blob: 7f7748a0f9e1f738fd1ffcaceccdee2ae54d8d35 [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);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700273
274 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);
294}
295
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700296/*
297 * Replace item expected in radix tree by a new item, while holding tree lock.
298 */
299static int shmem_radix_tree_replace(struct address_space *mapping,
300 pgoff_t index, void *expected, void *replacement)
301{
302 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700303 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700304
305 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700306 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700307 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700308 if (!pslot)
309 return -ENOENT;
310 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700311 if (item != expected)
312 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700313 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700314 return 0;
315}
316
317/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700318 * Sometimes, before we decide whether to proceed or to fail, we must check
319 * that an entry was not already brought back from swap by a racing thread.
320 *
321 * Checking page is not enough: by the time a SwapCache page is locked, it
322 * might be reused, and again be SwapCache, using the same swap as before.
323 */
324static bool shmem_confirm_swap(struct address_space *mapping,
325 pgoff_t index, swp_entry_t swap)
326{
327 void *item;
328
329 rcu_read_lock();
330 item = radix_tree_lookup(&mapping->page_tree, index);
331 rcu_read_unlock();
332 return item == swp_to_radix_entry(swap);
333}
334
335/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700336 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
337 *
338 * SHMEM_HUGE_NEVER:
339 * disables huge pages for the mount;
340 * SHMEM_HUGE_ALWAYS:
341 * enables huge pages for the mount;
342 * SHMEM_HUGE_WITHIN_SIZE:
343 * only allocate huge pages if the page will be fully within i_size,
344 * also respect fadvise()/madvise() hints;
345 * SHMEM_HUGE_ADVISE:
346 * only allocate huge pages if requested with fadvise()/madvise();
347 */
348
349#define SHMEM_HUGE_NEVER 0
350#define SHMEM_HUGE_ALWAYS 1
351#define SHMEM_HUGE_WITHIN_SIZE 2
352#define SHMEM_HUGE_ADVISE 3
353
354/*
355 * Special values.
356 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
357 *
358 * SHMEM_HUGE_DENY:
359 * disables huge on shm_mnt and all mounts, for emergency use;
360 * SHMEM_HUGE_FORCE:
361 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
362 *
363 */
364#define SHMEM_HUGE_DENY (-1)
365#define SHMEM_HUGE_FORCE (-2)
366
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700367#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700368/* ifdef here to avoid bloating shmem.o when not necessary */
369
370int shmem_huge __read_mostly;
371
372static int shmem_parse_huge(const char *str)
373{
374 if (!strcmp(str, "never"))
375 return SHMEM_HUGE_NEVER;
376 if (!strcmp(str, "always"))
377 return SHMEM_HUGE_ALWAYS;
378 if (!strcmp(str, "within_size"))
379 return SHMEM_HUGE_WITHIN_SIZE;
380 if (!strcmp(str, "advise"))
381 return SHMEM_HUGE_ADVISE;
382 if (!strcmp(str, "deny"))
383 return SHMEM_HUGE_DENY;
384 if (!strcmp(str, "force"))
385 return SHMEM_HUGE_FORCE;
386 return -EINVAL;
387}
388
389static const char *shmem_format_huge(int huge)
390{
391 switch (huge) {
392 case SHMEM_HUGE_NEVER:
393 return "never";
394 case SHMEM_HUGE_ALWAYS:
395 return "always";
396 case SHMEM_HUGE_WITHIN_SIZE:
397 return "within_size";
398 case SHMEM_HUGE_ADVISE:
399 return "advise";
400 case SHMEM_HUGE_DENY:
401 return "deny";
402 case SHMEM_HUGE_FORCE:
403 return "force";
404 default:
405 VM_BUG_ON(1);
406 return "bad_val";
407 }
408}
409
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700410static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
411 struct shrink_control *sc, unsigned long nr_to_split)
412{
413 LIST_HEAD(list), *pos, *next;
414 struct inode *inode;
415 struct shmem_inode_info *info;
416 struct page *page;
417 unsigned long batch = sc ? sc->nr_to_scan : 128;
418 int removed = 0, split = 0;
419
420 if (list_empty(&sbinfo->shrinklist))
421 return SHRINK_STOP;
422
423 spin_lock(&sbinfo->shrinklist_lock);
424 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
425 info = list_entry(pos, struct shmem_inode_info, shrinklist);
426
427 /* pin the inode */
428 inode = igrab(&info->vfs_inode);
429
430 /* inode is about to be evicted */
431 if (!inode) {
432 list_del_init(&info->shrinklist);
433 removed++;
434 goto next;
435 }
436
437 /* Check if there's anything to gain */
438 if (round_up(inode->i_size, PAGE_SIZE) ==
439 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
440 list_del_init(&info->shrinklist);
441 removed++;
442 iput(inode);
443 goto next;
444 }
445
446 list_move(&info->shrinklist, &list);
447next:
448 if (!--batch)
449 break;
450 }
451 spin_unlock(&sbinfo->shrinklist_lock);
452
453 list_for_each_safe(pos, next, &list) {
454 int ret;
455
456 info = list_entry(pos, struct shmem_inode_info, shrinklist);
457 inode = &info->vfs_inode;
458
459 if (nr_to_split && split >= nr_to_split) {
460 iput(inode);
461 continue;
462 }
463
464 page = find_lock_page(inode->i_mapping,
465 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
466 if (!page)
467 goto drop;
468
469 if (!PageTransHuge(page)) {
470 unlock_page(page);
471 put_page(page);
472 goto drop;
473 }
474
475 ret = split_huge_page(page);
476 unlock_page(page);
477 put_page(page);
478
479 if (ret) {
480 /* split failed: leave it on the list */
481 iput(inode);
482 continue;
483 }
484
485 split++;
486drop:
487 list_del_init(&info->shrinklist);
488 removed++;
489 iput(inode);
490 }
491
492 spin_lock(&sbinfo->shrinklist_lock);
493 list_splice_tail(&list, &sbinfo->shrinklist);
494 sbinfo->shrinklist_len -= removed;
495 spin_unlock(&sbinfo->shrinklist_lock);
496
497 return split;
498}
499
500static long shmem_unused_huge_scan(struct super_block *sb,
501 struct shrink_control *sc)
502{
503 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
504
505 if (!READ_ONCE(sbinfo->shrinklist_len))
506 return SHRINK_STOP;
507
508 return shmem_unused_huge_shrink(sbinfo, sc, 0);
509}
510
511static long shmem_unused_huge_count(struct super_block *sb,
512 struct shrink_control *sc)
513{
514 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
515 return READ_ONCE(sbinfo->shrinklist_len);
516}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700517#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700518
519#define shmem_huge SHMEM_HUGE_DENY
520
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700521static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
522 struct shrink_control *sc, unsigned long nr_to_split)
523{
524 return 0;
525}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700526#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700527
528/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700529 * Like add_to_page_cache_locked, but error if expected item has gone.
530 */
531static int shmem_add_to_page_cache(struct page *page,
532 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700533 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700534{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700535 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700536
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700537 VM_BUG_ON_PAGE(PageTail(page), page);
538 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800539 VM_BUG_ON_PAGE(!PageLocked(page), page);
540 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700541 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700542
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700543 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700544 page->mapping = mapping;
545 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700546
Hugh Dickinsb065b432012-07-11 14:02:48 -0700547 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700548 if (PageTransHuge(page)) {
549 void __rcu **results;
550 pgoff_t idx;
551 int i;
552
553 error = 0;
554 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
555 &results, &idx, index, 1) &&
556 idx < index + HPAGE_PMD_NR) {
557 error = -EEXIST;
558 }
559
560 if (!error) {
561 for (i = 0; i < HPAGE_PMD_NR; i++) {
562 error = radix_tree_insert(&mapping->page_tree,
563 index + i, page + i);
564 VM_BUG_ON(error);
565 }
566 count_vm_event(THP_FILE_ALLOC);
567 }
568 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700569 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700570 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700571 error = shmem_radix_tree_replace(mapping, index, expected,
572 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700573 }
574
Hugh Dickinsb065b432012-07-11 14:02:48 -0700575 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700576 mapping->nrpages += nr;
577 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700578 __inc_node_page_state(page, NR_SHMEM_THPS);
579 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
580 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700581 spin_unlock_irq(&mapping->tree_lock);
582 } else {
583 page->mapping = NULL;
584 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700585 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700586 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700587 return error;
588}
589
590/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700591 * Like delete_from_page_cache, but substitutes swap for page.
592 */
593static void shmem_delete_from_page_cache(struct page *page, void *radswap)
594{
595 struct address_space *mapping = page->mapping;
596 int error;
597
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700598 VM_BUG_ON_PAGE(PageCompound(page), page);
599
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700600 spin_lock_irq(&mapping->tree_lock);
601 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
602 page->mapping = NULL;
603 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700604 __dec_node_page_state(page, NR_FILE_PAGES);
605 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700606 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300607 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700608 BUG_ON(error);
609}
610
611/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700612 * Remove swap entry from radix tree, free the swap and its page cache.
613 */
614static int shmem_free_swap(struct address_space *mapping,
615 pgoff_t index, void *radswap)
616{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700617 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700618
619 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700620 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700621 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700622 if (old != radswap)
623 return -ENOENT;
624 free_swap_and_cache(radix_to_swp_entry(radswap));
625 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700626}
627
628/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800629 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800630 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800631 *
632 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
633 * as long as the inode doesn't go away and racy results are not a problem.
634 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800635unsigned long shmem_partial_swap_usage(struct address_space *mapping,
636 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800637{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800638 struct radix_tree_iter iter;
639 void **slot;
640 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800641 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800642
643 rcu_read_lock();
644
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800645 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
646 if (iter.index >= end)
647 break;
648
649 page = radix_tree_deref_slot(slot);
650
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700651 if (radix_tree_deref_retry(page)) {
652 slot = radix_tree_iter_retry(&iter);
653 continue;
654 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800655
656 if (radix_tree_exceptional_entry(page))
657 swapped++;
658
659 if (need_resched()) {
660 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700661 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800662 }
663 }
664
665 rcu_read_unlock();
666
667 return swapped << PAGE_SHIFT;
668}
669
670/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800671 * Determine (in bytes) how many of the shmem object's pages mapped by the
672 * given vma is swapped out.
673 *
674 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
675 * as long as the inode doesn't go away and racy results are not a problem.
676 */
677unsigned long shmem_swap_usage(struct vm_area_struct *vma)
678{
679 struct inode *inode = file_inode(vma->vm_file);
680 struct shmem_inode_info *info = SHMEM_I(inode);
681 struct address_space *mapping = inode->i_mapping;
682 unsigned long swapped;
683
684 /* Be careful as we don't hold info->lock */
685 swapped = READ_ONCE(info->swapped);
686
687 /*
688 * The easier cases are when the shmem object has nothing in swap, or
689 * the vma maps it whole. Then we can simply use the stats that we
690 * already track.
