blob: 2ac19a61d5655b82d6aa6b01a37eb6d05ea3a72a [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;
1365 pgoff_t idx, hindex = round_down(index, HPAGE_PMD_NR);
1366 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
1372 rcu_read_lock();
1373 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1374 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1375 rcu_read_unlock();
1376 return NULL;
1377 }
1378 rcu_read_unlock();
1379
1380 shmem_pseudo_vma_init(&pvma, info, hindex);
1381 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1382 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1383 shmem_pseudo_vma_destroy(&pvma);
1384 if (page)
1385 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001386 return page;
1387}
1388
1389static struct page *shmem_alloc_page(gfp_t gfp,
1390 struct shmem_inode_info *info, pgoff_t index)
1391{
1392 struct vm_area_struct pvma;
1393 struct page *page;
1394
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001395 shmem_pseudo_vma_init(&pvma, info, index);
1396 page = alloc_page_vma(gfp, &pvma, 0);
1397 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001398
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001399 return page;
1400}
1401
1402static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1403 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1404 pgoff_t index, bool huge)
1405{
1406 struct page *page;
1407 int nr;
1408 int err = -ENOSPC;
1409
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001410 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001411 huge = false;
1412 nr = huge ? HPAGE_PMD_NR : 1;
1413
1414 if (shmem_acct_block(info->flags, nr))
1415 goto failed;
1416 if (sbinfo->max_blocks) {
1417 if (percpu_counter_compare(&sbinfo->used_blocks,
1418 sbinfo->max_blocks - nr) > 0)
1419 goto unacct;
1420 percpu_counter_add(&sbinfo->used_blocks, nr);
1421 }
1422
1423 if (huge)
1424 page = shmem_alloc_hugepage(gfp, info, index);
1425 else
1426 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001427 if (page) {
1428 __SetPageLocked(page);
1429 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001430 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001431 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001432
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001433 err = -ENOMEM;
1434 if (sbinfo->max_blocks)
1435 percpu_counter_add(&sbinfo->used_blocks, -nr);
1436unacct:
1437 shmem_unacct_blocks(info->flags, nr);
1438failed:
1439 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001443 * When a page is moved from swapcache to shmem filecache (either by the
1444 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1445 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1446 * ignorance of the mapping it belongs to. If that mapping has special
1447 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1448 * we may need to copy to a suitable page before moving to filecache.
1449 *
1450 * In a future release, this may well be extended to respect cpuset and
1451 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1452 * but for now it is a simple matter of zone.
1453 */
1454static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1455{
1456 return page_zonenum(page) > gfp_zone(gfp);
1457}
1458
1459static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1460 struct shmem_inode_info *info, pgoff_t index)
1461{
1462 struct page *oldpage, *newpage;
1463 struct address_space *swap_mapping;
1464 pgoff_t swap_index;
1465 int error;
1466
1467 oldpage = *pagep;
1468 swap_index = page_private(oldpage);
1469 swap_mapping = page_mapping(oldpage);
1470
1471 /*
1472 * We have arrived here because our zones are constrained, so don't
1473 * limit chance of success by further cpuset and node constraints.
1474 */
1475 gfp &= ~GFP_CONSTRAINT_MASK;
1476 newpage = shmem_alloc_page(gfp, info, index);
1477 if (!newpage)
1478 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001479
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001480 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001481 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001482 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001483
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001484 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001485 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001486 SetPageSwapCache(newpage);
1487
1488 /*
1489 * Our caller will very soon move newpage out of swapcache, but it's
1490 * a nice clean interface for us to replace oldpage by newpage there.
1491 */
1492 spin_lock_irq(&swap_mapping->tree_lock);
1493 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1494 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001495 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001496 __inc_node_page_state(newpage, NR_FILE_PAGES);
1497 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001498 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001499 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001500
Hugh Dickins0142ef62012-06-07 14:21:09 -07001501 if (unlikely(error)) {
1502 /*
1503 * Is this possible? I think not, now that our callers check
1504 * both PageSwapCache and page_private after getting page lock;
1505 * but be defensive. Reverse old to newpage for clear and free.
1506 */
1507 oldpage = newpage;
1508 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001509 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001510 lru_cache_add_anon(newpage);
1511 *pagep = newpage;
1512 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001513
1514 ClearPageSwapCache(oldpage);
1515 set_page_private(oldpage, 0);
1516
1517 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001518 put_page(oldpage);
1519 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001520 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001521}
1522
1523/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001524 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 *
1526 * If we allocate a new one we do not mark it dirty. That's up to the
1527 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001528 * entry since a page cannot live in both the swap and page cache.
1529 *
1530 * fault_mm and fault_type are only supplied by shmem_fault:
1531 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001533static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001534 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1535 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001538 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001540 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001541 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001542 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001544 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001545 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001547 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001548 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001550 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001552 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1553 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001555 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001556 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001557 if (radix_tree_exceptional_entry(page)) {
1558 swap = radix_to_swp_entry(page);
1559 page = NULL;
1560 }
1561
Hugh Dickins75edd342016-05-19 17:12:44 -07001562 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001563 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001564 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001565 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001566 }
1567
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001568 if (page && sgp == SGP_WRITE)
1569 mark_page_accessed(page);
1570
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001571 /* fallocated page? */
1572 if (page && !PageUptodate(page)) {
1573 if (sgp != SGP_READ)
1574 goto clear;
1575 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001576 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001577 page = NULL;
1578 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001579 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001580 *pagep = page;
1581 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
1584 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001585 * Fast cache lookup did not find it:
1586 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001588 info = SHMEM_I(inode);
1589 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001590 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 if (swap.val) {
1593 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001594 page = lookup_swap_cache(swap);
1595 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001596 /* Or update major stats only when swapin succeeds?? */
1597 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001598 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001599 count_vm_event(PGMAJFAULT);
1600 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1601 }
1602 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001603 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001604 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001605 error = -ENOMEM;
1606 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
1609
1610 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001611 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001612 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001613 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001614 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001615 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001616 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001617 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001619 goto failed;
1620 }
1621 wait_on_page_writeback(page);
1622
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001623 if (shmem_should_replace_page(page, gfp)) {
1624 error = shmem_replace_page(&page, gfp, info, index);
1625 if (error)
1626 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 }
1628
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001629 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001630 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001631 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001632 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001633 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001634 /*
1635 * We already confirmed swap under page lock, and make
1636 * no memory allocation here, so usually no possibility
1637 * of error; but free_swap_and_cache() only trylocks a
1638 * page, so it is just possible that the entry has been
1639 * truncated or holepunched since swap was confirmed.
