 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002820 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002821 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002822 if (error && error != -EOPNOTSUPP)
2823 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002824
Al Viro718deb62009-12-16 19:35:36 -05002825 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002827 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 d_instantiate(dentry, inode);
2829 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 }
2831 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002832out_iput:
2833 iput(inode);
2834 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835}
2836
Al Viro60545d02013-06-07 01:20:27 -04002837static int
2838shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2839{
2840 struct inode *inode;
2841 int error = -ENOSPC;
2842
2843 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2844 if (inode) {
2845 error = security_inode_init_security(inode, dir,
2846 NULL,
2847 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002848 if (error && error != -EOPNOTSUPP)
2849 goto out_iput;
2850 error = simple_acl_create(dir, inode);
2851 if (error)
2852 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002853 d_tmpfile(dentry, inode);
2854 }
2855 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002856out_iput:
2857 iput(inode);
2858 return error;
Al Viro60545d02013-06-07 01:20:27 -04002859}
2860
Al Viro18bb1db2011-07-26 01:41:39 -04002861static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862{
2863 int error;
2864
2865 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2866 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002867 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 return 0;
2869}
2870
Al Viro4acdaf22011-07-26 01:42:34 -04002871static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002872 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873{
2874 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2875}
2876
2877/*
2878 * Link a file..
2879 */
2880static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2881{
David Howells75c3cfa2015-03-17 22:26:12 +00002882 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002883 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884
2885 /*
2886 * No ordinary (disk based) filesystem counts links as inodes;
2887 * but each new link needs a new dentry, pinning lowmem, and
2888 * tmpfs dentries cannot be pruned until they are unlinked.
2889 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002890 ret = shmem_reserve_inode(inode->i_sb);
2891 if (ret)
2892 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
2894 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002895 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002896 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002897 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 dget(dentry); /* Extra pinning count for the created dentry */
2899 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002900out:
2901 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902}
2903
2904static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2905{
David Howells75c3cfa2015-03-17 22:26:12 +00002906 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002908 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2909 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910
2911 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002912 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002913 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 dput(dentry); /* Undo the count from "create" - this does all the work */
2915 return 0;
2916}
2917
2918static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2919{
2920 if (!simple_empty(dentry))
2921 return -ENOTEMPTY;
2922
David Howells75c3cfa2015-03-17 22:26:12 +00002923 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002924 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 return shmem_unlink(dir, dentry);
2926}
2927
Miklos Szeredi37456772014-07-23 15:15:34 +02002928static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2929{
David Howellse36cb0b2015-01-29 12:02:35 +00002930 bool old_is_dir = d_is_dir(old_dentry);
2931 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002932
2933 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2934 if (old_is_dir) {
2935 drop_nlink(old_dir);
2936 inc_nlink(new_dir);
2937 } else {
2938 drop_nlink(new_dir);
2939 inc_nlink(old_dir);
2940 }
2941 }
2942 old_dir->i_ctime = old_dir->i_mtime =
2943 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002944 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002945 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02002946
2947 return 0;
2948}
2949
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002950static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2951{
2952 struct dentry *whiteout;
2953 int error;
2954
2955 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2956 if (!whiteout)
2957 return -ENOMEM;
2958
2959 error = shmem_mknod(old_dir, whiteout,
2960 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2961 dput(whiteout);
2962 if (error)
2963 return error;
2964
2965 /*
2966 * Cheat and hash the whiteout while the old dentry is still in
2967 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2968 *
2969 * d_lookup() will consistently find one of them at this point,
2970 * not sure which one, but that isn't even important.
2971 */
2972 d_rehash(whiteout);
2973 return 0;
2974}
2975
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976/*
2977 * The VFS layer already does all the dentry stuff for rename,
2978 * we just have to decrement the usage count for the target if
2979 * it exists so that the VFS layer correctly free's it when it
2980 * gets overwritten.