691 */
692 if (!swapped)
693 return 0;
694
695 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
696 return swapped << PAGE_SHIFT;
697
698 /* Here comes the more involved part */
699 return shmem_partial_swap_usage(mapping,
700 linear_page_index(vma, vma->vm_start),
701 linear_page_index(vma, vma->vm_end));
702}
703
704/*
Hugh Dickins24513262012-01-20 14:34:21 -0800705 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
706 */
707void shmem_unlock_mapping(struct address_space *mapping)
708{
709 struct pagevec pvec;
710 pgoff_t indices[PAGEVEC_SIZE];
711 pgoff_t index = 0;
712
713 pagevec_init(&pvec, 0);
714 /*
715 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
716 */
717 while (!mapping_unevictable(mapping)) {
718 /*
719 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
720 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
721 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700722 pvec.nr = find_get_entries(mapping, index,
723 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800724 if (!pvec.nr)
725 break;
726 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700727 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800728 check_move_unevictable_pages(pvec.pages, pvec.nr);
729 pagevec_release(&pvec);
730 cond_resched();
731 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700732}
733
734/*
735 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700736 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700737 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700738static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
739 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700741 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300743 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
744 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
745 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
746 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700747 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700748 pgoff_t indices[PAGEVEC_SIZE];
749 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700750 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700751 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700753 if (lend == -1)
754 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700755
756 pagevec_init(&pvec, 0);
757 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700758 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700759 pvec.nr = find_get_entries(mapping, index,
760 min(end - index, (pgoff_t)PAGEVEC_SIZE),
761 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700762 if (!pvec.nr)
763 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700764 for (i = 0; i < pagevec_count(&pvec); i++) {
765 struct page *page = pvec.pages[i];
766
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700767 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700768 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700769 break;
770
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700771 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700772 if (unfalloc)
773 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700774 nr_swaps_freed += !shmem_free_swap(mapping,
775 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700776 continue;
777 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700778
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700779 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
780
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700781 if (!trylock_page(page))
782 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700783
784 if (PageTransTail(page)) {
785 /* Middle of THP: zero out the page */
786 clear_highpage(page);
787 unlock_page(page);
788 continue;
789 } else if (PageTransHuge(page)) {
790 if (index == round_down(end, HPAGE_PMD_NR)) {
791 /*
792 * Range ends in the middle of THP:
793 * zero out the page
794 */
795 clear_highpage(page);
796 unlock_page(page);
797 continue;
798 }
799 index += HPAGE_PMD_NR - 1;
800 i += HPAGE_PMD_NR - 1;
801 }
802
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700803 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700804 VM_BUG_ON_PAGE(PageTail(page), page);
805 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800806 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700807 truncate_inode_page(mapping, page);
808 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700809 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700810 unlock_page(page);
811 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700812 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800813 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700814 cond_resched();
815 index++;
816 }
817
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700818 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700819 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700820 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700821 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300822 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700823 if (start > end) {
824 top = partial_end;
825 partial_end = 0;
826 }
827 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700828 set_page_dirty(page);
829 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300830 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700831 }
832 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700833 if (partial_end) {
834 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700835 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700836 if (page) {
837 zero_user_segment(page, 0, partial_end);
838 set_page_dirty(page);
839 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300840 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700841 }
842 }
843 if (start >= end)
844 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700845
846 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700847 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700848 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700849
850 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700851 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700852 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700853 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700854 /* If all gone or hole-punch or unfalloc, we're done */
855 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700856 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700857 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700858 index = start;
859 continue;
860 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700861 for (i = 0; i < pagevec_count(&pvec); i++) {
862 struct page *page = pvec.pages[i];
863
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700864 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700865 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700866 break;
867
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700868 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700869 if (unfalloc)
870 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700871 if (shmem_free_swap(mapping, index, page)) {
872 /* Swap was replaced by page: retry */
873 index--;
874 break;
875 }
876 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700877 continue;
878 }
879
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700880 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700881
882 if (PageTransTail(page)) {
883 /* Middle of THP: zero out the page */
884 clear_highpage(page);
885 unlock_page(page);
886 /*
887 * Partial thp truncate due 'start' in middle
888 * of THP: don't need to look on these pages
889 * again on !pvec.nr restart.
890 */
891 if (index != round_down(end, HPAGE_PMD_NR))
892 start++;
893 continue;
894 } else if (PageTransHuge(page)) {
895 if (index == round_down(end, HPAGE_PMD_NR)) {
896 /*
897 * Range ends in the middle of THP:
898 * zero out the page
899 */
900 clear_highpage(page);
901 unlock_page(page);
902 continue;
903 }
904 index += HPAGE_PMD_NR - 1;
905 i += HPAGE_PMD_NR - 1;
906 }
907
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700908 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700909 VM_BUG_ON_PAGE(PageTail(page), page);
910 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800911 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700912 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700913 } else {
914 /* Page was replaced by swap: retry */
915 unlock_page(page);
916 index--;
917 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700918 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700919 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700920 unlock_page(page);
921 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700922 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800923 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700924 index++;
925 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700926
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700927 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700928 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700930 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700931}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700933void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
934{
935 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700936 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700938EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Yu Zhao44a30222015-09-08 15:03:33 -0700940static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
941 struct kstat *stat)
942{
943 struct inode *inode = dentry->d_inode;
944 struct shmem_inode_info *info = SHMEM_I(inode);
945
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800946 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700947 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800948 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700949 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800950 }
Yu Zhao44a30222015-09-08 15:03:33 -0700951 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700952 return 0;
953}
954
Hugh Dickins94c1e622011-06-27 16:18:03 -0700955static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956{
David Howells75c3cfa2015-03-17 22:26:12 +0000957 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700958 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700959 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 int error;
961
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200962 error = inode_change_ok(inode, attr);
963 if (error)
964 return error;
965
Hugh Dickins94c1e622011-06-27 16:18:03 -0700966 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
967 loff_t oldsize = inode->i_size;
968 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000969
David Herrmann40e041a2014-08-08 14:25:27 -0700970 /* protected by i_mutex */
971 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
972 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
973 return -EPERM;
974
Hugh Dickins94c1e622011-06-27 16:18:03 -0700975 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700976 error = shmem_reacct_size(SHMEM_I(inode)->flags,
977 oldsize, newsize);
978 if (error)
979 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700980 i_size_write(inode, newsize);
981 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
982 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700983 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700984 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800985 if (oldsize > holebegin)
986 unmap_mapping_range(inode->i_mapping,
987 holebegin, 0, 1);
988 if (info->alloced)
989 shmem_truncate_range(inode,
990 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700991 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800992 if (oldsize > holebegin)
993 unmap_mapping_range(inode->i_mapping,
994 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700995
996 /*
997 * Part of the huge page can be beyond i_size: subject
998 * to shrink under memory pressure.
999 */
1000 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1001 spin_lock(&sbinfo->shrinklist_lock);
1002 if (list_empty(&info->shrinklist)) {
1003 list_add_tail(&info->shrinklist,
1004 &sbinfo->shrinklist);
1005 sbinfo->shrinklist_len++;
1006 }
1007 spin_unlock(&sbinfo->shrinklist_lock);
1008 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 }
1011
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001012 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001013 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001014 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 return error;
1016}
1017
Al Viro1f895f72010-06-05 19:10:41 -04001018static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001021 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001023 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 shmem_unacct_size(info->flags, inode->i_size);
1025 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001026 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001027 if (!list_empty(&info->shrinklist)) {
1028 spin_lock(&sbinfo->shrinklist_lock);
1029 if (!list_empty(&info->shrinklist)) {
1030 list_del_init(&info->shrinklist);
1031 sbinfo->shrinklist_len--;
1032 }
1033 spin_unlock(&sbinfo->shrinklist_lock);
1034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001036 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001038 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
Al Viro3ed47db2016-01-22 18:08:52 -05001040 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001041
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001042 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001043 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001044 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001045 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001048/*
1049 * If swap found in inode, free it and move page from swapcache to filecache.
1050 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001051static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001052 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001054 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001055 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001056 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001057 gfp_t gfp;
1058 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001060 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001061 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001062 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001063 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001064
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001065 /*
1066 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001067 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001068 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001069 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001070 */
1071 if (shmem_swaplist.next != &info->swaplist)
1072 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001073
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001074 gfp = mapping_gfp_mask(mapping);
1075 if (shmem_should_replace_page(*pagep, gfp)) {
1076 mutex_unlock(&shmem_swaplist_mutex);
1077 error = shmem_replace_page(pagep, gfp, info, index);
1078 mutex_lock(&shmem_swaplist_mutex);
1079 /*
1080 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001081 * allocation, but the inode might have been freed while we
1082 * dropped it: although a racing shmem_evict_inode() cannot
1083 * complete without emptying the radix_tree, our page lock
1084 * on this swapcache page is not enough to prevent that -
1085 * free_swap_and_cache() of our swap entry will only
1086 * trylock_page(), removing swap from radix_tree whatever.
1087 *
1088 * We must not proceed to shmem_add_to_page_cache() if the
1089 * inode has been freed, but of course we cannot rely on
1090 * inode or mapping or info to check that. However, we can
1091 * safely check if our swap entry is still in use (and here
1092 * it can't have got reused for another page): if it's still
1093 * in use, then the inode cannot have been freed yet, and we
1094 * can safely proceed (if it's no longer in use, that tells
1095 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001096 */
1097 if (!page_swapcount(*pagep))
1098 error = -ENOENT;
1099 }
1100
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001101 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001102 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1103 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1104 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001105 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001106 if (!error)
1107 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001108 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001109 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001110 /*
1111 * Truncation and eviction use free_swap_and_cache(), which
1112 * only does trylock page: if we raced, best clean up here.
1113 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001114 delete_from_swap_cache(*pagep);
1115 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001116 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001117 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001118 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001119 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001120 swap_free(swap);
1121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001123 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124}
1125
1126/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001127 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001129int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001131 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001133 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001134 int error = 0;
1135
1136 /*
1137 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001138 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001139 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001140 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001141 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001142
1143 /*
1144 * Charge page using GFP_KERNEL while we can wait, before taking
1145 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1146 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001147 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001148 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1149 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001150 if (error)
1151 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001152 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001153 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001155 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001156 list_for_each_safe(this, next, &shmem_swaplist) {
1157 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001158 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001159 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001160 else
1161 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001162 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001163 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001164 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001165 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001167 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001168
Johannes Weiner00501b52014-08-08 14:19:20 -07001169 if (error) {
1170 if (error != -ENOMEM)
1171 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001172 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001173 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001174 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001175out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001176 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001177 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001178 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179}
1180
1181/*
1182 * Move the page from the page cache to the swap cache.
1183 */
1184static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1185{
1186 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001189 swp_entry_t swap;
1190 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001192 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 mapping = page->mapping;
1195 index = page->index;
1196 inode = mapping->host;
1197 info = SHMEM_I(inode);
1198 if (info->flags & VM_LOCKED)
1199 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001200 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 goto redirty;
1202
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001203 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001204 * Our capabilities prevent regular writeback or sync from ever calling
1205 * shmem_writepage; but a stacking filesystem might use ->writepage of
1206 * its underlying filesystem, in which case tmpfs should write out to
1207 * swap only in response to memory pressure, and not for the writeback
1208 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001209 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001210 if (!wbc->for_reclaim) {
1211 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1212 goto redirty;
1213 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001214
1215 /*
1216 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1217 * value into swapfile.c, the only way we can correctly account for a
1218 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001219 *
1220 * That's okay for a page already fallocated earlier, but if we have
1221 * not yet completed the fallocation, then (a) we want to keep track
1222 * of this page in case we have to undo it, and (b) it may not be a
1223 * good idea to continue anyway, once we're pushing into swap. So
1224 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001225 */
1226 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001227 if (inode->i_private) {
1228 struct shmem_falloc *shmem_falloc;
1229 spin_lock(&inode->i_lock);
1230 shmem_falloc = inode->i_private;
1231 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001232 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001233 index >= shmem_falloc->start &&
1234 index < shmem_falloc->next)
1235 shmem_falloc->nr_unswapped++;
1236 else
1237 shmem_falloc = NULL;
1238 spin_unlock(&inode->i_lock);
1239 if (shmem_falloc)
1240 goto redirty;
1241 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001242 clear_highpage(page);
1243 flush_dcache_page(page);
1244 SetPageUptodate(page);
1245 }
1246
Hugh Dickins48f170f2011-07-25 17:12:37 -07001247 swap = get_swap_page();
1248 if (!swap.val)
1249 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001250
Vladimir Davydov37e84352016-01-20 15:02:56 -08001251 if (mem_cgroup_try_charge_swap(page, swap))
1252 goto free_swap;
1253
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001254 /*
1255 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001256 * if it's not already there. Do it now before the page is
1257 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001258 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001259 * we've incremented swapped, because shmem_unuse_inode() will
1260 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001261 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001262 mutex_lock(&shmem_swaplist_mutex);
1263 if (list_empty(&info->swaplist))
1264 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001265
Hugh Dickins48f170f2011-07-25 17:12:37 -07001266 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001267 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001268 shmem_recalc_inode(inode);
1269 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001270 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001271
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001272 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001273 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1274
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001275 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001276 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001277 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 return 0;
1279 }
1280
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001281 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001282free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001283 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284redirty:
1285 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001286 if (wbc->for_reclaim)
1287 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1288 unlock_page(page);
1289 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290}
1291
Hugh Dickins75edd342016-05-19 17:12:44 -07001292#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001293static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001294{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001295 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001296
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001297 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001298 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001299
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001300 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001301
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001302 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001303}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001304
1305static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1306{
1307 struct mempolicy *mpol = NULL;
1308 if (sbinfo->mpol) {
1309 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1310 mpol = sbinfo->mpol;
1311 mpol_get(mpol);
1312 spin_unlock(&sbinfo->stat_lock);
1313 }
1314 return mpol;
1315}
Hugh Dickins75edd342016-05-19 17:12:44 -07001316#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1317static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1318{
1319}
1320static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1321{
1322 return NULL;
1323}
1324#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1325#ifndef CONFIG_NUMA
1326#define vm_policy vm_private_data
1327#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001328
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001329static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1330 struct shmem_inode_info *info, pgoff_t index)
1331{
1332 /* Create a pseudo vma that just contains the policy */
1333 vma->vm_start = 0;
1334 /* Bias interleave by inode number to distribute better across nodes */
1335 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1336 vma->vm_ops = NULL;
1337 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1338}
1339
1340static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1341{
1342 /* Drop reference taken by mpol_shared_policy_lookup() */
1343 mpol_cond_put(vma->vm_policy);
1344}
1345
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001346static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1347 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001350 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001352 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001353 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001354 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001355
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001356 return page;
1357}
Mel Gorman18a2f372012-12-05 14:01:41 -08001358
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001359static struct page *shmem_alloc_hugepage(gfp_t gfp,
1360 struct shmem_inode_info *info, pgoff_t index)
1361{
1362 struct vm_area_struct pvma;
1363 struct inode *inode = &info->vfs_inode;
1364 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001365 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001366 void __rcu **results;
1367 struct page *page;
1368
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001369 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001370 return NULL;
1371
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001372 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001373 rcu_read_lock();
1374 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1375 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1376 rcu_read_unlock();
1377 return NULL;
1378 }
1379 rcu_read_unlock();
1380
1381 shmem_pseudo_vma_init(&pvma, info, hindex);
1382 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1383 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1384 shmem_pseudo_vma_destroy(&pvma);
1385 if (page)
1386 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001387 return page;
1388}
1389
1390static struct page *shmem_alloc_page(gfp_t gfp,
1391 struct shmem_inode_info *info, pgoff_t index)
1392{
1393 struct vm_area_struct pvma;
1394 struct page *page;
1395
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001396 shmem_pseudo_vma_init(&pvma, info, index);
1397 page = alloc_page_vma(gfp, &pvma, 0);
1398 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001399
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001400 return page;
1401}
1402
1403static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1404 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1405 pgoff_t index, bool huge)
1406{
1407 struct page *page;
1408 int nr;
1409 int err = -ENOSPC;
1410
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001411 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001412 huge = false;
1413 nr = huge ? HPAGE_PMD_NR : 1;
1414
1415 if (shmem_acct_block(info->flags, nr))
1416 goto failed;
1417 if (sbinfo->max_blocks) {
1418 if (percpu_counter_compare(&sbinfo->used_blocks,
1419 sbinfo->max_blocks - nr) > 0)
1420 goto unacct;
1421 percpu_counter_add(&sbinfo->used_blocks, nr);
1422 }
1423
1424 if (huge)
1425 page = shmem_alloc_hugepage(gfp, info, index);
1426 else
1427 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001428 if (page) {
1429 __SetPageLocked(page);
1430 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001431 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001432 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001433
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001434 err = -ENOMEM;
1435 if (sbinfo->max_blocks)
1436 percpu_counter_add(&sbinfo->used_blocks, -nr);
1437unacct:
1438 shmem_unacct_blocks(info->flags, nr);
1439failed:
1440 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001444 * When a page is moved from swapcache to shmem filecache (either by the
1445 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1446 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1447 * ignorance of the mapping it belongs to. If that mapping has special
1448 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1449 * we may need to copy to a suitable page before moving to filecache.