1640 * shmem_undo_range() will have done some of the
1641 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001642 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001643 * Reset swap.val? No, leave it so "failed" goes back to
1644 * "repeat": reading a hole and writing should succeed.
1645 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001646 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001647 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001648 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001649 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001650 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001651 if (error)
1652 goto failed;
1653
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001654 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001655
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001656 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001657 info->swapped--;
1658 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001659 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001660
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001661 if (sgp == SGP_WRITE)
1662 mark_page_accessed(page);
1663
Hugh Dickins27ab7002011-07-25 17:12:36 -07001664 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001665 set_page_dirty(page);
1666 swap_free(swap);
1667
Hugh Dickins54af60422011-08-03 16:21:24 -07001668 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001669 /* shmem_symlink() */
1670 if (mapping->a_ops != &shmem_aops)
1671 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001672 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001673 goto alloc_nohuge;
1674 if (shmem_huge == SHMEM_HUGE_FORCE)
1675 goto alloc_huge;
1676 switch (sbinfo->huge) {
1677 loff_t i_size;
1678 pgoff_t off;
1679 case SHMEM_HUGE_NEVER:
1680 goto alloc_nohuge;
1681 case SHMEM_HUGE_WITHIN_SIZE:
1682 off = round_up(index, HPAGE_PMD_NR);
1683 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1684 if (i_size >= HPAGE_PMD_SIZE &&
1685 i_size >> PAGE_SHIFT >= off)
1686 goto alloc_huge;
1687 /* fallthrough */
1688 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001689 if (sgp_huge == SGP_HUGE)
1690 goto alloc_huge;
1691 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001692 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001695alloc_huge:
1696 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1697 index, true);
1698 if (IS_ERR(page)) {
1699alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1700 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001701 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001702 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001703 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001704 error = PTR_ERR(page);
1705 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001706 if (error != -ENOSPC)
1707 goto failed;
1708 /*
1709 * Try to reclaim some spece by splitting a huge page
1710 * beyond i_size on the filesystem.
1711 */
1712 while (retry--) {
1713 int ret;
1714 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1715 if (ret == SHRINK_STOP)
1716 break;
1717 if (ret)
1718 goto alloc_nohuge;
1719 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001720 goto failed;
1721 }
1722
1723 if (PageTransHuge(page))
1724 hindex = round_down(index, HPAGE_PMD_NR);
1725 else
1726 hindex = index;
1727
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001728 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001729 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001730
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001731 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001732 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001733 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001734 goto unacct;
1735 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1736 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001737 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001738 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001739 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001740 radix_tree_preload_end();
1741 }
1742 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001743 mem_cgroup_cancel_charge(page, memcg,
1744 PageTransHuge(page));
1745 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001746 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001747 mem_cgroup_commit_charge(page, memcg, false,
1748 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001749 lru_cache_add_anon(page);
1750
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001751 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001752 info->alloced += 1 << compound_order(page);
1753 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001754 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001755 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001756 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001757
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001758 if (PageTransHuge(page) &&
1759 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1760 hindex + HPAGE_PMD_NR - 1) {
1761 /*
1762 * Part of the huge page is beyond i_size: subject
1763 * to shrink under memory pressure.
1764 */
1765 spin_lock(&sbinfo->shrinklist_lock);
1766 if (list_empty(&info->shrinklist)) {
1767 list_add_tail(&info->shrinklist,
1768 &sbinfo->shrinklist);
1769 sbinfo->shrinklist_len++;
1770 }
1771 spin_unlock(&sbinfo->shrinklist_lock);
1772 }
1773
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001774 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001775 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1776 */
1777 if (sgp == SGP_FALLOC)
1778 sgp = SGP_WRITE;
1779clear:
1780 /*
1781 * Let SGP_WRITE caller clear ends if write does not fill page;
1782 * but SGP_FALLOC on a page fallocated earlier must initialize
1783 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001784 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001785 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1786 struct page *head = compound_head(page);
1787 int i;
1788
1789 for (i = 0; i < (1 << compound_order(head)); i++) {
1790 clear_highpage(head + i);
1791 flush_dcache_page(head + i);
1792 }
1793 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001794 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001795 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001796
Hugh Dickins54af60422011-08-03 16:21:24 -07001797 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001798 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001799 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001800 if (alloced) {
1801 ClearPageDirty(page);
1802 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001803 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001804 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001805 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001806 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001807 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001808 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001809 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001810 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001811 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001812
Nick Piggind0217ac2007-07-19 01:47:03 -07001813 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001814 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001816unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001817 if (sbinfo->max_blocks)
1818 percpu_counter_sub(&sbinfo->used_blocks,
1819 1 << compound_order(page));
1820 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1821
1822 if (PageTransHuge(page)) {
1823 unlock_page(page);
1824 put_page(page);
1825 goto alloc_nohuge;
1826 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001827failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001828 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001829 error = -EEXIST;
1830unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001831 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001832 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001833 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001834 }
1835 if (error == -ENOSPC && !once++) {
1836 info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001837 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001838 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001839 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001840 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001841 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001842 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001843 goto repeat;
1844 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845}
1846
1847static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1848{
Al Viro496ad9a2013-01-23 17:07:38 -05001849 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001850 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001851 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001853 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001855 /*
1856 * Trinity finds that probing a hole which tmpfs is punching can
1857 * prevent the hole-punch from ever completing: which in turn
1858 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001859 * faulting pages into the hole while it's being punched. Although
1860 * shmem_undo_range() does remove the additions, it may be unable to
1861 * keep up, as each new page needs its own unmap_mapping_range() call,
1862 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1863 *
1864 * It does not matter if we sometimes reach this check just before the
1865 * hole-punch begins, so that one fault then races with the punch:
1866 * we just need to make racing faults a rare case.