2981 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002982static 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 -07002983{
David Howells75c3cfa2015-03-17 22:26:12 +00002984 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 int they_are_dirs = S_ISDIR(inode->i_mode);
2986
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002987 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002988 return -EINVAL;
2989
Miklos Szeredi37456772014-07-23 15:15:34 +02002990 if (flags & RENAME_EXCHANGE)
2991 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2992
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 if (!simple_empty(new_dentry))
2994 return -ENOTEMPTY;
2995
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002996 if (flags & RENAME_WHITEOUT) {
2997 int error;
2998
2999 error = shmem_whiteout(old_dir, old_dentry);
3000 if (error)
3001 return error;
3002 }
3003
David Howells75c3cfa2015-03-17 22:26:12 +00003004 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003006 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003007 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003008 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003011 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003012 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 }
3014
3015 old_dir->i_size -= BOGO_DIRENT_SIZE;
3016 new_dir->i_size += BOGO_DIRENT_SIZE;
3017 old_dir->i_ctime = old_dir->i_mtime =
3018 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003019 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 return 0;
3021}
3022
3023static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3024{
3025 int error;
3026 int len;
3027 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003028 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 struct shmem_inode_info *info;
3030
3031 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003032 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 return -ENAMETOOLONG;
3034
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003035 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 if (!inode)
3037 return -ENOSPC;
3038
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003039 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003040 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003041 if (error) {
3042 if (error != -EOPNOTSUPP) {
3043 iput(inode);
3044 return error;
3045 }
3046 error = 0;
3047 }
3048
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 info = SHMEM_I(inode);
3050 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003051 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003052 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3053 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003054 iput(inode);
3055 return -ENOMEM;
3056 }
3057 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003059 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003060 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 if (error) {
3062 iput(inode);
3063 return error;
3064 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003065 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003067 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003068 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003070 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003071 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003074 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 d_instantiate(dentry, inode);
3076 dget(dentry);
3077 return 0;
3078}
3079
Al Virofceef392015-12-29 15:58:39 -05003080static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081{
Al Virofceef392015-12-29 15:58:39 -05003082 mark_page_accessed(arg);
3083 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084}
3085
Al Viro6b255392015-11-17 10:20:54 -05003086static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003087 struct inode *inode,
3088 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003091 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003092 if (!dentry) {
3093 page = find_get_page(inode->i_mapping, 0);
3094 if (!page)
3095 return ERR_PTR(-ECHILD);
3096 if (!PageUptodate(page)) {
3097 put_page(page);
3098 return ERR_PTR(-ECHILD);
3099 }
3100 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003101 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003102 if (error)
3103 return ERR_PTR(error);
3104 unlock_page(page);
3105 }
Al Virofceef392015-12-29 15:58:39 -05003106 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003107 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108}
3109
Eric Parisb09e0fa2011-05-24 17:12:39 -07003110#ifdef CONFIG_TMPFS_XATTR
3111/*
3112 * Superblocks without xattr inode operations may get some security.* xattr
3113 * support from the LSM "for free". As soon as we have any other xattrs
3114 * like ACLs, we also need to implement the security.* handlers at
3115 * filesystem level, though.
3116 */
3117
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003118/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003119 * Callback for security_inode_init_security() for acquiring xattrs.