1450 *
1451 * In a future release, this may well be extended to respect cpuset and
1452 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1453 * but for now it is a simple matter of zone.
1454 */
1455static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1456{
1457 return page_zonenum(page) > gfp_zone(gfp);
1458}
1459
1460static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1461 struct shmem_inode_info *info, pgoff_t index)
1462{
1463 struct page *oldpage, *newpage;
1464 struct address_space *swap_mapping;
1465 pgoff_t swap_index;
1466 int error;
1467
1468 oldpage = *pagep;
1469 swap_index = page_private(oldpage);
1470 swap_mapping = page_mapping(oldpage);
1471
1472 /*
1473 * We have arrived here because our zones are constrained, so don't
1474 * limit chance of success by further cpuset and node constraints.
1475 */
1476 gfp &= ~GFP_CONSTRAINT_MASK;
1477 newpage = shmem_alloc_page(gfp, info, index);
1478 if (!newpage)
1479 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001480
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001481 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001482 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001483 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001484
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001485 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001486 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001487 SetPageSwapCache(newpage);
1488
1489 /*
1490 * Our caller will very soon move newpage out of swapcache, but it's
1491 * a nice clean interface for us to replace oldpage by newpage there.
1492 */
1493 spin_lock_irq(&swap_mapping->tree_lock);
1494 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1495 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001496 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001497 __inc_node_page_state(newpage, NR_FILE_PAGES);
1498 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001499 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001500 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001501
Hugh Dickins0142ef62012-06-07 14:21:09 -07001502 if (unlikely(error)) {
1503 /*
1504 * Is this possible? I think not, now that our callers check
1505 * both PageSwapCache and page_private after getting page lock;
1506 * but be defensive. Reverse old to newpage for clear and free.
1507 */
1508 oldpage = newpage;
1509 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001510 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001511 lru_cache_add_anon(newpage);
1512 *pagep = newpage;
1513 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001514
1515 ClearPageSwapCache(oldpage);
1516 set_page_private(oldpage, 0);
1517
1518 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001519 put_page(oldpage);
1520 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001521 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001522}
1523
1524/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001525 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 *
1527 * If we allocate a new one we do not mark it dirty. That's up to the
1528 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001529 * entry since a page cannot live in both the swap and page cache.
1530 *
1531 * fault_mm and fault_type are only supplied by shmem_fault:
1532 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001534static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001535 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1536 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
1538 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001539 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001541 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001542 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001543 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001545 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001546 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001548 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001549 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001551 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001553 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1554 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001556 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001557 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001558 if (radix_tree_exceptional_entry(page)) {
1559 swap = radix_to_swp_entry(page);
1560 page = NULL;
1561 }
1562
Hugh Dickins75edd342016-05-19 17:12:44 -07001563 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001564 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001565 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001566 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001567 }
1568
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001569 if (page && sgp == SGP_WRITE)
1570 mark_page_accessed(page);
1571
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001572 /* fallocated page? */
1573 if (page && !PageUptodate(page)) {
1574 if (sgp != SGP_READ)
1575 goto clear;
1576 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001577 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001578 page = NULL;
1579 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001580 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001581 *pagep = page;
1582 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001586 * Fast cache lookup did not find it:
1587 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001589 info = SHMEM_I(inode);
1590 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001591 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001592
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 if (swap.val) {
1594 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001595 page = lookup_swap_cache(swap);
1596 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001597 /* Or update major stats only when swapin succeeds?? */
1598 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001599 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001600 count_vm_event(PGMAJFAULT);
1601 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1602 }
1603 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001604 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001605 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001606 error = -ENOMEM;
1607 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
1610
1611 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001612 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001613 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001614 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001615 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001616 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001617 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001618 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001620 goto failed;
1621 }
1622 wait_on_page_writeback(page);
1623
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001624 if (shmem_should_replace_page(page, gfp)) {
1625 error = shmem_replace_page(&page, gfp, info, index);
1626 if (error)
1627 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 }
1629
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001630 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001631 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001632 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001633 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001634 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001635 /*
1636 * We already confirmed swap under page lock, and make
1637 * no memory allocation here, so usually no possibility
1638 * of error; but free_swap_and_cache() only trylocks a
1639 * page, so it is just possible that the entry has been
1640 * truncated or holepunched since swap was confirmed.
1641 * shmem_undo_range() will have done some of the
1642 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001643 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001644 * Reset swap.val? No, leave it so "failed" goes back to
1645 * "repeat": reading a hole and writing should succeed.
1646 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001647 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001648 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001649 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001650 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001651 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001652 if (error)
1653 goto failed;
1654
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001655 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001656
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001657 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001658 info->swapped--;
1659 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001660 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001661
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001662 if (sgp == SGP_WRITE)
1663 mark_page_accessed(page);
1664
Hugh Dickins27ab7002011-07-25 17:12:36 -07001665 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001666 set_page_dirty(page);
1667 swap_free(swap);
1668
Hugh Dickins54af60422011-08-03 16:21:24 -07001669 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001670 /* shmem_symlink() */
1671 if (mapping->a_ops != &shmem_aops)
1672 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001673 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001674 goto alloc_nohuge;
1675 if (shmem_huge == SHMEM_HUGE_FORCE)
1676 goto alloc_huge;
1677 switch (sbinfo->huge) {
1678 loff_t i_size;
1679 pgoff_t off;
1680 case SHMEM_HUGE_NEVER:
1681 goto alloc_nohuge;
1682 case SHMEM_HUGE_WITHIN_SIZE:
1683 off = round_up(index, HPAGE_PMD_NR);
1684 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1685 if (i_size >= HPAGE_PMD_SIZE &&
1686 i_size >> PAGE_SHIFT >= off)
1687 goto alloc_huge;
1688 /* fallthrough */
1689 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001690 if (sgp_huge == SGP_HUGE)
1691 goto alloc_huge;
1692 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001693 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001696alloc_huge:
1697 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1698 index, true);
1699 if (IS_ERR(page)) {
1700alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1701 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001702 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001703 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001704 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001705 error = PTR_ERR(page);
1706 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001707 if (error != -ENOSPC)
1708 goto failed;
1709 /*
1710 * Try to reclaim some spece by splitting a huge page
1711 * beyond i_size on the filesystem.
1712 */
1713 while (retry--) {
1714 int ret;
1715 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1716 if (ret == SHRINK_STOP)
1717 break;
1718 if (ret)
1719 goto alloc_nohuge;
1720 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001721 goto failed;
1722 }
1723
1724 if (PageTransHuge(page))
1725 hindex = round_down(index, HPAGE_PMD_NR);
1726 else
1727 hindex = index;
1728
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001729 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001730 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001731
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001732 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001733 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001734 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001735 goto unacct;
1736 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1737 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001738 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001739 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001740 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001741 radix_tree_preload_end();
1742 }
1743 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001744 mem_cgroup_cancel_charge(page, memcg,
1745 PageTransHuge(page));
1746 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001747 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001748 mem_cgroup_commit_charge(page, memcg, false,
1749 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001750 lru_cache_add_anon(page);
1751
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001752 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001753 info->alloced += 1 << compound_order(page);
1754 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001755 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001756 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001757 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001758
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001759 if (PageTransHuge(page) &&
1760 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1761 hindex + HPAGE_PMD_NR - 1) {
1762 /*
1763 * Part of the huge page is beyond i_size: subject
1764 * to shrink under memory pressure.
1765 */
1766 spin_lock(&sbinfo->shrinklist_lock);
1767 if (list_empty(&info->shrinklist)) {
1768 list_add_tail(&info->shrinklist,
1769 &sbinfo->shrinklist);
1770 sbinfo->shrinklist_len++;
1771 }
1772 spin_unlock(&sbinfo->shrinklist_lock);
1773 }
1774
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001775 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001776 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1777 */
1778 if (sgp == SGP_FALLOC)
1779 sgp = SGP_WRITE;
1780clear:
1781 /*
1782 * Let SGP_WRITE caller clear ends if write does not fill page;
1783 * but SGP_FALLOC on a page fallocated earlier must initialize
1784 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001785 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001786 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1787 struct page *head = compound_head(page);
1788 int i;
1789
1790 for (i = 0; i < (1 << compound_order(head)); i++) {
1791 clear_highpage(head + i);
1792 flush_dcache_page(head + i);
1793 }
1794 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001795 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001796 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001797
Hugh Dickins54af60422011-08-03 16:21:24 -07001798 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001799 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001800 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001801 if (alloced) {
1802 ClearPageDirty(page);
1803 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001804 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001805 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001806 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001807 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001808 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001809 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001810 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001811 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001812 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001813
Nick Piggind0217ac2007-07-19 01:47:03 -07001814 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001815 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001817unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001818 if (sbinfo->max_blocks)
1819 percpu_counter_sub(&sbinfo->used_blocks,
1820 1 << compound_order(page));
1821 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1822
1823 if (PageTransHuge(page)) {
1824 unlock_page(page);
1825 put_page(page);
1826 goto alloc_nohuge;
1827 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001828failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001829 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001830 error = -EEXIST;
1831unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001832 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001833 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001834 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001835 }
1836 if (error == -ENOSPC && !once++) {
1837 info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001838 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001839 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001840 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001841 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001842 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001843 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001844 goto repeat;
1845 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846}
1847
1848static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1849{
Al Viro496ad9a2013-01-23 17:07:38 -05001850 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001851 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001852 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001854 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001856 /*
1857 * Trinity finds that probing a hole which tmpfs is punching can
1858 * prevent the hole-punch from ever completing: which in turn
1859 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001860 * faulting pages into the hole while it's being punched. Although
1861 * shmem_undo_range() does remove the additions, it may be unable to
1862 * keep up, as each new page needs its own unmap_mapping_range() call,
1863 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1864 *
1865 * It does not matter if we sometimes reach this check just before the
1866 * hole-punch begins, so that one fault then races with the punch:
1867 * we just need to make racing faults a rare case.