1867 *
1868 * The implementation below would be much simpler if we just used a
1869 * standard mutex or completion: but we cannot take i_mutex in fault,
1870 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001871 */
1872 if (unlikely(inode->i_private)) {
1873 struct shmem_falloc *shmem_falloc;
1874
1875 spin_lock(&inode->i_lock);
1876 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001877 if (shmem_falloc &&
1878 shmem_falloc->waitq &&
1879 vmf->pgoff >= shmem_falloc->start &&
1880 vmf->pgoff < shmem_falloc->next) {
1881 wait_queue_head_t *shmem_falloc_waitq;
1882 DEFINE_WAIT(shmem_fault_wait);
1883
1884 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001885 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1886 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001887 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001888 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001889 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001890 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001891
1892 shmem_falloc_waitq = shmem_falloc->waitq;
1893 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1894 TASK_UNINTERRUPTIBLE);
1895 spin_unlock(&inode->i_lock);
1896 schedule();
1897
1898 /*
1899 * shmem_falloc_waitq points into the shmem_fallocate()
1900 * stack of the hole-punching task: shmem_falloc_waitq
1901 * is usually invalid by the time we reach here, but
1902 * finish_wait() does not dereference it in that case;
1903 * though i_lock needed lest racing with wake_up_all().
1904 */
1905 spin_lock(&inode->i_lock);
1906 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1907 spin_unlock(&inode->i_lock);
1908 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001909 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001910 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001911 }
1912
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001913 sgp = SGP_CACHE;
1914 if (vma->vm_flags & VM_HUGEPAGE)
1915 sgp = SGP_HUGE;
1916 else if (vma->vm_flags & VM_NOHUGEPAGE)
1917 sgp = SGP_NOHUGE;
1918
1919 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001920 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 if (error)
1922 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001923 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924}
1925
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001926unsigned long shmem_get_unmapped_area(struct file *file,
1927 unsigned long uaddr, unsigned long len,
1928 unsigned long pgoff, unsigned long flags)
1929{
1930 unsigned long (*get_area)(struct file *,
1931 unsigned long, unsigned long, unsigned long, unsigned long);
1932 unsigned long addr;
1933 unsigned long offset;
1934 unsigned long inflated_len;
1935 unsigned long inflated_addr;
1936 unsigned long inflated_offset;
1937
1938 if (len > TASK_SIZE)
1939 return -ENOMEM;
1940
1941 get_area = current->mm->get_unmapped_area;
1942 addr = get_area(file, uaddr, len, pgoff, flags);
1943
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001944 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001945 return addr;
1946 if (IS_ERR_VALUE(addr))
1947 return addr;
1948 if (addr & ~PAGE_MASK)
1949 return addr;
1950 if (addr > TASK_SIZE - len)
1951 return addr;
1952
1953 if (shmem_huge == SHMEM_HUGE_DENY)
1954 return addr;
1955 if (len < HPAGE_PMD_SIZE)
1956 return addr;
1957 if (flags & MAP_FIXED)
1958 return addr;
1959 /*
1960 * Our priority is to support MAP_SHARED mapped hugely;
1961 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1962 * But if caller specified an address hint, respect that as before.
1963 */
1964 if (uaddr)
1965 return addr;
1966
1967 if (shmem_huge != SHMEM_HUGE_FORCE) {
1968 struct super_block *sb;
1969
1970 if (file) {
1971 VM_BUG_ON(file->f_op != &shmem_file_operations);
1972 sb = file_inode(file)->i_sb;
1973 } else {
1974 /*
1975 * Called directly from mm/mmap.c, or drivers/char/mem.c
1976 * for "/dev/zero", to create a shared anonymous object.
1977 */
1978 if (IS_ERR(shm_mnt))
1979 return addr;
1980 sb = shm_mnt->mnt_sb;
1981 }
1982 if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER)
1983 return addr;
1984 }
1985
1986 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
1987 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
1988 return addr;
1989 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
1990 return addr;
1991
1992 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
1993 if (inflated_len > TASK_SIZE)
1994 return addr;
1995 if (inflated_len < len)
1996 return addr;
1997
1998 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
1999 if (IS_ERR_VALUE(inflated_addr))
2000 return addr;
2001 if (inflated_addr & ~PAGE_MASK)
2002 return addr;
2003
2004 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2005 inflated_addr += offset - inflated_offset;
2006 if (inflated_offset > offset)
2007 inflated_addr += HPAGE_PMD_SIZE;
2008
2009 if (inflated_addr > TASK_SIZE - len)
2010 return addr;
2011 return inflated_addr;
2012}
2013
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002015static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016{
Al Viro496ad9a2013-01-23 17:07:38 -05002017 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002018 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019}
2020
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002021static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2022 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023{
Al Viro496ad9a2013-01-23 17:07:38 -05002024 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002025 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002027 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2028 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029}
2030#endif
2031
2032int shmem_lock(struct file *file, int lock, struct user_struct *user)
2033{
Al Viro496ad9a2013-01-23 17:07:38 -05002034 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 struct shmem_inode_info *info = SHMEM_I(inode);
2036 int retval = -ENOMEM;
2037
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002038 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 if (lock && !(info->flags & VM_LOCKED)) {
2040 if (!user_shm_lock(inode->i_size, user))
2041 goto out_nomem;
2042 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002043 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 }
2045 if (!lock && (info->flags & VM_LOCKED) && user) {
2046 user_shm_unlock(inode->i_size, user);
2047 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002048 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002051
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002053 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 return retval;
2055}
2056
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002057static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058{
2059 file_accessed(file);
2060 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002061 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002062 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2063 (vma->vm_end & HPAGE_PMD_MASK)) {
2064 khugepaged_enter(vma, vma->vm_flags);
2065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 return 0;
2067}
2068
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002069static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002070 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071{
2072 struct inode *inode;
2073 struct shmem_inode_info *info;
2074 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2075
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002076 if (shmem_reserve_inode(sb))
2077 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
2079 inode = new_inode(sb);
2080 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002081 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002082 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07002085 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 info = SHMEM_I(inode);
2087 memset(info, 0, (char *)inode - (char *)info);
2088 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002089 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002090 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002091 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002093 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002094 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
2096 switch (mode & S_IFMT) {
2097 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002098 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 init_special_inode(inode, mode, dev);
2100 break;
2101 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002102 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 inode->i_op = &shmem_inode_operations;
2104 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002105 mpol_shared_policy_init(&info->policy,
2106 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 break;
2108 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002109 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 /* Some things misbehave if size == 0 on a directory */
2111 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2112 inode->i_op = &shmem_dir_inode_operations;
2113 inode->i_fop = &simple_dir_operations;
2114 break;
2115 case S_IFLNK:
2116 /*
2117 * Must not load anything in the rbtree,
2118 * mpol_free_shared_policy will not be called.