3120 */
3121static int shmem_initxattrs(struct inode *inode,
3122 const struct xattr *xattr_array,
3123 void *fs_info)
3124{
3125 struct shmem_inode_info *info = SHMEM_I(inode);
3126 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003127 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003128 size_t len;
3129
3130 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003131 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003132 if (!new_xattr)
3133 return -ENOMEM;
3134
3135 len = strlen(xattr->name) + 1;
3136 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3137 GFP_KERNEL);
3138 if (!new_xattr->name) {
3139 kfree(new_xattr);
3140 return -ENOMEM;
3141 }
3142
3143 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3144 XATTR_SECURITY_PREFIX_LEN);
3145 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3146 xattr->name, len);
3147
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003148 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003149 }
3150
3151 return 0;
3152}
3153
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003154static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003155 struct dentry *unused, struct inode *inode,
3156 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003157{
Al Virob2968212016-04-10 20:48:24 -04003158 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003159
3160 name = xattr_full_name(handler, name);
3161 return simple_xattr_get(&info->xattrs, name, buffer, size);
3162}
3163
3164static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003165 struct dentry *unused, struct inode *inode,
3166 const char *name, const void *value,
3167 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003168{
Al Viro59301222016-05-27 10:19:30 -04003169 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003170
3171 name = xattr_full_name(handler, name);
3172 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3173}
3174
3175static const struct xattr_handler shmem_security_xattr_handler = {
3176 .prefix = XATTR_SECURITY_PREFIX,
3177 .get = shmem_xattr_handler_get,
3178 .set = shmem_xattr_handler_set,
3179};
3180
3181static const struct xattr_handler shmem_trusted_xattr_handler = {
3182 .prefix = XATTR_TRUSTED_PREFIX,
3183 .get = shmem_xattr_handler_get,
3184 .set = shmem_xattr_handler_set,
3185};
3186
Eric Parisb09e0fa2011-05-24 17:12:39 -07003187static const struct xattr_handler *shmem_xattr_handlers[] = {
3188#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003189 &posix_acl_access_xattr_handler,
3190 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003191#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003192 &shmem_security_xattr_handler,
3193 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003194 NULL
3195};
3196
Eric Parisb09e0fa2011-05-24 17:12:39 -07003197static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3198{
David Howells75c3cfa2015-03-17 22:26:12 +00003199 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003200 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003201}
3202#endif /* CONFIG_TMPFS_XATTR */
3203
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003204static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003206 .get_link = simple_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,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003209#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210};
3211
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003212static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003214 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003215#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003216 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003217#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003218};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003219
David M. Grimes91828a42006-10-17 00:09:45 -07003220static struct dentry *shmem_get_parent(struct dentry *child)
3221{
3222 return ERR_PTR(-ESTALE);
3223}
3224
3225static int shmem_match(struct inode *ino, void *vfh)
3226{
3227 __u32 *fh = vfh;
3228 __u64 inum = fh[2];
3229 inum = (inum << 32) | fh[1];
3230 return ino->i_ino == inum && fh[0] == ino->i_generation;
3231}
3232
Christoph Hellwig480b1162007-10-21 16:42:13 -07003233static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3234 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003235{
David M. Grimes91828a42006-10-17 00:09:45 -07003236 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003237 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003238 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003239
Christoph Hellwig480b1162007-10-21 16:42:13 -07003240 if (fh_len < 3)
3241 return NULL;
3242
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003243 inum = fid->raw[2];
3244 inum = (inum << 32) | fid->raw[1];
3245
Christoph Hellwig480b1162007-10-21 16:42:13 -07003246 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3247 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003248 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003249 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003250 iput(inode);
3251 }
3252
Christoph Hellwig480b1162007-10-21 16:42:13 -07003253 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003254}
3255
Al Virob0b03822012-04-02 14:34:06 -04003256static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3257 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003258{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303259 if (*len < 3) {
3260 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003261 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303262 }
David M. Grimes91828a42006-10-17 00:09:45 -07003263
Al Viro1d3382cb2010-10-23 15:19:20 -04003264 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003265 /* Unfortunately insert_inode_hash is not idempotent,
3266 * so as we hash inodes here rather than at creation
3267 * time, we need a lock to ensure we only try
3268 * to do it once
3269 */
3270 static DEFINE_SPINLOCK(lock);
3271 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003272 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003273 __insert_inode_hash(inode,
3274 inode->i_ino + inode->i_generation);
3275 spin_unlock(&lock);
3276 }
3277
3278 fh[0] = inode->i_generation;
3279 fh[1] = inode->i_ino;
3280 fh[2] = ((__u64)inode->i_ino) >> 32;
3281
3282 *len = 3;
3283 return 1;
3284}
3285
Christoph Hellwig39655162007-10-21 16:42:17 -07003286static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003287 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003288 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003289 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003290};
3291
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003292static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3293 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294{
3295 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003296 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003297 uid_t uid;
3298 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003300 while (options != NULL) {
3301 this_char = options;
3302 for (;;) {
3303 /*
3304 * NUL-terminate this option: unfortunately,
3305 * mount options form a comma-separated list,
3306 * but mpol's nodelist may also contain commas.