1868 *
1869 * The implementation below would be much simpler if we just used a
1870 * standard mutex or completion: but we cannot take i_mutex in fault,
1871 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001872 */
1873 if (unlikely(inode->i_private)) {
1874 struct shmem_falloc *shmem_falloc;
1875
1876 spin_lock(&inode->i_lock);
1877 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001878 if (shmem_falloc &&
1879 shmem_falloc->waitq &&
1880 vmf->pgoff >= shmem_falloc->start &&
1881 vmf->pgoff < shmem_falloc->next) {
1882 wait_queue_head_t *shmem_falloc_waitq;
1883 DEFINE_WAIT(shmem_fault_wait);
1884
1885 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001886 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1887 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001888 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001889 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001890 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001891 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001892
1893 shmem_falloc_waitq = shmem_falloc->waitq;
1894 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1895 TASK_UNINTERRUPTIBLE);
1896 spin_unlock(&inode->i_lock);
1897 schedule();
1898
1899 /*
1900 * shmem_falloc_waitq points into the shmem_fallocate()
1901 * stack of the hole-punching task: shmem_falloc_waitq
1902 * is usually invalid by the time we reach here, but
1903 * finish_wait() does not dereference it in that case;
1904 * though i_lock needed lest racing with wake_up_all().
1905 */
1906 spin_lock(&inode->i_lock);
1907 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1908 spin_unlock(&inode->i_lock);
1909 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001910 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001911 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001912 }
1913
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001914 sgp = SGP_CACHE;
1915 if (vma->vm_flags & VM_HUGEPAGE)
1916 sgp = SGP_HUGE;
1917 else if (vma->vm_flags & VM_NOHUGEPAGE)
1918 sgp = SGP_NOHUGE;
1919
1920 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001921 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (error)
1923 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001924 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925}
1926
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001927unsigned long shmem_get_unmapped_area(struct file *file,
1928 unsigned long uaddr, unsigned long len,
1929 unsigned long pgoff, unsigned long flags)
1930{
1931 unsigned long (*get_area)(struct file *,
1932 unsigned long, unsigned long, unsigned long, unsigned long);
1933 unsigned long addr;
1934 unsigned long offset;
1935 unsigned long inflated_len;
1936 unsigned long inflated_addr;
1937 unsigned long inflated_offset;
1938
1939 if (len > TASK_SIZE)
1940 return -ENOMEM;
1941
1942 get_area = current->mm->get_unmapped_area;
1943 addr = get_area(file, uaddr, len, pgoff, flags);
1944
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001945 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001946 return addr;
1947 if (IS_ERR_VALUE(addr))
1948 return addr;
1949 if (addr & ~PAGE_MASK)
1950 return addr;
1951 if (addr > TASK_SIZE - len)
1952 return addr;
1953
1954 if (shmem_huge == SHMEM_HUGE_DENY)
1955 return addr;
1956 if (len < HPAGE_PMD_SIZE)
1957 return addr;
1958 if (flags & MAP_FIXED)
1959 return addr;
1960 /*
1961 * Our priority is to support MAP_SHARED mapped hugely;
1962 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1963 * But if caller specified an address hint, respect that as before.
1964 */
1965 if (uaddr)
1966 return addr;
1967
1968 if (shmem_huge != SHMEM_HUGE_FORCE) {
1969 struct super_block *sb;
1970
1971 if (file) {
1972 VM_BUG_ON(file->f_op != &shmem_file_operations);
1973 sb = file_inode(file)->i_sb;
1974 } else {
1975 /*
1976 * Called directly from mm/mmap.c, or drivers/char/mem.c
1977 * for "/dev/zero", to create a shared anonymous object.
1978 */
1979 if (IS_ERR(shm_mnt))
1980 return addr;
1981 sb = shm_mnt->mnt_sb;
1982 }
1983 if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER)
1984 return addr;
1985 }
1986
1987 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
1988 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
1989 return addr;
1990 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
1991 return addr;
1992
1993 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
1994 if (inflated_len > TASK_SIZE)
1995 return addr;
1996 if (inflated_len < len)
1997 return addr;
1998
1999 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2000 if (IS_ERR_VALUE(inflated_addr))
2001 return addr;
2002 if (inflated_addr & ~PAGE_MASK)
2003 return addr;
2004
2005 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2006 inflated_addr += offset - inflated_offset;
2007 if (inflated_offset > offset)
2008 inflated_addr += HPAGE_PMD_SIZE;
2009
2010 if (inflated_addr > TASK_SIZE - len)
2011 return addr;
2012 return inflated_addr;
2013}
2014
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002016static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Al Viro496ad9a2013-01-23 17:07:38 -05002018 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002019 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020}
2021
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002022static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2023 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024{
Al Viro496ad9a2013-01-23 17:07:38 -05002025 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002026 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002028 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2029 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031#endif
2032
2033int shmem_lock(struct file *file, int lock, struct user_struct *user)
2034{
Al Viro496ad9a2013-01-23 17:07:38 -05002035 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 struct shmem_inode_info *info = SHMEM_I(inode);
2037 int retval = -ENOMEM;
2038
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002039 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 if (lock && !(info->flags & VM_LOCKED)) {
2041 if (!user_shm_lock(inode->i_size, user))
2042 goto out_nomem;
2043 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002044 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 }
2046 if (!lock && (info->flags & VM_LOCKED) && user) {
2047 user_shm_unlock(inode->i_size, user);
2048 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002049 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 }
2051 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002054 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return retval;
2056}
2057
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002058static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059{
2060 file_accessed(file);
2061 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002062 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002063 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2064 (vma->vm_end & HPAGE_PMD_MASK)) {
2065 khugepaged_enter(vma, vma->vm_flags);
2066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 return 0;
2068}
2069
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002070static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002071 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072{
2073 struct inode *inode;
2074 struct shmem_inode_info *info;
2075 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2076
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002077 if (shmem_reserve_inode(sb))
2078 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
2080 inode = new_inode(sb);
2081 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002082 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002083 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07002086 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 info = SHMEM_I(inode);
2088 memset(info, 0, (char *)inode - (char *)info);
2089 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002090 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002091 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002092 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002094 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002095 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
2097 switch (mode & S_IFMT) {
2098 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002099 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 init_special_inode(inode, mode, dev);
2101 break;
2102 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002103 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 inode->i_op = &shmem_inode_operations;
2105 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002106 mpol_shared_policy_init(&info->policy,
2107 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 break;
2109 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002110 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 /* Some things misbehave if size == 0 on a directory */
2112 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2113 inode->i_op = &shmem_dir_inode_operations;
2114 inode->i_fop = &simple_dir_operations;
2115 break;
2116 case S_IFLNK:
2117 /*
2118 * Must not load anything in the rbtree,
2119 * mpol_free_shared_policy will not be called.
2120 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002121 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 break;
2123 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002124 } else
2125 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 return inode;
2127}
2128
Johannes Weiner0cd61442014-04-03 14:47:46 -07002129bool shmem_mapping(struct address_space *mapping)
2130{
Sasha Levinf0774d82015-02-23 05:38:00 -05002131 if (!mapping->host)
2132 return false;
2133
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002134 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002135}
2136
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002138static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002139static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002141#ifdef CONFIG_TMPFS_XATTR
2142static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2143#else
2144#define shmem_initxattrs NULL
2145#endif
2146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002148shmem_write_begin(struct file *file, struct address_space *mapping,
2149 loff_t pos, unsigned len, unsigned flags,
2150 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151{
Nick Piggin800d15a2007-10-16 01:25:03 -07002152 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002153 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002154 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002155
2156 /* i_mutex is held by caller */
2157 if (unlikely(info->seals)) {
2158 if (info->seals & F_SEAL_WRITE)
2159 return -EPERM;
2160 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2161 return -EPERM;
2162 }
2163
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002164 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002165}
2166
2167static int
2168shmem_write_end(struct file *file, struct address_space *mapping,
2169 loff_t pos, unsigned len, unsigned copied,
2170 struct page *page, void *fsdata)
2171{
2172 struct inode *inode = mapping->host;
2173
Hugh Dickinsd3602442008-02-04 22:28:44 -08002174 if (pos + copied > inode->i_size)
2175 i_size_write(inode, pos + copied);
2176
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002177 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002178 struct page *head = compound_head(page);
2179 if (PageTransCompound(page)) {
2180 int i;
2181
2182 for (i = 0; i < HPAGE_PMD_NR; i++) {
2183 if (head + i == page)
2184 continue;
2185 clear_highpage(head + i);
2186 flush_dcache_page(head + i);
2187 }
2188 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002189 if (copied < PAGE_SIZE) {
2190 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002191 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002192 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002193 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002194 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002195 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002196 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002197 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002198 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002199
Nick Piggin800d15a2007-10-16 01:25:03 -07002200 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201}
2202
Al Viro2ba5bbe2014-04-02 20:00:02 -04002203static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204{
Al Viro6e58e792014-02-03 17:07:03 -05002205 struct file *file = iocb->ki_filp;
2206 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002208 pgoff_t index;
2209 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002210 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002211 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002212 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002213 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002214
2215 /*
2216 * Might this read be for a stacking filesystem? Then when reading
2217 * holes of a sparse file, we actually need to allocate those pages,
2218 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2219 */
Al Viro777eda22014-12-17 04:46:46 -05002220 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002221 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002223 index = *ppos >> PAGE_SHIFT;
2224 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226 for (;;) {
2227 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002228 pgoff_t end_index;
2229 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 loff_t i_size = i_size_read(inode);
2231
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002232 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 if (index > end_index)
2234 break;
2235 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002236 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 if (nr <= offset)
2238 break;
2239 }
2240
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002241 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002242 if (error) {
2243 if (error == -EINVAL)
2244 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 break;
2246 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002247 if (page) {
2248 if (sgp == SGP_CACHE)
2249 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002250 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253 /*
2254 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002255 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002257 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002259 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002261 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 if (nr <= offset) {
2263 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002264 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 break;
2266 }
2267 }
2268 nr -= offset;
2269
2270 if (page) {
2271 /*
2272 * If users can be writing to this page using arbitrary
2273 * virtual addresses, take care about potential aliasing
2274 * before reading the page on the kernel side.
2275 */
2276 if (mapping_writably_mapped(mapping))
2277 flush_dcache_page(page);
2278 /*
2279 * Mark the page accessed if we read the beginning.