2119 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002120 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 break;
2122 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002123 } else
2124 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 return inode;
2126}
2127
Johannes Weiner0cd61442014-04-03 14:47:46 -07002128bool shmem_mapping(struct address_space *mapping)
2129{
Sasha Levinf0774d82015-02-23 05:38:00 -05002130 if (!mapping->host)
2131 return false;
2132
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002133 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002134}
2135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002137static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002138static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002140#ifdef CONFIG_TMPFS_XATTR
2141static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2142#else
2143#define shmem_initxattrs NULL
2144#endif
2145
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002147shmem_write_begin(struct file *file, struct address_space *mapping,
2148 loff_t pos, unsigned len, unsigned flags,
2149 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
Nick Piggin800d15a2007-10-16 01:25:03 -07002151 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002152 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002153 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002154
2155 /* i_mutex is held by caller */
2156 if (unlikely(info->seals)) {
2157 if (info->seals & F_SEAL_WRITE)
2158 return -EPERM;
2159 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2160 return -EPERM;
2161 }
2162
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002163 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002164}
2165
2166static int
2167shmem_write_end(struct file *file, struct address_space *mapping,
2168 loff_t pos, unsigned len, unsigned copied,
2169 struct page *page, void *fsdata)
2170{
2171 struct inode *inode = mapping->host;
2172
Hugh Dickinsd3602442008-02-04 22:28:44 -08002173 if (pos + copied > inode->i_size)
2174 i_size_write(inode, pos + copied);
2175
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002176 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002177 struct page *head = compound_head(page);
2178 if (PageTransCompound(page)) {
2179 int i;
2180
2181 for (i = 0; i < HPAGE_PMD_NR; i++) {
2182 if (head + i == page)
2183 continue;
2184 clear_highpage(head + i);
2185 flush_dcache_page(head + i);
2186 }
2187 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002188 if (copied < PAGE_SIZE) {
2189 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002190 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002191 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002192 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002193 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002194 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002195 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002196 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002197 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002198
Nick Piggin800d15a2007-10-16 01:25:03 -07002199 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200}
2201
Al Viro2ba5bbe2014-04-02 20:00:02 -04002202static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203{
Al Viro6e58e792014-02-03 17:07:03 -05002204 struct file *file = iocb->ki_filp;
2205 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002207 pgoff_t index;
2208 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002209 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002210 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002211 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002212 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002213
2214 /*
2215 * Might this read be for a stacking filesystem? Then when reading
2216 * holes of a sparse file, we actually need to allocate those pages,
2217 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2218 */
Al Viro777eda22014-12-17 04:46:46 -05002219 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002220 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002222 index = *ppos >> PAGE_SHIFT;
2223 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
2225 for (;;) {
2226 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002227 pgoff_t end_index;
2228 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 loff_t i_size = i_size_read(inode);
2230
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002231 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 if (index > end_index)
2233 break;
2234 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002235 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 if (nr <= offset)
2237 break;
2238 }
2239
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002240 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002241 if (error) {
2242 if (error == -EINVAL)
2243 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 break;
2245 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002246 if (page) {
2247 if (sgp == SGP_CACHE)
2248 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002249 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
2252 /*
2253 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002254 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002256 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002258 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002260 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 if (nr <= offset) {
2262 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002263 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 break;
2265 }
2266 }
2267 nr -= offset;
2268
2269 if (page) {
2270 /*
2271 * If users can be writing to this page using arbitrary
2272 * virtual addresses, take care about potential aliasing
2273 * before reading the page on the kernel side.
2274 */
2275 if (mapping_writably_mapped(mapping))
2276 flush_dcache_page(page);
2277 /*
2278 * Mark the page accessed if we read the beginning.
2279 */
2280 if (!offset)
2281 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002282 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002284 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287 /*
2288 * Ok, we have the page, and it's up-to-date, so
2289 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002291 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002292 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002294 index += offset >> PAGE_SHIFT;
2295 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002297 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002298 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 break;
Al Viro6e58e792014-02-03 17:07:03 -05002300 if (ret < nr) {
2301 error = -EFAULT;
2302 break;
2303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 cond_resched();
2305 }
2306
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002307 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002308 file_accessed(file);
2309 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310}
2311
Hugh Dickins708e3502011-07-25 17:12:32 -07002312static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2313 struct pipe_inode_info *pipe, size_t len,
2314 unsigned int flags)
2315{
2316 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002317 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07002318 unsigned int loff, nr_pages, req_pages;
2319 struct page *pages[PIPE_DEF_BUFFERS];
2320 struct partial_page partial[PIPE_DEF_BUFFERS];
2321 struct page *page;
2322 pgoff_t index, end_index;
2323 loff_t isize, left;
2324 int error, page_nr;
2325 struct splice_pipe_desc spd = {
2326 .pages = pages,
2327 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02002328 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07002329 .flags = flags,
2330 .ops = &page_cache_pipe_buf_ops,
2331 .spd_release = spd_release_page,
2332 };
2333
Hugh Dickins71f0e072011-07-25 17:12:33 -07002334 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07002335 if (unlikely(*ppos >= isize))
2336 return 0;
2337
2338 left = isize - *ppos;
2339 if (unlikely(left < len))
2340 len = left;
2341
2342 if (splice_grow_spd(pipe, &spd))
2343 return -ENOMEM;
2344