3307 */
3308 options = strchr(options, ',');
3309 if (options == NULL)
3310 break;
3311 options++;
3312 if (!isdigit(*options)) {
3313 options[-1] = '\0';
3314 break;
3315 }
3316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 if (!*this_char)
3318 continue;
3319 if ((value = strchr(this_char,'=')) != NULL) {
3320 *value++ = 0;
3321 } else {
Joe Perches11705322016-03-17 14:19:50 -07003322 pr_err("tmpfs: No value for mount option '%s'\n",
3323 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003324 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 }
3326
3327 if (!strcmp(this_char,"size")) {
3328 unsigned long long size;
3329 size = memparse(value,&rest);
3330 if (*rest == '%') {
3331 size <<= PAGE_SHIFT;
3332 size *= totalram_pages;
3333 do_div(size, 100);
3334 rest++;
3335 }
3336 if (*rest)
3337 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003338 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003339 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003341 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342 if (*rest)
3343 goto bad_val;
3344 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003345 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 if (*rest)
3347 goto bad_val;
3348 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003349 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003351 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 if (*rest)
3353 goto bad_val;
3354 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003355 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003357 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 if (*rest)
3359 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003360 sbinfo->uid = make_kuid(current_user_ns(), uid);
3361 if (!uid_valid(sbinfo->uid))
3362 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003364 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003366 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 if (*rest)
3368 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003369 sbinfo->gid = make_kgid(current_user_ns(), gid);
3370 if (!gid_valid(sbinfo->gid))
3371 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003372#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003373 } else if (!strcmp(this_char, "huge")) {
3374 int huge;
3375 huge = shmem_parse_huge(value);
3376 if (huge < 0)
3377 goto bad_val;
3378 if (!has_transparent_hugepage() &&
3379 huge != SHMEM_HUGE_NEVER)
3380 goto bad_val;
3381 sbinfo->huge = huge;
3382#endif
3383#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003384 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003385 mpol_put(mpol);
3386 mpol = NULL;
3387 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003388 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003389#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 } else {
Joe Perches11705322016-03-17 14:19:50 -07003391 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003392 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 }
3394 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003395 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 return 0;
3397
3398bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003399 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003401error:
3402 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 return 1;
3404
3405}
3406
3407static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3408{
3409 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003410 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003411 unsigned long inodes;
3412 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413
Greg Thelen5f001102013-02-22 16:36:01 -08003414 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003415 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003416 return error;
3417
3418 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003419 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003420 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003421 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003422 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003423 goto out;
3424 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003425 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003426 * but we must separately disallow unlimited->limited, because
3427 * in that case we have no record of how much is already in use.
3428 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003429 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003430 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003431 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003432 goto out;
3433
3434 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003435 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003436 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003437 sbinfo->max_inodes = config.max_inodes;
3438 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003439
Greg Thelen5f001102013-02-22 16:36:01 -08003440 /*
3441 * Preserve previous mempolicy unless mpol remount option was specified.
3442 */
3443 if (config.mpol) {
3444 mpol_put(sbinfo->mpol);
3445 sbinfo->mpol = config.mpol; /* transfers initial ref */
3446 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003447out:
3448 spin_unlock(&sbinfo->stat_lock);
3449 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003451
Al Viro34c80b12011-12-08 21:32:45 -05003452static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003453{
Al Viro34c80b12011-12-08 21:32:45 -05003454 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003455
3456 if (sbinfo->max_blocks != shmem_default_max_blocks())
3457 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003458 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003459 if (sbinfo->max_inodes != shmem_default_max_inodes())
3460 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3461 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003462 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003463 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3464 seq_printf(seq, ",uid=%u",
3465 from_kuid_munged(&init_user_ns, sbinfo->uid));
3466 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3467 seq_printf(seq, ",gid=%u",
3468 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003469#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003470 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3471 if (sbinfo->huge)
3472 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3473#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003474 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003475 return 0;