2280 */
2281 if (!offset)
2282 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002283 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002285 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
2288 /*
2289 * Ok, we have the page, and it's up-to-date, so
2290 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002292 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002293 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002295 index += offset >> PAGE_SHIFT;
2296 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002298 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002299 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 break;
Al Viro6e58e792014-02-03 17:07:03 -05002301 if (ret < nr) {
2302 error = -EFAULT;
2303 break;
2304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 cond_resched();
2306 }
2307
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002308 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002309 file_accessed(file);
2310 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311}
2312
Hugh Dickins708e3502011-07-25 17:12:32 -07002313static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2314 struct pipe_inode_info *pipe, size_t len,
2315 unsigned int flags)
2316{
2317 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002318 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07002319 unsigned int loff, nr_pages, req_pages;
2320 struct page *pages[PIPE_DEF_BUFFERS];
2321 struct partial_page partial[PIPE_DEF_BUFFERS];
2322 struct page *page;
2323 pgoff_t index, end_index;
2324 loff_t isize, left;
2325 int error, page_nr;
2326 struct splice_pipe_desc spd = {
2327 .pages = pages,
2328 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02002329 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07002330 .flags = flags,
2331 .ops = &page_cache_pipe_buf_ops,
2332 .spd_release = spd_release_page,
2333 };
2334
Hugh Dickins71f0e072011-07-25 17:12:33 -07002335 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07002336 if (unlikely(*ppos >= isize))
2337 return 0;
2338
2339 left = isize - *ppos;
2340 if (unlikely(left < len))
2341 len = left;
2342
2343 if (splice_grow_spd(pipe, &spd))
2344 return -ENOMEM;
2345
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002346 index = *ppos >> PAGE_SHIFT;
2347 loff = *ppos & ~PAGE_MASK;
2348 req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04002349 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07002350
Hugh Dickins708e3502011-07-25 17:12:32 -07002351 spd.nr_pages = find_get_pages_contig(mapping, index,
2352 nr_pages, spd.pages);
2353 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07002354 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002355
Hugh Dickins708e3502011-07-25 17:12:32 -07002356 while (spd.nr_pages < nr_pages) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002357 error = shmem_getpage(inode, index, &page, SGP_CACHE);
Hugh Dickins71f0e072011-07-25 17:12:33 -07002358 if (error)
2359 break;
2360 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07002361 spd.pages[spd.nr_pages++] = page;
2362 index++;
2363 }
2364
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002365 index = *ppos >> PAGE_SHIFT;
Hugh Dickins708e3502011-07-25 17:12:32 -07002366 nr_pages = spd.nr_pages;
2367 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002368
Hugh Dickins708e3502011-07-25 17:12:32 -07002369 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
2370 unsigned int this_len;
2371
2372 if (!len)
2373 break;
2374
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002375 this_len = min_t(unsigned long, len, PAGE_SIZE - loff);
Hugh Dickins708e3502011-07-25 17:12:32 -07002376 page = spd.pages[page_nr];
2377
Hugh Dickins71f0e072011-07-25 17:12:33 -07002378 if (!PageUptodate(page) || page->mapping != mapping) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002379 error = shmem_getpage(inode, index, &page, SGP_CACHE);
Hugh Dickins71f0e072011-07-25 17:12:33 -07002380 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07002381 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002382 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002383 put_page(spd.pages[page_nr]);
Hugh Dickins71f0e072011-07-25 17:12:33 -07002384 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07002385 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07002386
2387 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002388 end_index = (isize - 1) >> PAGE_SHIFT;
Hugh Dickins708e3502011-07-25 17:12:32 -07002389 if (unlikely(!isize || index > end_index))
2390 break;
2391
Hugh Dickins708e3502011-07-25 17:12:32 -07002392 if (end_index == index) {
2393 unsigned int plen;
2394
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002395 plen = ((isize - 1) & ~PAGE_MASK) + 1;
Hugh Dickins708e3502011-07-25 17:12:32 -07002396 if (plen <= loff)
2397 break;
2398
Hugh Dickins708e3502011-07-25 17:12:32 -07002399 this_len = min(this_len, plen - loff);
2400 len = this_len;
2401 }
2402
2403 spd.partial[page_nr].offset = loff;
2404 spd.partial[page_nr].len = this_len;
2405 len -= this_len;
2406 loff = 0;
2407 spd.nr_pages++;
2408 index++;
2409 }
2410
Hugh Dickins708e3502011-07-25 17:12:32 -07002411 while (page_nr < nr_pages)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002412 put_page(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07002413
2414 if (spd.nr_pages)
2415 error = splice_to_pipe(pipe, &spd);
2416
Eric Dumazet047fe362012-06-12 15:24:40 +02002417 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07002418
2419 if (error > 0) {
2420 *ppos += error;
2421 file_accessed(in);
2422 }
2423 return error;
2424}
2425
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002426/*
2427 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2428 */
2429static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002430 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002431{
2432 struct page *page;
2433 struct pagevec pvec;
2434 pgoff_t indices[PAGEVEC_SIZE];
2435 bool done = false;
2436 int i;
2437
2438 pagevec_init(&pvec, 0);
2439 pvec.nr = 1; /* start small: we may be there already */
2440 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002441 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002442 pvec.nr, pvec.pages, indices);
2443 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002444 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002445 index = end;
2446 break;
2447 }
2448 for (i = 0; i < pvec.nr; i++, index++) {
2449 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002450 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002451 done = true;
2452 break;
2453 }
2454 index = indices[i];
2455 }
2456 page = pvec.pages[i];
2457 if (page && !radix_tree_exceptional_entry(page)) {
2458 if (!PageUptodate(page))
2459 page = NULL;
2460 }
2461 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002462 (page && whence == SEEK_DATA) ||
2463 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002464 done = true;
2465 break;
2466 }
2467 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002468 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002469 pagevec_release(&pvec);
2470 pvec.nr = PAGEVEC_SIZE;
2471 cond_resched();
2472 }
2473 return index;
2474}
2475
Andrew Morton965c8e52012-12-17 15:59:39 -08002476static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002477{
2478 struct address_space *mapping = file->f_mapping;
2479 struct inode *inode = mapping->host;
2480 pgoff_t start, end;
2481 loff_t new_offset;
2482
Andrew Morton965c8e52012-12-17 15:59:39 -08002483 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2484 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002485 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002486 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002487 /* We're holding i_mutex so we can access i_size directly */
2488
2489 if (offset < 0)
2490 offset = -EINVAL;
2491 else if (offset >= inode->i_size)
2492 offset = -ENXIO;
2493 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002494 start = offset >> PAGE_SHIFT;
2495 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002496 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002497 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002498 if (new_offset > offset) {
2499 if (new_offset < inode->i_size)
2500 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002501 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002502 offset = -ENXIO;
2503 else
2504 offset = inode->i_size;
2505 }
2506 }
2507
Hugh Dickins387aae62013-08-04 11:30:25 -07002508 if (offset >= 0)
2509 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002510 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002511 return offset;
2512}
2513
David Herrmann05f65b52014-08-08 14:25:36 -07002514/*
2515 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2516 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2517 */
2518#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2519#define LAST_SCAN 4 /* about 150ms max */
2520
2521static void shmem_tag_pins(struct address_space *mapping)
2522{
2523 struct radix_tree_iter iter;
2524 void **slot;
2525 pgoff_t start;
2526 struct page *page;
2527
2528 lru_add_drain();
2529 start = 0;
2530 rcu_read_lock();
2531
David Herrmann05f65b52014-08-08 14:25:36 -07002532 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2533 page = radix_tree_deref_slot(slot);
2534 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002535 if (radix_tree_deref_retry(page)) {
2536 slot = radix_tree_iter_retry(&iter);
2537 continue;
2538 }
David Herrmann05f65b52014-08-08 14:25:36 -07002539 } else if (page_count(page) - page_mapcount(page) > 1) {
2540 spin_lock_irq(&mapping->tree_lock);
2541 radix_tree_tag_set(&mapping->page_tree, iter.index,
2542 SHMEM_TAG_PINNED);
2543 spin_unlock_irq(&mapping->tree_lock);
2544 }
2545
2546 if (need_resched()) {
2547 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002548 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002549 }
2550 }
2551 rcu_read_unlock();
2552}
2553
2554/*
2555 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2556 * via get_user_pages(), drivers might have some pending I/O without any active
2557 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2558 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2559 * them to be dropped.
2560 * The caller must guarantee that no new user will acquire writable references
2561 * to those pages to avoid races.
2562 */
David Herrmann40e041a2014-08-08 14:25:27 -07002563static int shmem_wait_for_pins(struct address_space *mapping)
2564{
David Herrmann05f65b52014-08-08 14:25:36 -07002565 struct radix_tree_iter iter;
2566 void **slot;
2567 pgoff_t start;
2568 struct page *page;
2569 int error, scan;
2570
2571 shmem_tag_pins(mapping);
2572
2573 error = 0;
2574 for (scan = 0; scan <= LAST_SCAN; scan++) {
2575 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2576 break;
2577
2578 if (!scan)
2579 lru_add_drain_all();
2580 else if (schedule_timeout_killable((HZ << scan) / 200))
2581 scan = LAST_SCAN;
2582
2583 start = 0;
2584 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002585 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2586 start, SHMEM_TAG_PINNED) {
2587
2588 page = radix_tree_deref_slot(slot);
2589 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002590 if (radix_tree_deref_retry(page)) {
2591 slot = radix_tree_iter_retry(&iter);
2592 continue;
2593 }
David Herrmann05f65b52014-08-08 14:25:36 -07002594
2595 page = NULL;
2596 }
2597
2598 if (page &&
2599 page_count(page) - page_mapcount(page) != 1) {
2600 if (scan < LAST_SCAN)
2601 goto continue_resched;
2602
2603 /*
2604 * On the last scan, we clean up all those tags
2605 * we inserted; but make a note that we still
2606 * found pages pinned.
2607 */
2608 error = -EBUSY;
2609 }
2610
2611 spin_lock_irq(&mapping->tree_lock);
2612 radix_tree_tag_clear(&mapping->page_tree,
2613 iter.index, SHMEM_TAG_PINNED);
2614 spin_unlock_irq(&mapping->tree_lock);
2615continue_resched:
2616 if (need_resched()) {
2617 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002618 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002619 }
2620 }
2621 rcu_read_unlock();
2622 }
2623
2624 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002625}
2626
2627#define F_ALL_SEALS (F_SEAL_SEAL | \
2628 F_SEAL_SHRINK | \
2629 F_SEAL_GROW | \
2630 F_SEAL_WRITE)
2631
2632int shmem_add_seals(struct file *file, unsigned int seals)
2633{
2634 struct inode *inode = file_inode(file);
2635 struct shmem_inode_info *info = SHMEM_I(inode);
2636 int error;
2637
2638 /*
2639 * SEALING
2640 * Sealing allows multiple parties to share a shmem-file but restrict
2641 * access to a specific subset of file operations. Seals can only be
2642 * added, but never removed. This way, mutually untrusted parties can
2643 * share common memory regions with a well-defined policy. A malicious
2644 * peer can thus never perform unwanted operations on a shared object.
2645 *
2646 * Seals are only supported on special shmem-files and always affect
2647 * the whole underlying inode. Once a seal is set, it may prevent some
2648 * kinds of access to the file. Currently, the following seals are
2649 * defined:
2650 * SEAL_SEAL: Prevent further seals from being set on this file
2651 * SEAL_SHRINK: Prevent the file from shrinking
2652 * SEAL_GROW: Prevent the file from growing
2653 * SEAL_WRITE: Prevent write access to the file
2654 *
2655 * As we don't require any trust relationship between two parties, we
2656 * must prevent seals from being removed. Therefore, sealing a file
2657 * only adds a given set of seals to the file, it never touches
2658 * existing seals. Furthermore, the "setting seals"-operation can be
2659 * sealed itself, which basically prevents any further seal from being
2660 * added.
2661 *
2662 * Semantics of sealing are only defined on volatile files. Only
2663 * anonymous shmem files support sealing. More importantly, seals are
2664 * never written to disk. Therefore, there's no plan to support it on
2665 * other file types.