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002345 index = *ppos >> PAGE_SHIFT;
2346 loff = *ppos & ~PAGE_MASK;
2347 req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT;
Al Viroa786c062014-04-11 12:01:03 -04002348 nr_pages = min(req_pages, spd.nr_pages_max);
Hugh Dickins708e3502011-07-25 17:12:32 -07002349
Hugh Dickins708e3502011-07-25 17:12:32 -07002350 spd.nr_pages = find_get_pages_contig(mapping, index,
2351 nr_pages, spd.pages);
2352 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07002353 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002354
Hugh Dickins708e3502011-07-25 17:12:32 -07002355 while (spd.nr_pages < nr_pages) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002356 error = shmem_getpage(inode, index, &page, SGP_CACHE);
Hugh Dickins71f0e072011-07-25 17:12:33 -07002357 if (error)
2358 break;
2359 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07002360 spd.pages[spd.nr_pages++] = page;
2361 index++;
2362 }
2363
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002364 index = *ppos >> PAGE_SHIFT;
Hugh Dickins708e3502011-07-25 17:12:32 -07002365 nr_pages = spd.nr_pages;
2366 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002367
Hugh Dickins708e3502011-07-25 17:12:32 -07002368 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
2369 unsigned int this_len;
2370
2371 if (!len)
2372 break;
2373
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002374 this_len = min_t(unsigned long, len, PAGE_SIZE - loff);
Hugh Dickins708e3502011-07-25 17:12:32 -07002375 page = spd.pages[page_nr];
2376
Hugh Dickins71f0e072011-07-25 17:12:33 -07002377 if (!PageUptodate(page) || page->mapping != mapping) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002378 error = shmem_getpage(inode, index, &page, SGP_CACHE);
Hugh Dickins71f0e072011-07-25 17:12:33 -07002379 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07002380 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07002381 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002382 put_page(spd.pages[page_nr]);
Hugh Dickins71f0e072011-07-25 17:12:33 -07002383 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07002384 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07002385
2386 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002387 end_index = (isize - 1) >> PAGE_SHIFT;
Hugh Dickins708e3502011-07-25 17:12:32 -07002388 if (unlikely(!isize || index > end_index))
2389 break;
2390
Hugh Dickins708e3502011-07-25 17:12:32 -07002391 if (end_index == index) {
2392 unsigned int plen;
2393
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002394 plen = ((isize - 1) & ~PAGE_MASK) + 1;
Hugh Dickins708e3502011-07-25 17:12:32 -07002395 if (plen <= loff)
2396 break;
2397
Hugh Dickins708e3502011-07-25 17:12:32 -07002398 this_len = min(this_len, plen - loff);
2399 len = this_len;
2400 }
2401
2402 spd.partial[page_nr].offset = loff;
2403 spd.partial[page_nr].len = this_len;
2404 len -= this_len;
2405 loff = 0;
2406 spd.nr_pages++;
2407 index++;
2408 }
2409
Hugh Dickins708e3502011-07-25 17:12:32 -07002410 while (page_nr < nr_pages)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002411 put_page(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07002412
2413 if (spd.nr_pages)
2414 error = splice_to_pipe(pipe, &spd);
2415
Eric Dumazet047fe362012-06-12 15:24:40 +02002416 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07002417
2418 if (error > 0) {
2419 *ppos += error;
2420 file_accessed(in);
2421 }
2422 return error;
2423}
2424
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002425/*
2426 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2427 */
2428static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002429 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002430{
2431 struct page *page;
2432 struct pagevec pvec;
2433 pgoff_t indices[PAGEVEC_SIZE];
2434 bool done = false;
2435 int i;
2436
2437 pagevec_init(&pvec, 0);
2438 pvec.nr = 1; /* start small: we may be there already */
2439 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002440 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002441 pvec.nr, pvec.pages, indices);
2442 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002443 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002444 index = end;
2445 break;
2446 }
2447 for (i = 0; i < pvec.nr; i++, index++) {
2448 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002449 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002450 done = true;
2451 break;
2452 }
2453 index = indices[i];
2454 }
2455 page = pvec.pages[i];
2456 if (page && !radix_tree_exceptional_entry(page)) {
2457 if (!PageUptodate(page))
2458 page = NULL;
2459 }
2460 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002461 (page && whence == SEEK_DATA) ||
2462 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002463 done = true;
2464 break;
2465 }
2466 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002467 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002468 pagevec_release(&pvec);
2469 pvec.nr = PAGEVEC_SIZE;
2470 cond_resched();
2471 }
2472 return index;
2473}
2474
Andrew Morton965c8e52012-12-17 15:59:39 -08002475static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002476{
2477 struct address_space *mapping = file->f_mapping;
2478 struct inode *inode = mapping->host;
2479 pgoff_t start, end;
2480 loff_t new_offset;
2481
Andrew Morton965c8e52012-12-17 15:59:39 -08002482 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2483 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002484 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002485 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002486 /* We're holding i_mutex so we can access i_size directly */
2487
2488 if (offset < 0)
2489 offset = -EINVAL;
2490 else if (offset >= inode->i_size)
2491 offset = -ENXIO;
2492 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002493 start = offset >> PAGE_SHIFT;
2494 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002495 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002496 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002497 if (new_offset > offset) {
2498 if (new_offset < inode->i_size)
2499 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002500 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002501 offset = -ENXIO;
2502 else
2503 offset = inode->i_size;
2504 }
2505 }
2506
Hugh Dickins387aae62013-08-04 11:30:25 -07002507 if (offset >= 0)
2508 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002509 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002510 return offset;
2511}
2512
David Herrmann05f65b52014-08-08 14:25:36 -07002513/*
2514 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2515 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2516 */
2517#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2518#define LAST_SCAN 4 /* about 150ms max */
2519
2520static void shmem_tag_pins(struct address_space *mapping)
2521{
2522 struct radix_tree_iter iter;
2523 void **slot;
2524 pgoff_t start;
2525 struct page *page;
2526
2527 lru_add_drain();
2528 start = 0;
2529 rcu_read_lock();
2530
David Herrmann05f65b52014-08-08 14:25:36 -07002531 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2532 page = radix_tree_deref_slot(slot);
2533 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002534 if (radix_tree_deref_retry(page)) {
2535 slot = radix_tree_iter_retry(&iter);
2536 continue;
2537 }
David Herrmann05f65b52014-08-08 14:25:36 -07002538 } else if (page_count(page) - page_mapcount(page) > 1) {
2539 spin_lock_irq(&mapping->tree_lock);
2540 radix_tree_tag_set(&mapping->page_tree, iter.index,
2541 SHMEM_TAG_PINNED);
2542 spin_unlock_irq(&mapping->tree_lock);
2543 }
2544
2545 if (need_resched()) {
2546 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002547 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002548 }
2549 }
2550 rcu_read_unlock();
2551}
2552
2553/*
2554 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2555 * via get_user_pages(), drivers might have some pending I/O without any active
2556 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2557 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2558 * them to be dropped.
2559 * The caller must guarantee that no new user will acquire writable references
2560 * to those pages to avoid races.