3476}
David Herrmann9183df22014-08-08 14:25:29 -07003477
3478#define MFD_NAME_PREFIX "memfd:"
3479#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3480#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3481
3482#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3483
3484SYSCALL_DEFINE2(memfd_create,
3485 const char __user *, uname,
3486 unsigned int, flags)
3487{
3488 struct shmem_inode_info *info;
3489 struct file *file;
3490 int fd, error;
3491 char *name;
3492 long len;
3493
3494 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3495 return -EINVAL;
3496
3497 /* length includes terminating zero */
3498 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3499 if (len <= 0)
3500 return -EFAULT;
3501 if (len > MFD_NAME_MAX_LEN + 1)
3502 return -EINVAL;
3503
3504 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3505 if (!name)
3506 return -ENOMEM;
3507
3508 strcpy(name, MFD_NAME_PREFIX);
3509 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3510 error = -EFAULT;
3511 goto err_name;
3512 }
3513
3514 /* terminating-zero may have changed after strnlen_user() returned */
3515 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3516 error = -EFAULT;
3517 goto err_name;
3518 }
3519
3520 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3521 if (fd < 0) {
3522 error = fd;
3523 goto err_name;
3524 }
3525
3526 file = shmem_file_setup(name, 0, VM_NORESERVE);
3527 if (IS_ERR(file)) {
3528 error = PTR_ERR(file);
3529 goto err_fd;
3530 }
3531 info = SHMEM_I(file_inode(file));
3532 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3533 file->f_flags |= O_RDWR | O_LARGEFILE;
3534 if (flags & MFD_ALLOW_SEALING)
3535 info->seals &= ~F_SEAL_SEAL;
3536
3537 fd_install(fd, file);
3538 kfree(name);
3539 return fd;
3540
3541err_fd:
3542 put_unused_fd(fd);
3543err_name:
3544 kfree(name);
3545 return error;
3546}
3547
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003548#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549
3550static void shmem_put_super(struct super_block *sb)
3551{
Hugh Dickins602586a2010-08-17 15:23:56 -07003552 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3553
3554 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003555 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003556 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 sb->s_fs_info = NULL;
3558}
3559
Kay Sievers2b2af542009-04-30 15:23:42 +02003560int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561{
3562 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003563 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003564 int err = -ENOMEM;
3565
3566 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003567 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003568 L1_CACHE_BYTES), GFP_KERNEL);
3569 if (!sbinfo)
3570 return -ENOMEM;
3571
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003572 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003573 sbinfo->uid = current_fsuid();
3574 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003575 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003577#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 /*
3579 * Per default we only allow half of the physical ram per
3580 * tmpfs instance, limiting inodes to one per page of lowmem;
3581 * but the internal instance is left unlimited.
3582 */
Al Viroca4e0512013-08-31 12:57:10 -04003583 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003584 sbinfo->max_blocks = shmem_default_max_blocks();
3585 sbinfo->max_inodes = shmem_default_max_inodes();
3586 if (shmem_parse_options(data, sbinfo, false)) {
3587 err = -EINVAL;
3588 goto failed;
3589 }
Al Viroca4e0512013-08-31 12:57:10 -04003590 } else {
3591 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 }
David M. Grimes91828a42006-10-17 00:09:45 -07003593 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003594 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595#else
3596 sb->s_flags |= MS_NOUSER;
3597#endif
3598
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003599 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003600 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003601 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003602 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003603 spin_lock_init(&sbinfo->shrinklist_lock);
3604 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003605
Hugh Dickins285b2c42011-08-03 16:21:20 -07003606 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003607 sb->s_blocksize = PAGE_SIZE;
3608 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609 sb->s_magic = TMPFS_MAGIC;
3610 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003611 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003612#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003613 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003614#endif
3615#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003616 sb->s_flags |= MS_POSIXACL;
3617#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003618
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003619 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620 if (!inode)
3621 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003622 inode->i_uid = sbinfo->uid;
3623 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003624 sb->s_root = d_make_root(inode);
3625 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003626 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 return 0;
3628
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629failed:
3630 shmem_put_super(sb);
3631 return err;
3632}
3633
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003634static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635
3636static struct inode *shmem_alloc_inode(struct super_block *sb)
3637{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003638 struct shmem_inode_info *info;
3639 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3640 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003642 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643}
3644
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003645static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003646{
3647 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003648 if (S_ISLNK(inode->i_mode))
3649 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003650 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3651}
3652
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653static void shmem_destroy_inode(struct inode *inode)
3654{