2666 */
2667
2668 if (file->f_op != &shmem_file_operations)
2669 return -EINVAL;
2670 if (!(file->f_mode & FMODE_WRITE))
2671 return -EPERM;
2672 if (seals & ~(unsigned int)F_ALL_SEALS)
2673 return -EINVAL;
2674
Al Viro59551022016-01-22 15:40:57 -05002675 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002676
2677 if (info->seals & F_SEAL_SEAL) {
2678 error = -EPERM;
2679 goto unlock;
2680 }
2681
2682 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2683 error = mapping_deny_writable(file->f_mapping);
2684 if (error)
2685 goto unlock;
2686
2687 error = shmem_wait_for_pins(file->f_mapping);
2688 if (error) {
2689 mapping_allow_writable(file->f_mapping);
2690 goto unlock;
2691 }
2692 }
2693
2694 info->seals |= seals;
2695 error = 0;
2696
2697unlock:
Al Viro59551022016-01-22 15:40:57 -05002698 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002699 return error;
2700}
2701EXPORT_SYMBOL_GPL(shmem_add_seals);
2702
2703int shmem_get_seals(struct file *file)
2704{
2705 if (file->f_op != &shmem_file_operations)
2706 return -EINVAL;
2707
2708 return SHMEM_I(file_inode(file))->seals;
2709}
2710EXPORT_SYMBOL_GPL(shmem_get_seals);
2711
2712long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2713{
2714 long error;
2715
2716 switch (cmd) {
2717 case F_ADD_SEALS:
2718 /* disallow upper 32bit */
2719 if (arg > UINT_MAX)
2720 return -EINVAL;
2721
2722 error = shmem_add_seals(file, arg);
2723 break;
2724 case F_GET_SEALS:
2725 error = shmem_get_seals(file);
2726 break;
2727 default:
2728 error = -EINVAL;
2729 break;
2730 }
2731
2732 return error;
2733}
2734
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002735static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2736 loff_t len)
2737{
Al Viro496ad9a2013-01-23 17:07:38 -05002738 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002739 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002740 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002741 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002742 pgoff_t start, index, end;
2743 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002744
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002745 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2746 return -EOPNOTSUPP;
2747
Al Viro59551022016-01-22 15:40:57 -05002748 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002749
2750 if (mode & FALLOC_FL_PUNCH_HOLE) {
2751 struct address_space *mapping = file->f_mapping;
2752 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2753 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002754 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002755
David Herrmann40e041a2014-08-08 14:25:27 -07002756 /* protected by i_mutex */
2757 if (info->seals & F_SEAL_WRITE) {
2758 error = -EPERM;
2759 goto out;
2760 }
2761
Hugh Dickins8e205f72014-07-23 14:00:10 -07002762 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002763 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2764 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2765 spin_lock(&inode->i_lock);
2766 inode->i_private = &shmem_falloc;
2767 spin_unlock(&inode->i_lock);
2768
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002769 if ((u64)unmap_end > (u64)unmap_start)
2770 unmap_mapping_range(mapping, unmap_start,
2771 1 + unmap_end - unmap_start, 0);
2772 shmem_truncate_range(inode, offset, offset + len - 1);
2773 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002774
2775 spin_lock(&inode->i_lock);
2776 inode->i_private = NULL;
2777 wake_up_all(&shmem_falloc_waitq);
2778 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002779 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002780 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002781 }
2782
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002783 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2784 error = inode_newsize_ok(inode, offset + len);
2785 if (error)
2786 goto out;
2787
David Herrmann40e041a2014-08-08 14:25:27 -07002788 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2789 error = -EPERM;
2790 goto out;
2791 }
2792
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002793 start = offset >> PAGE_SHIFT;
2794 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002795 /* Try to avoid a swapstorm if len is impossible to satisfy */
2796 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2797 error = -ENOSPC;
2798 goto out;
2799 }
2800
Hugh Dickins8e205f72014-07-23 14:00:10 -07002801 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002802 shmem_falloc.start = start;
2803 shmem_falloc.next = start;
2804 shmem_falloc.nr_falloced = 0;
2805 shmem_falloc.nr_unswapped = 0;
2806 spin_lock(&inode->i_lock);
2807 inode->i_private = &shmem_falloc;
2808 spin_unlock(&inode->i_lock);
2809
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002810 for (index = start; index < end; index++) {
2811 struct page *page;
2812
2813 /*
2814 * Good, the fallocate(2) manpage permits EINTR: we may have
2815 * been interrupted because we are using up too much memory.
2816 */
2817 if (signal_pending(current))
2818 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002819 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2820 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002821 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002822 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002823 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002824 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002825 if (index > start) {
2826 shmem_undo_range(inode,
2827 (loff_t)start << PAGE_SHIFT,
2828 ((loff_t)index << PAGE_SHIFT) - 1, true);
2829 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002830 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002831 }
2832
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002833 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002834 * Inform shmem_writepage() how far we have reached.
2835 * No need for lock or barrier: we have the page lock.
2836 */
2837 shmem_falloc.next++;
2838 if (!PageUptodate(page))
2839 shmem_falloc.nr_falloced++;
2840
2841 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002842 * If !PageUptodate, leave it that way so that freeable pages
2843 * can be recognized if we need to rollback on error later.
2844 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002845 * than free the pages we are allocating (and SGP_CACHE pages
2846 * might still be clean: we now need to mark those dirty too).
2847 */
2848 set_page_dirty(page);
2849 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002850 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002851 cond_resched();
2852 }
2853
2854 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2855 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002856 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002857undone:
2858 spin_lock(&inode->i_lock);
2859 inode->i_private = NULL;
2860 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002861out:
Al Viro59551022016-01-22 15:40:57 -05002862 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002863 return error;
2864}
2865
David Howells726c3342006-06-23 02:02:58 -07002866static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867{
David Howells726c3342006-06-23 02:02:58 -07002868 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869
2870 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002871 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002873 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002875 buf->f_bavail =
2876 buf->f_bfree = sbinfo->max_blocks -
2877 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002878 }
2879 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 buf->f_files = sbinfo->max_inodes;
2881 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 }
2883 /* else leave those fields 0 like simple_statfs */
2884 return 0;
2885}
2886
2887/*
2888 * File creation. Allocate an inode, and we're done..
2889 */
2890static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002891shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002893 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 int error = -ENOSPC;
2895
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002896 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002898 error = simple_acl_create(dir, inode);
2899 if (error)
2900 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002901 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002902 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002903 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002904 if (error && error != -EOPNOTSUPP)
2905 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002906
Al Viro718deb62009-12-16 19:35:36 -05002907 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 dir->i_size += BOGO_DIRENT_SIZE;
2909 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2910 d_instantiate(dentry, inode);
2911 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 }
2913 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002914out_iput:
2915 iput(inode);
2916 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917}
2918
Al Viro60545d02013-06-07 01:20:27 -04002919static int
2920shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2921{
2922 struct inode *inode;
2923 int error = -ENOSPC;
2924
2925 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2926 if (inode) {
2927 error = security_inode_init_security(inode, dir,
2928 NULL,
2929 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002930 if (error && error != -EOPNOTSUPP)
2931 goto out_iput;
2932 error = simple_acl_create(dir, inode);
2933 if (error)
2934 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002935 d_tmpfile(dentry, inode);
2936 }
2937 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002938out_iput:
2939 iput(inode);
2940 return error;
Al Viro60545d02013-06-07 01:20:27 -04002941}
2942
Al Viro18bb1db2011-07-26 01:41:39 -04002943static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944{
2945 int error;
2946
2947 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2948 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002949 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 return 0;
2951}
2952
Al Viro4acdaf22011-07-26 01:42:34 -04002953static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002954 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955{
2956 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2957}
2958
2959/*
2960 * Link a file..
2961 */
2962static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2963{
David Howells75c3cfa2015-03-17 22:26:12 +00002964 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002965 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966
2967 /*
2968 * No ordinary (disk based) filesystem counts links as inodes;
2969 * but each new link needs a new dentry, pinning lowmem, and
2970 * tmpfs dentries cannot be pruned until they are unlinked.
2971 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002972 ret = shmem_reserve_inode(inode->i_sb);
2973 if (ret)
2974 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
2976 dir->i_size += BOGO_DIRENT_SIZE;
2977 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002978 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002979 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 dget(dentry); /* Extra pinning count for the created dentry */
2981 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002982out:
2983 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984}
2985
2986static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2987{
David Howells75c3cfa2015-03-17 22:26:12 +00002988 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002990 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2991 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
2993 dir->i_size -= BOGO_DIRENT_SIZE;
2994 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002995 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 dput(dentry); /* Undo the count from "create" - this does all the work */
2997 return 0;
2998}
2999
3000static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3001{
3002 if (!simple_empty(dentry))
3003 return -ENOTEMPTY;
3004
David Howells75c3cfa2015-03-17 22:26:12 +00003005 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003006 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 return shmem_unlink(dir, dentry);
3008}
3009
Miklos Szeredi37456772014-07-23 15:15:34 +02003010static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3011{
David Howellse36cb0b2015-01-29 12:02:35 +00003012 bool old_is_dir = d_is_dir(old_dentry);
3013 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003014
3015 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3016 if (old_is_dir) {
3017 drop_nlink(old_dir);
3018 inc_nlink(new_dir);
3019 } else {
3020 drop_nlink(new_dir);
3021 inc_nlink(old_dir);
3022 }
3023 }
3024 old_dir->i_ctime = old_dir->i_mtime =
3025 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003026 d_inode(old_dentry)->i_ctime =
3027 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
Miklos Szeredi37456772014-07-23 15:15:34 +02003028
3029 return 0;
3030}
3031
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003032static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3033{
3034 struct dentry *whiteout;
3035 int error;
3036
3037 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3038 if (!whiteout)
3039 return -ENOMEM;
3040
3041 error = shmem_mknod(old_dir, whiteout,
3042 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3043 dput(whiteout);
3044 if (error)
3045 return error;
3046
3047 /*
3048 * Cheat and hash the whiteout while the old dentry is still in
3049 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3050 *
3051 * d_lookup() will consistently find one of them at this point,
3052 * not sure which one, but that isn't even important.
3053 */
3054 d_rehash(whiteout);
3055 return 0;
3056}
3057
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058/*
3059 * The VFS layer already does all the dentry stuff for rename,
3060 * we just have to decrement the usage count for the target if
3061 * it exists so that the VFS layer correctly free's it when it
3062 * gets overwritten.
3063 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003064static 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 -07003065{
David Howells75c3cfa2015-03-17 22:26:12 +00003066 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 int they_are_dirs = S_ISDIR(inode->i_mode);
3068
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003069 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003070 return -EINVAL;
3071
Miklos Szeredi37456772014-07-23 15:15:34 +02003072 if (flags & RENAME_EXCHANGE)
3073 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3074
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 if (!simple_empty(new_dentry))
3076 return -ENOTEMPTY;
3077
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003078 if (flags & RENAME_WHITEOUT) {
3079 int error;
3080
3081 error = shmem_whiteout(old_dir, old_dentry);
3082 if (error)
3083 return error;
3084 }
3085
David Howells75c3cfa2015-03-17 22:26:12 +00003086 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003088 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003089 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003090 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003091 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003093 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003094 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 }
3096
3097 old_dir->i_size -= BOGO_DIRENT_SIZE;
3098 new_dir->i_size += BOGO_DIRENT_SIZE;
3099 old_dir->i_ctime = old_dir->i_mtime =
3100 new_dir->i_ctime = new_dir->i_mtime =
3101 inode->i_ctime = CURRENT_TIME;
3102 return 0;
3103}
3104
3105static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3106{
3107 int error;
3108 int len;
3109 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003110 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 struct shmem_inode_info *info;
3112
3113 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003114 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 return -ENAMETOOLONG;
3116
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003117 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 if (!inode)
3119 return -ENOSPC;
3120
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003121 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003122 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003123 if (error) {
3124 if (error != -EOPNOTSUPP) {
3125 iput(inode);
3126 return error;
3127 }
3128 error = 0;
3129 }
3130
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 info = SHMEM_I(inode);
3132 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003133 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003134 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3135 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003136 iput(inode);
3137 return -ENOMEM;
3138 }
3139 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003141 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003142 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 if (error) {
3144 iput(inode);
3145 return error;
3146 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003147 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003149 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003150 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003152 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003153 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 dir->i_size += BOGO_DIRENT_SIZE;
3156 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
3157 d_instantiate(dentry, inode);
3158 dget(dentry);
3159 return 0;
3160}
3161
Al Virofceef392015-12-29 15:58:39 -05003162static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163{
Al Virofceef392015-12-29 15:58:39 -05003164 mark_page_accessed(arg);
3165 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166}
3167
Al Viro6b255392015-11-17 10:20:54 -05003168static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003169 struct inode *inode,
3170 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003173 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003174 if (!dentry) {
3175 page = find_get_page(inode->i_mapping, 0);
3176 if (!page)
3177 return ERR_PTR(-ECHILD);
3178 if (!PageUptodate(page)) {
3179 put_page(page);
3180 return ERR_PTR(-ECHILD);
3181 }
3182 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003183 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003184 if (error)
3185 return ERR_PTR(error);
3186 unlock_page(page);
3187 }
Al Virofceef392015-12-29 15:58:39 -05003188 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003189 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190}
3191
Eric Parisb09e0fa2011-05-24 17:12:39 -07003192#ifdef CONFIG_TMPFS_XATTR
3193/*
3194 * Superblocks without xattr inode operations may get some security.* xattr
3195 * support from the LSM "for free". As soon as we have any other xattrs
3196 * like ACLs, we also need to implement the security.* handlers at
3197 * filesystem level, though.
3198 */
3199
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003200/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003201 * Callback for security_inode_init_security() for acquiring xattrs.