2561 */
David Herrmann40e041a2014-08-08 14:25:27 -07002562static int shmem_wait_for_pins(struct address_space *mapping)
2563{
David Herrmann05f65b52014-08-08 14:25:36 -07002564 struct radix_tree_iter iter;
2565 void **slot;
2566 pgoff_t start;
2567 struct page *page;
2568 int error, scan;
2569
2570 shmem_tag_pins(mapping);
2571
2572 error = 0;
2573 for (scan = 0; scan <= LAST_SCAN; scan++) {
2574 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2575 break;
2576
2577 if (!scan)
2578 lru_add_drain_all();
2579 else if (schedule_timeout_killable((HZ << scan) / 200))
2580 scan = LAST_SCAN;
2581
2582 start = 0;
2583 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002584 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2585 start, SHMEM_TAG_PINNED) {
2586
2587 page = radix_tree_deref_slot(slot);
2588 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002589 if (radix_tree_deref_retry(page)) {
2590 slot = radix_tree_iter_retry(&iter);
2591 continue;
2592 }
David Herrmann05f65b52014-08-08 14:25:36 -07002593
2594 page = NULL;
2595 }
2596
2597 if (page &&
2598 page_count(page) - page_mapcount(page) != 1) {
2599 if (scan < LAST_SCAN)
2600 goto continue_resched;
2601
2602 /*
2603 * On the last scan, we clean up all those tags
2604 * we inserted; but make a note that we still
2605 * found pages pinned.
2606 */
2607 error = -EBUSY;
2608 }
2609
2610 spin_lock_irq(&mapping->tree_lock);
2611 radix_tree_tag_clear(&mapping->page_tree,
2612 iter.index, SHMEM_TAG_PINNED);
2613 spin_unlock_irq(&mapping->tree_lock);
2614continue_resched:
2615 if (need_resched()) {
2616 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002617 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002618 }
2619 }
2620 rcu_read_unlock();
2621 }
2622
2623 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002624}
2625
2626#define F_ALL_SEALS (F_SEAL_SEAL | \
2627 F_SEAL_SHRINK | \
2628 F_SEAL_GROW | \
2629 F_SEAL_WRITE)
2630
2631int shmem_add_seals(struct file *file, unsigned int seals)
2632{
2633 struct inode *inode = file_inode(file);
2634 struct shmem_inode_info *info = SHMEM_I(inode);
2635 int error;
2636
2637 /*
2638 * SEALING
2639 * Sealing allows multiple parties to share a shmem-file but restrict
2640 * access to a specific subset of file operations. Seals can only be
2641 * added, but never removed. This way, mutually untrusted parties can
2642 * share common memory regions with a well-defined policy. A malicious
2643 * peer can thus never perform unwanted operations on a shared object.
2644 *
2645 * Seals are only supported on special shmem-files and always affect
2646 * the whole underlying inode. Once a seal is set, it may prevent some
2647 * kinds of access to the file. Currently, the following seals are
2648 * defined:
2649 * SEAL_SEAL: Prevent further seals from being set on this file
2650 * SEAL_SHRINK: Prevent the file from shrinking
2651 * SEAL_GROW: Prevent the file from growing
2652 * SEAL_WRITE: Prevent write access to the file
2653 *
2654 * As we don't require any trust relationship between two parties, we
2655 * must prevent seals from being removed. Therefore, sealing a file
2656 * only adds a given set of seals to the file, it never touches
2657 * existing seals. Furthermore, the "setting seals"-operation can be
2658 * sealed itself, which basically prevents any further seal from being
2659 * added.
2660 *
2661 * Semantics of sealing are only defined on volatile files. Only
2662 * anonymous shmem files support sealing. More importantly, seals are
2663 * never written to disk. Therefore, there's no plan to support it on
2664 * other file types.
2665 */
2666
2667 if (file->f_op != &shmem_file_operations)
2668 return -EINVAL;
2669 if (!(file->f_mode & FMODE_WRITE))
2670 return -EPERM;
2671 if (seals & ~(unsigned int)F_ALL_SEALS)
2672 return -EINVAL;
2673
Al Viro59551022016-01-22 15:40:57 -05002674 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002675
2676 if (info->seals & F_SEAL_SEAL) {
2677 error = -EPERM;
2678 goto unlock;
2679 }
2680
2681 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2682 error = mapping_deny_writable(file->f_mapping);
2683 if (error)
2684 goto unlock;
2685
2686 error = shmem_wait_for_pins(file->f_mapping);
2687 if (error) {
2688 mapping_allow_writable(file->f_mapping);
2689 goto unlock;
2690 }
2691 }
2692
2693 info->seals |= seals;
2694 error = 0;
2695
2696unlock:
Al Viro59551022016-01-22 15:40:57 -05002697 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002698 return error;
2699}
2700EXPORT_SYMBOL_GPL(shmem_add_seals);
2701
2702int shmem_get_seals(struct file *file)
2703{
2704 if (file->f_op != &shmem_file_operations)
2705 return -EINVAL;
2706
2707 return SHMEM_I(file_inode(file))->seals;
2708}
2709EXPORT_SYMBOL_GPL(shmem_get_seals);
2710
2711long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2712{
2713 long error;
2714
2715 switch (cmd) {
2716 case F_ADD_SEALS:
2717 /* disallow upper 32bit */
2718 if (arg > UINT_MAX)
2719 return -EINVAL;
2720
2721 error = shmem_add_seals(file, arg);
2722 break;
2723 case F_GET_SEALS:
2724 error = shmem_get_seals(file);
2725 break;
2726 default:
2727 error = -EINVAL;
2728 break;
2729 }
2730
2731 return error;
2732}
2733
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002734static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2735 loff_t len)
2736{
Al Viro496ad9a2013-01-23 17:07:38 -05002737 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002738 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002739 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002740 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002741 pgoff_t start, index, end;
2742 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002743
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002744 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2745 return -EOPNOTSUPP;
2746
Al Viro59551022016-01-22 15:40:57 -05002747 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002748
2749 if (mode & FALLOC_FL_PUNCH_HOLE) {
2750 struct address_space *mapping = file->f_mapping;
2751 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2752 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002753 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002754
David Herrmann40e041a2014-08-08 14:25:27 -07002755 /* protected by i_mutex */
2756 if (info->seals & F_SEAL_WRITE) {
2757 error = -EPERM;
2758 goto out;
2759 }
2760
Hugh Dickins8e205f72014-07-23 14:00:10 -07002761 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002762 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2763 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2764 spin_lock(&inode->i_lock);
2765 inode->i_private = &shmem_falloc;
2766 spin_unlock(&inode->i_lock);
2767
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002768 if ((u64)unmap_end > (u64)unmap_start)
2769 unmap_mapping_range(mapping, unmap_start,
2770 1 + unmap_end - unmap_start, 0);
2771 shmem_truncate_range(inode, offset, offset + len - 1);
2772 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002773
2774 spin_lock(&inode->i_lock);
2775 inode->i_private = NULL;
2776 wake_up_all(&shmem_falloc_waitq);
2777 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002778 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002779 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002780 }
2781
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002782 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2783 error = inode_newsize_ok(inode, offset + len);
2784 if (error)
2785 goto out;
2786
David Herrmann40e041a2014-08-08 14:25:27 -07002787 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2788 error = -EPERM;
2789 goto out;
2790 }
2791
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002792 start = offset >> PAGE_SHIFT;
2793 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002794 /* Try to avoid a swapstorm if len is impossible to satisfy */
2795 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2796 error = -ENOSPC;
2797 goto out;
2798 }
2799
Hugh Dickins8e205f72014-07-23 14:00:10 -07002800 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002801 shmem_falloc.start = start;
2802 shmem_falloc.next = start;
2803 shmem_falloc.nr_falloced = 0;
2804 shmem_falloc.nr_unswapped = 0;
2805 spin_lock(&inode->i_lock);
2806 inode->i_private = &shmem_falloc;
2807 spin_unlock(&inode->i_lock);
2808
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002809 for (index = start; index < end; index++) {
2810 struct page *page;
2811
2812 /*
2813 * Good, the fallocate(2) manpage permits EINTR: we may have
2814 * been interrupted because we are using up too much memory.