Al Viro09208d12011-07-26 03:15:03 -04003655 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003657 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658}
3659
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003660static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003662 struct shmem_inode_info *info = foo;
3663 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664}
3665
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003666static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667{
3668 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3669 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003670 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671 return 0;
3672}
3673
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003674static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003676 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677}
3678
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003679static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003681 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003683 .write_begin = shmem_write_begin,
3684 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003686#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003687 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003688#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003689 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690};
3691
Helge Deller15ad7cd2006-12-06 20:40:36 -08003692static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003694 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003696 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003697 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003698 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003699 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003700 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003701 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003702 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703#endif
3704};
3705
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003706static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003707 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003708 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003709#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003710 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003711 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003712#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713};
3714
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003715static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716#ifdef CONFIG_TMPFS
3717 .create = shmem_create,
3718 .lookup = simple_lookup,
3719 .link = shmem_link,
3720 .unlink = shmem_unlink,
3721 .symlink = shmem_symlink,
3722 .mkdir = shmem_mkdir,
3723 .rmdir = shmem_rmdir,
3724 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003725 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003726 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003728#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003729 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003730#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003731#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003732 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003733 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003734#endif
3735};
3736
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003737static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003738#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003739 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003740#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003741#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003742 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003743 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003744#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745};
3746
Hugh Dickins759b9772007-03-05 00:30:28 -08003747static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748 .alloc_inode = shmem_alloc_inode,
3749 .destroy_inode = shmem_destroy_inode,
3750#ifdef CONFIG_TMPFS
3751 .statfs = shmem_statfs,
3752 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003753 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754#endif
Al Viro1f895f72010-06-05 19:10:41 -04003755 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 .drop_inode = generic_delete_inode,
3757 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003758#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3759 .nr_cached_objects = shmem_unused_huge_count,
3760 .free_cached_objects = shmem_unused_huge_scan,
3761#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762};
3763
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003764static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003765 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003766 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767#ifdef CONFIG_NUMA
3768 .set_policy = shmem_set_policy,
3769 .get_policy = shmem_get_policy,
3770#endif
3771};
3772
Al Viro3c26ff62010-07-25 11:46:36 +04003773static struct dentry *shmem_mount(struct file_system_type *fs_type,
3774 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775{
Al Viro3c26ff62010-07-25 11:46:36 +04003776 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777}
3778
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003779static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 .owner = THIS_MODULE,
3781 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003782 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003784 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003787int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788{
3789 int error;
3790
Rob Landley16203a72013-09-11 14:26:12 -07003791 /* If rootfs called this, don't re-init */
3792 if (shmem_inode_cachep)
3793 return 0;
3794
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003795 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796 if (error)
3797 goto out3;
3798
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003799 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003801 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 goto out2;
3803 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003804
Al Viroca4e0512013-08-31 12:57:10 -04003805 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 if (IS_ERR(shm_mnt)) {
3807 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003808 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 goto out1;