3202 */
3203static int shmem_initxattrs(struct inode *inode,
3204 const struct xattr *xattr_array,
3205 void *fs_info)
3206{
3207 struct shmem_inode_info *info = SHMEM_I(inode);
3208 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003209 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003210 size_t len;
3211
3212 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003213 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003214 if (!new_xattr)
3215 return -ENOMEM;
3216
3217 len = strlen(xattr->name) + 1;
3218 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3219 GFP_KERNEL);
3220 if (!new_xattr->name) {
3221 kfree(new_xattr);
3222 return -ENOMEM;
3223 }
3224
3225 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3226 XATTR_SECURITY_PREFIX_LEN);
3227 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3228 xattr->name, len);
3229
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003230 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003231 }
3232
3233 return 0;
3234}
3235
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003236static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003237 struct dentry *unused, struct inode *inode,
3238 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003239{
Al Virob2968212016-04-10 20:48:24 -04003240 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003241
3242 name = xattr_full_name(handler, name);
3243 return simple_xattr_get(&info->xattrs, name, buffer, size);
3244}
3245
3246static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003247 struct dentry *unused, struct inode *inode,
3248 const char *name, const void *value,
3249 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003250{
Al Viro59301222016-05-27 10:19:30 -04003251 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003252
3253 name = xattr_full_name(handler, name);
3254 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3255}
3256
3257static const struct xattr_handler shmem_security_xattr_handler = {
3258 .prefix = XATTR_SECURITY_PREFIX,
3259 .get = shmem_xattr_handler_get,
3260 .set = shmem_xattr_handler_set,
3261};
3262
3263static const struct xattr_handler shmem_trusted_xattr_handler = {
3264 .prefix = XATTR_TRUSTED_PREFIX,
3265 .get = shmem_xattr_handler_get,
3266 .set = shmem_xattr_handler_set,
3267};
3268
Eric Parisb09e0fa2011-05-24 17:12:39 -07003269static const struct xattr_handler *shmem_xattr_handlers[] = {
3270#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003271 &posix_acl_access_xattr_handler,
3272 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003273#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003274 &shmem_security_xattr_handler,
3275 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003276 NULL
3277};
3278
Eric Parisb09e0fa2011-05-24 17:12:39 -07003279static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3280{
David Howells75c3cfa2015-03-17 22:26:12 +00003281 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003282 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003283}
3284#endif /* CONFIG_TMPFS_XATTR */
3285
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003286static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003288 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003289#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003290 .setxattr = generic_setxattr,
3291 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003292 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003293 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003294#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295};
3296
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003297static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003299 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003300#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003301 .setxattr = generic_setxattr,
3302 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003303 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003304 .removexattr = generic_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003305#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003306};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003307
David M. Grimes91828a42006-10-17 00:09:45 -07003308static struct dentry *shmem_get_parent(struct dentry *child)
3309{
3310 return ERR_PTR(-ESTALE);
3311}
3312
3313static int shmem_match(struct inode *ino, void *vfh)
3314{
3315 __u32 *fh = vfh;
3316 __u64 inum = fh[2];
3317 inum = (inum << 32) | fh[1];
3318 return ino->i_ino == inum && fh[0] == ino->i_generation;
3319}
3320
Christoph Hellwig480b1162007-10-21 16:42:13 -07003321static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3322 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003323{
David M. Grimes91828a42006-10-17 00:09:45 -07003324 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003325 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003326 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003327
Christoph Hellwig480b1162007-10-21 16:42:13 -07003328 if (fh_len < 3)
3329 return NULL;
3330
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003331 inum = fid->raw[2];
3332 inum = (inum << 32) | fid->raw[1];
3333
Christoph Hellwig480b1162007-10-21 16:42:13 -07003334 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3335 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003336 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003337 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003338 iput(inode);
3339 }
3340
Christoph Hellwig480b1162007-10-21 16:42:13 -07003341 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003342}
3343
Al Virob0b03822012-04-02 14:34:06 -04003344static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3345 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003346{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303347 if (*len < 3) {
3348 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003349 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303350 }
David M. Grimes91828a42006-10-17 00:09:45 -07003351
Al Viro1d3382cb2010-10-23 15:19:20 -04003352 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003353 /* Unfortunately insert_inode_hash is not idempotent,
3354 * so as we hash inodes here rather than at creation
3355 * time, we need a lock to ensure we only try
3356 * to do it once
3357 */
3358 static DEFINE_SPINLOCK(lock);
3359 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003360 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003361 __insert_inode_hash(inode,
3362 inode->i_ino + inode->i_generation);
3363 spin_unlock(&lock);
3364 }
3365
3366 fh[0] = inode->i_generation;
3367 fh[1] = inode->i_ino;
3368 fh[2] = ((__u64)inode->i_ino) >> 32;
3369
3370 *len = 3;
3371 return 1;
3372}
3373
Christoph Hellwig39655162007-10-21 16:42:17 -07003374static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003375 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003376 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003377 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003378};
3379
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003380static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3381 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382{
3383 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003384 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003385 uid_t uid;
3386 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003388 while (options != NULL) {
3389 this_char = options;
3390 for (;;) {
3391 /*
3392 * NUL-terminate this option: unfortunately,
3393 * mount options form a comma-separated list,
3394 * but mpol's nodelist may also contain commas.
3395 */
3396 options = strchr(options, ',');
3397 if (options == NULL)
3398 break;
3399 options++;
3400 if (!isdigit(*options)) {
3401 options[-1] = '\0';
3402 break;
3403 }
3404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 if (!*this_char)
3406 continue;
3407 if ((value = strchr(this_char,'=')) != NULL) {
3408 *value++ = 0;
3409 } else {
Joe Perches11705322016-03-17 14:19:50 -07003410 pr_err("tmpfs: No value for mount option '%s'\n",
3411 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003412 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413 }
3414
3415 if (!strcmp(this_char,"size")) {
3416 unsigned long long size;
3417 size = memparse(value,&rest);
3418 if (*rest == '%') {
3419 size <<= PAGE_SHIFT;
3420 size *= totalram_pages;
3421 do_div(size, 100);
3422 rest++;
3423 }
3424 if (*rest)
3425 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003426 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003427 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003429 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 if (*rest)
3431 goto bad_val;
3432 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003433 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 if (*rest)
3435 goto bad_val;
3436 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003437 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003439 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 if (*rest)
3441 goto bad_val;
3442 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003443 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003445 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 if (*rest)
3447 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003448 sbinfo->uid = make_kuid(current_user_ns(), uid);
3449 if (!uid_valid(sbinfo->uid))
3450 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003452 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003454 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455 if (*rest)
3456 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003457 sbinfo->gid = make_kgid(current_user_ns(), gid);
3458 if (!gid_valid(sbinfo->gid))
3459 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003460#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003461 } else if (!strcmp(this_char, "huge")) {
3462 int huge;
3463 huge = shmem_parse_huge(value);
3464 if (huge < 0)
3465 goto bad_val;
3466 if (!has_transparent_hugepage() &&
3467 huge != SHMEM_HUGE_NEVER)
3468 goto bad_val;
3469 sbinfo->huge = huge;
3470#endif
3471#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003472 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003473 mpol_put(mpol);
3474 mpol = NULL;
3475 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003476 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003477#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 } else {
Joe Perches11705322016-03-17 14:19:50 -07003479 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003480 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 }
3482 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003483 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 return 0;
3485
3486bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003487 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003489error:
3490 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 return 1;
3492
3493}
3494
3495static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3496{
3497 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003498 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003499 unsigned long inodes;
3500 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501
Greg Thelen5f001102013-02-22 16:36:01 -08003502 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003503 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003504 return error;
3505
3506 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003507 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003508 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003509 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003510 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003511 goto out;
3512 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003513 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003514 * but we must separately disallow unlimited->limited, because
3515 * in that case we have no record of how much is already in use.
3516 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003517 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003518 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003519 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003520 goto out;
3521
3522 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003523 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003524 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003525 sbinfo->max_inodes = config.max_inodes;
3526 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003527
Greg Thelen5f001102013-02-22 16:36:01 -08003528 /*
3529 * Preserve previous mempolicy unless mpol remount option was specified.
3530 */
3531 if (config.mpol) {
3532 mpol_put(sbinfo->mpol);
3533 sbinfo->mpol = config.mpol; /* transfers initial ref */
3534 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003535out:
3536 spin_unlock(&sbinfo->stat_lock);
3537 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003539
Al Viro34c80b12011-12-08 21:32:45 -05003540static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003541{
Al Viro34c80b12011-12-08 21:32:45 -05003542 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003543
3544 if (sbinfo->max_blocks != shmem_default_max_blocks())
3545 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003546 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003547 if (sbinfo->max_inodes != shmem_default_max_inodes())
3548 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3549 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003550 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003551 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3552 seq_printf(seq, ",uid=%u",
3553 from_kuid_munged(&init_user_ns, sbinfo->uid));
3554 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3555 seq_printf(seq, ",gid=%u",
3556 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003557#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003558 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3559 if (sbinfo->huge)
3560 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3561#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003562 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003563 return 0;
3564}
David Herrmann9183df22014-08-08 14:25:29 -07003565
3566#define MFD_NAME_PREFIX "memfd:"
3567#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3568#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3569
3570#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3571
3572SYSCALL_DEFINE2(memfd_create,
3573 const char __user *, uname,
3574 unsigned int, flags)
3575{
3576 struct shmem_inode_info *info;
3577 struct file *file;
3578 int fd, error;
3579 char *name;
3580 long len;
3581
3582 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3583 return -EINVAL;
3584
3585 /* length includes terminating zero */
3586 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3587 if (len <= 0)
3588 return -EFAULT;
3589 if (len > MFD_NAME_MAX_LEN + 1)
3590 return -EINVAL;
3591
3592 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3593 if (!name)
3594 return -ENOMEM;
3595
3596 strcpy(name, MFD_NAME_PREFIX);
3597 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3598 error = -EFAULT;
3599 goto err_name;
3600 }
3601
3602 /* terminating-zero may have changed after strnlen_user() returned */
3603 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3604 error = -EFAULT;
3605 goto err_name;
3606 }
3607
3608 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3609 if (fd < 0) {
3610 error = fd;
3611 goto err_name;
3612 }
3613
3614 file = shmem_file_setup(name, 0, VM_NORESERVE);
3615 if (IS_ERR(file)) {
3616 error = PTR_ERR(file);
3617 goto err_fd;
3618 }
3619 info = SHMEM_I(file_inode(file));
3620 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3621 file->f_flags |= O_RDWR | O_LARGEFILE;
3622 if (flags & MFD_ALLOW_SEALING)
3623 info->seals &= ~F_SEAL_SEAL;
3624
3625 fd_install(fd, file);
3626 kfree(name);
3627 return fd;
3628
3629err_fd:
3630 put_unused_fd(fd);
3631err_name:
3632 kfree(name);
3633 return error;
3634}
3635
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003636#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637
3638static void shmem_put_super(struct super_block *sb)
3639{
Hugh Dickins602586a2010-08-17 15:23:56 -07003640 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3641
3642 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003643 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003644 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645 sb->s_fs_info = NULL;
3646}
3647
Kay Sievers2b2af542009-04-30 15:23:42 +02003648int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649{
3650 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003651 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003652 int err = -ENOMEM;
3653
3654 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003655 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003656 L1_CACHE_BYTES), GFP_KERNEL);
3657 if (!sbinfo)
3658 return -ENOMEM;
3659
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003660 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003661 sbinfo->uid = current_fsuid();
3662 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003663 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003665#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 /*
3667 * Per default we only allow half of the physical ram per
3668 * tmpfs instance, limiting inodes to one per page of lowmem;
3669 * but the internal instance is left unlimited.