2815 */
2816 if (signal_pending(current))
2817 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002818 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2819 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002820 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002821 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002822 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002823 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002824 if (index > start) {
2825 shmem_undo_range(inode,
2826 (loff_t)start << PAGE_SHIFT,
2827 ((loff_t)index << PAGE_SHIFT) - 1, true);
2828 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002829 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002830 }
2831
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002832 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002833 * Inform shmem_writepage() how far we have reached.
2834 * No need for lock or barrier: we have the page lock.
2835 */
2836 shmem_falloc.next++;
2837 if (!PageUptodate(page))
2838 shmem_falloc.nr_falloced++;
2839
2840 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002841 * If !PageUptodate, leave it that way so that freeable pages
2842 * can be recognized if we need to rollback on error later.
2843 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002844 * than free the pages we are allocating (and SGP_CACHE pages
2845 * might still be clean: we now need to mark those dirty too).
2846 */
2847 set_page_dirty(page);
2848 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002849 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002850 cond_resched();
2851 }
2852
2853 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2854 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002855 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002856undone:
2857 spin_lock(&inode->i_lock);
2858 inode->i_private = NULL;
2859 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002860out:
Al Viro59551022016-01-22 15:40:57 -05002861 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002862 return error;
2863}
2864
David Howells726c3342006-06-23 02:02:58 -07002865static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866{
David Howells726c3342006-06-23 02:02:58 -07002867 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
2869 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002870 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002872 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002874 buf->f_bavail =
2875 buf->f_bfree = sbinfo->max_blocks -
2876 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002877 }
2878 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 buf->f_files = sbinfo->max_inodes;
2880 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 }
2882 /* else leave those fields 0 like simple_statfs */
2883 return 0;
2884}
2885
2886/*
2887 * File creation. Allocate an inode, and we're done..
2888 */
2889static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002890shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002892 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 int error = -ENOSPC;
2894
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002895 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002897 error = simple_acl_create(dir, inode);
2898 if (error)
2899 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002900 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002901 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002902 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002903 if (error && error != -EOPNOTSUPP)
2904 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002905
Al Viro718deb62009-12-16 19:35:36 -05002906 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 dir->i_size += BOGO_DIRENT_SIZE;
2908 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2909 d_instantiate(dentry, inode);
2910 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 }
2912 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002913out_iput:
2914 iput(inode);
2915 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916}
2917
Al Viro60545d02013-06-07 01:20:27 -04002918static int
2919shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2920{
2921 struct inode *inode;
2922 int error = -ENOSPC;
2923
2924 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2925 if (inode) {
2926 error = security_inode_init_security(inode, dir,
2927 NULL,
2928 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002929 if (error && error != -EOPNOTSUPP)
2930 goto out_iput;
2931 error = simple_acl_create(dir, inode);
2932 if (error)
2933 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002934 d_tmpfile(dentry, inode);
2935 }
2936 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002937out_iput:
2938 iput(inode);
2939 return error;
Al Viro60545d02013-06-07 01:20:27 -04002940}
2941
Al Viro18bb1db2011-07-26 01:41:39 -04002942static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943{
2944 int error;
2945
2946 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2947 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002948 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 return 0;
2950}
2951
Al Viro4acdaf22011-07-26 01:42:34 -04002952static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002953 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954{
2955 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2956}
2957
2958/*
2959 * Link a file..
2960 */
2961static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2962{
David Howells75c3cfa2015-03-17 22:26:12 +00002963 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002964 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965
2966 /*
2967 * No ordinary (disk based) filesystem counts links as inodes;
2968 * but each new link needs a new dentry, pinning lowmem, and
2969 * tmpfs dentries cannot be pruned until they are unlinked.