3810 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003811
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003812#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5d8b3cc2017-08-25 15:55:33 -07003813 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003814 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3815 else
3816 shmem_huge = 0; /* just in case it was patched */
3817#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 return 0;
3819
3820out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003821 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003823 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824out3:
3825 shm_mnt = ERR_PTR(error);
3826 return error;
3827}
Matt Mackall853ac432009-01-06 14:40:20 -08003828
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003829#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003830static ssize_t shmem_enabled_show(struct kobject *kobj,
3831 struct kobj_attribute *attr, char *buf)
3832{
3833 int values[] = {
3834 SHMEM_HUGE_ALWAYS,
3835 SHMEM_HUGE_WITHIN_SIZE,
3836 SHMEM_HUGE_ADVISE,
3837 SHMEM_HUGE_NEVER,
3838 SHMEM_HUGE_DENY,
3839 SHMEM_HUGE_FORCE,
3840 };
3841 int i, count;
3842
3843 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3844 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3845
3846 count += sprintf(buf + count, fmt,
3847 shmem_format_huge(values[i]));
3848 }
3849 buf[count - 1] = '\n';
3850 return count;
3851}
3852
3853static ssize_t shmem_enabled_store(struct kobject *kobj,
3854 struct kobj_attribute *attr, const char *buf, size_t count)
3855{
3856 char tmp[16];
3857 int huge;
3858
3859 if (count + 1 > sizeof(tmp))
3860 return -EINVAL;
3861 memcpy(tmp, buf, count);
3862 tmp[count] = '\0';
3863 if (count && tmp[count - 1] == '\n')
3864 tmp[count - 1] = '\0';
3865
3866 huge = shmem_parse_huge(tmp);
3867 if (huge == -EINVAL)
3868 return -EINVAL;
3869 if (!has_transparent_hugepage() &&
3870 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3871 return -EINVAL;
3872
3873 shmem_huge = huge;
Kirill A. Shutemov5d8b3cc2017-08-25 15:55:33 -07003874 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003875 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3876 return count;
3877}
3878
3879struct kobj_attribute shmem_enabled_attr =
3880 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003881#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003882
Arnd Bergmann3b337192016-08-10 16:27:44 -07003883#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003884bool shmem_huge_enabled(struct vm_area_struct *vma)
3885{
3886 struct inode *inode = file_inode(vma->vm_file);
3887 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3888 loff_t i_size;
3889 pgoff_t off;
3890
3891 if (shmem_huge == SHMEM_HUGE_FORCE)
3892 return true;
3893 if (shmem_huge == SHMEM_HUGE_DENY)
3894 return false;
3895 switch (sbinfo->huge) {
3896 case SHMEM_HUGE_NEVER:
3897 return false;
3898 case SHMEM_HUGE_ALWAYS:
3899 return true;
3900 case SHMEM_HUGE_WITHIN_SIZE:
3901 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3902 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3903 if (i_size >= HPAGE_PMD_SIZE &&
3904 i_size >> PAGE_SHIFT >= off)
3905 return true;
3906 case SHMEM_HUGE_ADVISE:
3907 /* TODO: implement fadvise() hints */
3908 return (vma->vm_flags & VM_HUGEPAGE);
3909 default:
3910 VM_BUG_ON(1);
3911 return false;
3912 }
3913}
Arnd Bergmann3b337192016-08-10 16:27:44 -07003914#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003915
Matt Mackall853ac432009-01-06 14:40:20 -08003916#else /* !CONFIG_SHMEM */
3917
3918/*
3919 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3920 *
3921 * This is intended for small system where the benefits of the full
3922 * shmem code (swap-backed and resource-limited) are outweighed by
3923 * their complexity. On systems without swap this code should be
3924 * effectively equivalent, but much lighter weight.
3925 */
3926
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003927static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003928 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003929 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003930 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003931 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003932};
3933
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003934int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003935{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003936 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003937
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003938 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003939 BUG_ON(IS_ERR(shm_mnt));
3940
3941 return 0;
3942}
3943
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003944int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003945{
3946 return 0;
3947}
3948
Hugh Dickins3f96b792009-09-21 17:03:37 -07003949int shmem_lock(struct file *file, int lock, struct user_struct *user)
3950{
3951 return 0;
3952}
3953
Hugh Dickins24513262012-01-20 14:34:21 -08003954void shmem_unlock_mapping(struct address_space *mapping)
3955{
3956}
3957
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003958#ifdef CONFIG_MMU
3959unsigned long shmem_get_unmapped_area(struct file *file,
3960 unsigned long addr, unsigned long len,
3961 unsigned long pgoff, unsigned long flags)
3962{
3963 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3964}
3965#endif
3966
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003967void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003968{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003969 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003970}
3971EXPORT_SYMBOL_GPL(shmem_truncate_range);
3972
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003973#define shmem_vm_ops generic_file_vm_ops
3974#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003975#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003976#define shmem_acct_size(flags, size) 0
3977#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003978
3979#endif /* CONFIG_SHMEM */
3980
3981/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982
Rasmus Villemoes19938e32016-10-07 17:01:01 -07003983static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003984 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003985};
3986
Eric Parisc7277092013-12-02 11:24:19 +00003987static struct file *__shmem_file_setup(const char *name, loff_t size,
3988 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989{