3670 */
Al Viroca4e0512013-08-31 12:57:10 -04003671 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003672 sbinfo->max_blocks = shmem_default_max_blocks();
3673 sbinfo->max_inodes = shmem_default_max_inodes();
3674 if (shmem_parse_options(data, sbinfo, false)) {
3675 err = -EINVAL;
3676 goto failed;
3677 }
Al Viroca4e0512013-08-31 12:57:10 -04003678 } else {
3679 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 }
David M. Grimes91828a42006-10-17 00:09:45 -07003681 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003682 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683#else
3684 sb->s_flags |= MS_NOUSER;
3685#endif
3686
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003687 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003688 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003689 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003690 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003691 spin_lock_init(&sbinfo->shrinklist_lock);
3692 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003693
Hugh Dickins285b2c42011-08-03 16:21:20 -07003694 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003695 sb->s_blocksize = PAGE_SIZE;
3696 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 sb->s_magic = TMPFS_MAGIC;
3698 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003699 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003700#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003701 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003702#endif
3703#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003704 sb->s_flags |= MS_POSIXACL;
3705#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003706
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003707 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 if (!inode)
3709 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003710 inode->i_uid = sbinfo->uid;
3711 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003712 sb->s_root = d_make_root(inode);
3713 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003714 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715 return 0;
3716
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717failed:
3718 shmem_put_super(sb);
3719 return err;
3720}
3721
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003722static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003723
3724static struct inode *shmem_alloc_inode(struct super_block *sb)
3725{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003726 struct shmem_inode_info *info;
3727 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3728 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003730 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731}
3732
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003733static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003734{
3735 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003736 if (S_ISLNK(inode->i_mode))
3737 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003738 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3739}
3740
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741static void shmem_destroy_inode(struct inode *inode)
3742{
Al Viro09208d12011-07-26 03:15:03 -04003743 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003745 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746}
3747
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003748static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003750 struct shmem_inode_info *info = foo;
3751 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752}
3753
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003754static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755{
3756 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3757 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003758 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 return 0;
3760}
3761
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003762static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003764 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765}
3766
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003767static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003769 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003771 .write_begin = shmem_write_begin,
3772 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003774#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003775 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003776#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003777 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778};
3779
Helge Deller15ad7cd2006-12-06 20:40:36 -08003780static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003782 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003784 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003785 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003786 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003787 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003788 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003789 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003790 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791#endif
3792};
3793
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003794static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003795 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003796 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003797#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003798 .setxattr = generic_setxattr,
3799 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003800 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003801 .removexattr = generic_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003802 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003803#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804};
3805
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003806static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807#ifdef CONFIG_TMPFS
3808 .create = shmem_create,
3809 .lookup = simple_lookup,
3810 .link = shmem_link,
3811 .unlink = shmem_unlink,
3812 .symlink = shmem_symlink,
3813 .mkdir = shmem_mkdir,
3814 .rmdir = shmem_rmdir,
3815 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003816 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003817 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003819#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003820 .setxattr = generic_setxattr,
3821 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003822 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003823 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003824#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003825#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003826 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003827 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003828#endif
3829};
3830
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003831static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003832#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003833 .setxattr = generic_setxattr,
3834 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003835 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003836 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003837#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003838#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003839 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003840 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003841#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842};
3843
Hugh Dickins759b9772007-03-05 00:30:28 -08003844static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845 .alloc_inode = shmem_alloc_inode,
3846 .destroy_inode = shmem_destroy_inode,
3847#ifdef CONFIG_TMPFS
3848 .statfs = shmem_statfs,
3849 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003850 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851#endif
Al Viro1f895f72010-06-05 19:10:41 -04003852 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 .drop_inode = generic_delete_inode,
3854 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003855#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3856 .nr_cached_objects = shmem_unused_huge_count,
3857 .free_cached_objects = shmem_unused_huge_scan,
3858#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859};
3860
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003861static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003862 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003863 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864#ifdef CONFIG_NUMA
3865 .set_policy = shmem_set_policy,
3866 .get_policy = shmem_get_policy,
3867#endif
3868};
3869
Al Viro3c26ff62010-07-25 11:46:36 +04003870static struct dentry *shmem_mount(struct file_system_type *fs_type,
3871 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872{
Al Viro3c26ff62010-07-25 11:46:36 +04003873 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874}
3875
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003876static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877 .owner = THIS_MODULE,
3878 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003879 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003881 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003883
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003884int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885{
3886 int error;
3887
Rob Landley16203a72013-09-11 14:26:12 -07003888 /* If rootfs called this, don't re-init */
3889 if (shmem_inode_cachep)
3890 return 0;
3891
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003892 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 if (error)
3894 goto out3;
3895
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003896 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003898 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 goto out2;
3900 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003901
Al Viroca4e0512013-08-31 12:57:10 -04003902 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 if (IS_ERR(shm_mnt)) {
3904 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003905 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906 goto out1;
3907 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003908
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003909#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003910 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3911 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3912 else
3913 shmem_huge = 0; /* just in case it was patched */
3914#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915 return 0;
3916
3917out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003918 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003919out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003920 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921out3:
3922 shm_mnt = ERR_PTR(error);
3923 return error;
3924}
Matt Mackall853ac432009-01-06 14:40:20 -08003925
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003926#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003927static ssize_t shmem_enabled_show(struct kobject *kobj,
3928 struct kobj_attribute *attr, char *buf)
3929{
3930 int values[] = {
3931 SHMEM_HUGE_ALWAYS,
3932 SHMEM_HUGE_WITHIN_SIZE,
3933 SHMEM_HUGE_ADVISE,
3934 SHMEM_HUGE_NEVER,
3935 SHMEM_HUGE_DENY,
3936 SHMEM_HUGE_FORCE,
3937 };
3938 int i, count;
3939
3940 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3941 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3942
3943 count += sprintf(buf + count, fmt,
3944 shmem_format_huge(values[i]));
3945 }
3946 buf[count - 1] = '\n';
3947 return count;
3948}
3949
3950static ssize_t shmem_enabled_store(struct kobject *kobj,
3951 struct kobj_attribute *attr, const char *buf, size_t count)
3952{
3953 char tmp[16];
3954 int huge;
3955
3956 if (count + 1 > sizeof(tmp))
3957 return -EINVAL;
3958 memcpy(tmp, buf, count);
3959 tmp[count] = '\0';
3960 if (count && tmp[count - 1] == '\n')
3961 tmp[count - 1] = '\0';
3962
3963 huge = shmem_parse_huge(tmp);
3964 if (huge == -EINVAL)
3965 return -EINVAL;
3966 if (!has_transparent_hugepage() &&
3967 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3968 return -EINVAL;
3969
3970 shmem_huge = huge;
3971 if (shmem_huge < SHMEM_HUGE_DENY)
3972 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3973 return count;
3974}
3975
3976struct kobj_attribute shmem_enabled_attr =
3977 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003978
3979bool shmem_huge_enabled(struct vm_area_struct *vma)
3980{
3981 struct inode *inode = file_inode(vma->vm_file);
3982 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3983 loff_t i_size;
3984 pgoff_t off;
3985
3986 if (shmem_huge == SHMEM_HUGE_FORCE)
3987 return true;
3988 if (shmem_huge == SHMEM_HUGE_DENY)
3989 return false;
3990 switch (sbinfo->huge) {
3991 case SHMEM_HUGE_NEVER:
3992 return false;
3993 case SHMEM_HUGE_ALWAYS:
3994 return true;
3995 case SHMEM_HUGE_WITHIN_SIZE:
3996 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3997 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3998 if (i_size >= HPAGE_PMD_SIZE &&
3999 i_size >> PAGE_SHIFT >= off)
4000 return true;
4001 case SHMEM_HUGE_ADVISE:
4002 /* TODO: implement fadvise() hints */
4003 return (vma->vm_flags & VM_HUGEPAGE);
4004 default:
4005 VM_BUG_ON(1);
4006 return false;
4007 }
4008}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004009#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004010
Matt Mackall853ac432009-01-06 14:40:20 -08004011#else /* !CONFIG_SHMEM */
4012
4013/*
4014 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4015 *
4016 * This is intended for small system where the benefits of the full
4017 * shmem code (swap-backed and resource-limited) are outweighed by
4018 * their complexity. On systems without swap this code should be
4019 * effectively equivalent, but much lighter weight.
4020 */
4021
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004022static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004023 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04004024 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08004025 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004026 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004027};
4028
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004029int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004030{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004031 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004032
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004033 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004034 BUG_ON(IS_ERR(shm_mnt));
4035
4036 return 0;
4037}
4038
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004039int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08004040{
4041 return 0;
4042}
4043
Hugh Dickins3f96b792009-09-21 17:03:37 -07004044int shmem_lock(struct file *file, int lock, struct user_struct *user)
4045{
4046 return 0;
4047}
4048
Hugh Dickins24513262012-01-20 14:34:21 -08004049void shmem_unlock_mapping(struct address_space *mapping)
4050{
4051}
4052
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004053#ifdef CONFIG_MMU
4054unsigned long shmem_get_unmapped_area(struct file *file,
4055 unsigned long addr, unsigned long len,
4056 unsigned long pgoff, unsigned long flags)
4057{
4058 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4059}
4060#endif
4061
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004062void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004063{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004064 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004065}
4066EXPORT_SYMBOL_GPL(shmem_truncate_range);
4067
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004068#define shmem_vm_ops generic_file_vm_ops
4069#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004070#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004071#define shmem_acct_size(flags, size) 0
4072#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004073
4074#endif /* CONFIG_SHMEM */
4075
4076/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077
Al Viro34515382013-02-14 22:38:02 -05004078static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04004079 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05004080};
4081
Eric Parisc7277092013-12-02 11:24:19 +00004082static struct file *__shmem_file_setup(const char *name, loff_t size,
4083 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084{
Al Viro6b4d0b22013-02-14 21:37:26 -05004085 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04004087 struct path path;
Al Viro34515382013-02-14 22:38:02 -05004088 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 struct qstr this;
4090
4091 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05004092 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093
Hugh Dickins285b2c42011-08-03 16:21:20 -07004094 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 return ERR_PTR(-EINVAL);
4096
4097 if (shmem_acct_size(flags, size))
4098 return ERR_PTR(-ENOMEM);
4099
Al Viro6b4d0b22013-02-14 21:37:26 -05004100 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101 this.name = name;
4102 this.len = strlen(name);
4103 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004104 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004105 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004106 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004107 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004109 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004110
Al Viro6b4d0b22013-02-14 21:37:26 -05004111 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004112 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004114 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115
Eric Parisc7277092013-12-02 11:24:19 +00004116 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004117 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004119 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004120 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4121 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004122 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004123
Al Viro6b4d0b22013-02-14 21:37:26 -05004124 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004125 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004126 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004127 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004128
Al Viro6b4d0b22013-02-14 21:37:26 -05004129 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131put_memory:
4132 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004133put_path:
4134 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004135 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136}
Eric Parisc7277092013-12-02 11:24:19 +00004137
4138/**
4139 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4140 * kernel internal. There will be NO LSM permission checks against the
4141 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004142 * higher layer. The users are the big_key and shm implementations. LSM
4143 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004144 * @name: name for dentry (to be seen in /proc/<pid>/maps
4145 * @size: size to be set for the file
4146 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4147 */
4148struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4149{
4150 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4151}
4152
4153/**
4154 * shmem_file_setup - get an unlinked file living in tmpfs
4155 * @name: name for dentry (to be seen in /proc/<pid>/maps
4156 * @size: size to be set for the file
4157 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4158 */
4159struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4160{
4161 return __shmem_file_setup(name, size, flags, 0);
4162}
Keith Packard395e0dd2008-06-20 00:08:06 -07004163EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004164
Randy Dunlap46711812008-03-19 17:00:41 -07004165/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004167 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4168 */
4169int shmem_zero_setup(struct vm_area_struct *vma)
4170{
4171 struct file *file;
4172 loff_t size = vma->vm_end - vma->vm_start;
4173
Hugh Dickins66fc1302015-06-14 09:48:09 -07004174 /*
4175 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4176 * between XFS directory reading and selinux: since this file is only
4177 * accessible to the user through its mapping, use S_PRIVATE flag to
4178 * bypass file security, in the same way as shmem_kernel_file_setup().
4179 */
4180 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 if (IS_ERR(file))
4182 return PTR_ERR(file);
4183
4184 if (vma->vm_file)
4185 fput(vma->vm_file);
4186 vma->vm_file = file;
4187 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004188
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004189 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004190 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4191 (vma->vm_end & HPAGE_PMD_MASK)) {
4192 khugepaged_enter(vma, vma->vm_flags);
4193 }
4194
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 return 0;
4196}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004197
4198/**
4199 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4200 * @mapping: the page's address_space
4201 * @index: the page index
4202 * @gfp: the page allocator flags to use if allocating
4203 *
4204 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4205 * with any new page allocations done using the specified allocation flags.
4206 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4207 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4208 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4209 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004210 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4211 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004212 */
4213struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4214 pgoff_t index, gfp_t gfp)
4215{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004216#ifdef CONFIG_SHMEM
4217 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004218 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004219 int error;
4220
4221 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004222 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4223 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004224 if (error)
4225 page = ERR_PTR(error);
4226 else
4227 unlock_page(page);
4228 return page;
4229#else
4230 /*
4231 * The tiny !SHMEM case uses ramfs without swap
4232 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004233 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004234#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004235}
4236EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);