Al Viro6b4d0b22013-02-14 21:37:26 -05003990 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003992 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003993 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 struct qstr this;
3995
3996 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003997 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998
Hugh Dickins285b2c42011-08-03 16:21:20 -07003999 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 return ERR_PTR(-EINVAL);
4001
4002 if (shmem_acct_size(flags, size))
4003 return ERR_PTR(-ENOMEM);
4004
Al Viro6b4d0b22013-02-14 21:37:26 -05004005 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 this.name = name;
4007 this.len = strlen(name);
4008 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004009 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004010 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004011 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004012 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004014 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015
Al Viro6b4d0b22013-02-14 21:37:26 -05004016 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004017 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004019 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020
Eric Parisc7277092013-12-02 11:24:19 +00004021 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004022 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004024 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004025 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4026 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004027 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004028
Al Viro6b4d0b22013-02-14 21:37:26 -05004029 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004030 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004031 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004032 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004033
Al Viro6b4d0b22013-02-14 21:37:26 -05004034 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035
Linus Torvalds1da177e2005-04-16 15:20:36 -07004036put_memory:
4037 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004038put_path:
4039 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004040 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041}
Eric Parisc7277092013-12-02 11:24:19 +00004042
4043/**
4044 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4045 * kernel internal. There will be NO LSM permission checks against the
4046 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004047 * higher layer. The users are the big_key and shm implementations. LSM
4048 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004049 * @name: name for dentry (to be seen in /proc/<pid>/maps
4050 * @size: size to be set for the file
4051 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4052 */
4053struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4054{
4055 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4056}
4057
4058/**
4059 * shmem_file_setup - get an unlinked file living in tmpfs
4060 * @name: name for dentry (to be seen in /proc/<pid>/maps
4061 * @size: size to be set for the file
4062 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4063 */
4064struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4065{
4066 return __shmem_file_setup(name, size, flags, 0);
4067}
Keith Packard395e0dd2008-06-20 00:08:06 -07004068EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069
John Stultz6f193672011-12-15 18:51:04 -08004070void shmem_set_file(struct vm_area_struct *vma, struct file *file)
4071{
4072 if (vma->vm_file)
4073 fput(vma->vm_file);
4074 vma->vm_file = file;
4075 vma->vm_ops = &shmem_vm_ops;
4076}
4077
Randy Dunlap46711812008-03-19 17:00:41 -07004078/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4081 */
4082int shmem_zero_setup(struct vm_area_struct *vma)
4083{
4084 struct file *file;
4085 loff_t size = vma->vm_end - vma->vm_start;
4086
Hugh Dickins66fc1302015-06-14 09:48:09 -07004087 /*
4088 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4089 * between XFS directory reading and selinux: since this file is only
4090 * accessible to the user through its mapping, use S_PRIVATE flag to
4091 * bypass file security, in the same way as shmem_kernel_file_setup().
4092 */
4093 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 if (IS_ERR(file))
4095 return PTR_ERR(file);
4096
John Stultz6f193672011-12-15 18:51:04 -08004097 shmem_set_file(vma, file);
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004098
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004099 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004100 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4101 (vma->vm_end & HPAGE_PMD_MASK)) {
4102 khugepaged_enter(vma, vma->vm_flags);
4103 }
4104
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105 return 0;
4106}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004107
4108/**
4109 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4110 * @mapping: the page's address_space
4111 * @index: the page index
4112 * @gfp: the page allocator flags to use if allocating
4113 *
4114 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4115 * with any new page allocations done using the specified allocation flags.
4116 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4117 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4118 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4119 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004120 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4121 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004122 */
4123struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4124 pgoff_t index, gfp_t gfp)
4125{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004126#ifdef CONFIG_SHMEM
4127 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004128 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004129 int error;
4130
4131 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004132 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4133 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004134 if (error)
4135 page = ERR_PTR(error);
4136 else
4137 unlock_page(page);
4138 return page;
4139#else
4140 /*
4141 * The tiny !SHMEM case uses ramfs without swap
4142 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004143 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004144#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004145}
4146EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);