2970 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002971 ret = shmem_reserve_inode(inode->i_sb);
2972 if (ret)
2973 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974
2975 dir->i_size += BOGO_DIRENT_SIZE;
2976 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002977 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002978 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 dget(dentry); /* Extra pinning count for the created dentry */
2980 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002981out:
2982 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983}
2984
2985static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2986{
David Howells75c3cfa2015-03-17 22:26:12 +00002987 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002989 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2990 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991
2992 dir->i_size -= BOGO_DIRENT_SIZE;
2993 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002994 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 dput(dentry); /* Undo the count from "create" - this does all the work */
2996 return 0;
2997}
2998
2999static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3000{
3001 if (!simple_empty(dentry))
3002 return -ENOTEMPTY;
3003
David Howells75c3cfa2015-03-17 22:26:12 +00003004 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003005 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 return shmem_unlink(dir, dentry);
3007}
3008
Miklos Szeredi37456772014-07-23 15:15:34 +02003009static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3010{
David Howellse36cb0b2015-01-29 12:02:35 +00003011 bool old_is_dir = d_is_dir(old_dentry);
3012 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003013
3014 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3015 if (old_is_dir) {
3016 drop_nlink(old_dir);
3017 inc_nlink(new_dir);
3018 } else {
3019 drop_nlink(new_dir);
3020 inc_nlink(old_dir);
3021 }
3022 }
3023 old_dir->i_ctime = old_dir->i_mtime =
3024 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003025 d_inode(old_dentry)->i_ctime =
3026 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
Miklos Szeredi37456772014-07-23 15:15:34 +02003027
3028 return 0;
3029}
3030
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003031static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3032{
3033 struct dentry *whiteout;
3034 int error;
3035
3036 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3037 if (!whiteout)
3038 return -ENOMEM;
3039
3040 error = shmem_mknod(old_dir, whiteout,
3041 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3042 dput(whiteout);
3043 if (error)
3044 return error;
3045
3046 /*
3047 * Cheat and hash the whiteout while the old dentry is still in
3048 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3049 *
3050 * d_lookup() will consistently find one of them at this point,
3051 * not sure which one, but that isn't even important.
3052 */
3053 d_rehash(whiteout);
3054 return 0;
3055}
3056
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057/*
3058 * The VFS layer already does all the dentry stuff for rename,
3059 * we just have to decrement the usage count for the target if
3060 * it exists so that the VFS layer correctly free's it when it
3061 * gets overwritten.
3062 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003063static 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 -07003064{
David Howells75c3cfa2015-03-17 22:26:12 +00003065 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 int they_are_dirs = S_ISDIR(inode->i_mode);
3067
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003068 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003069 return -EINVAL;
3070
Miklos Szeredi37456772014-07-23 15:15:34 +02003071 if (flags & RENAME_EXCHANGE)
3072 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3073
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 if (!simple_empty(new_dentry))
3075 return -ENOTEMPTY;
3076
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003077 if (flags & RENAME_WHITEOUT) {
3078 int error;
3079
3080 error = shmem_whiteout(old_dir, old_dentry);
3081 if (error)
3082 return error;
3083 }
3084
David Howells75c3cfa2015-03-17 22:26:12 +00003085 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003087 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003088 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003089 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003092 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003093 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 }
3095
3096 old_dir->i_size -= BOGO_DIRENT_SIZE;
3097 new_dir->i_size += BOGO_DIRENT_SIZE;
3098 old_dir->i_ctime = old_dir->i_mtime =
3099 new_dir->i_ctime = new_dir->i_mtime =
3100 inode->i_ctime = CURRENT_TIME;
3101 return 0;
3102}
3103
3104static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3105{
3106 int error;
3107 int len;
3108 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003109 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110 struct shmem_inode_info *info;
3111
3112 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003113 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 return -ENAMETOOLONG;
3115
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003116 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 if (!inode)
3118 return -ENOSPC;
3119
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003120 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003121 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003122 if (error) {
3123 if (error != -EOPNOTSUPP) {
3124 iput(inode);
3125 return error;
3126 }
3127 error = 0;
3128 }
3129
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 info = SHMEM_I(inode);
3131 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003132 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003133 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3134 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003135 iput(inode);
3136 return -ENOMEM;
3137 }
3138 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003140 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003141 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 if (error) {
3143 iput(inode);
3144 return error;
3145 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003146 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003148 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003149 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003151 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003152 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 dir->i_size += BOGO_DIRENT_SIZE;
3155 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
3156 d_instantiate(dentry, inode);
3157 dget(dentry);
3158 return 0;
3159}
3160
Al Virofceef392015-12-29 15:58:39 -05003161static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162{
Al Virofceef392015-12-29 15:58:39 -05003163 mark_page_accessed(arg);
3164 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165}
3166
Al Viro6b255392015-11-17 10:20:54 -05003167static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003168 struct inode *inode,
3169 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003172 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003173 if (!dentry) {
3174 page = find_get_page(inode->i_mapping, 0);
3175 if (!page)
3176 return ERR_PTR(-ECHILD);
3177 if (!PageUptodate(page)) {
3178 put_page(page);
3179 return ERR_PTR(-ECHILD);
3180 }
3181 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003182 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003183 if (error)
3184 return ERR_PTR(error);
3185 unlock_page(page);
3186 }
Al Virofceef392015-12-29 15:58:39 -05003187 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003188 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189}
3190
Eric Parisb09e0fa2011-05-24 17:12:39 -07003191#ifdef CONFIG_TMPFS_XATTR
3192/*
3193 * Superblocks without xattr inode operations may get some security.* xattr
3194 * support from the LSM "for free". As soon as we have any other xattrs
3195 * like ACLs, we also need to implement the security.* handlers at
3196 * filesystem level, though.
3197 */
3198
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003199/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003200 * Callback for security_inode_init_security() for acquiring xattrs.
3201 */
3202static int shmem_initxattrs(struct inode *inode,
3203 const struct xattr *xattr_array,
3204 void *fs_info)
3205{
3206 struct shmem_inode_info *info = SHMEM_I(inode);
3207 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003208 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003209 size_t len;
3210
3211 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003212 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003213 if (!new_xattr)
3214 return -ENOMEM;
3215
3216 len = strlen(xattr->name) + 1;
3217 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3218 GFP_KERNEL);
3219 if (!new_xattr->name) {
3220 kfree(new_xattr);
3221 return -ENOMEM;
3222 }
3223
3224 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3225 XATTR_SECURITY_PREFIX_LEN);
3226 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3227 xattr->name, len);
3228
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003229 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003230 }
3231
3232 return 0;
3233}
3234
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003235static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003236 struct dentry *unused, struct inode *inode,
3237 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003238{
Al Virob2968212016-04-10 20:48:24 -04003239 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003240
3241 name = xattr_full_name(handler, name);
3242 return simple_xattr_get(&info->xattrs, name, buffer, size);
3243}
3244
3245static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003246 struct dentry *unused, struct inode *inode,
3247 const char *name, const void *value,
3248 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003249{
Al Viro59301222016-05-27 10:19:30 -04003250 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003251
3252 name = xattr_full_name(handler, name);
3253 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3254}
3255
3256static const struct xattr_handler shmem_security_xattr_handler = {
3257 .prefix = XATTR_SECURITY_PREFIX,
3258 .get = shmem_xattr_handler_get,
3259 .set = shmem_xattr_handler_set,
3260};
3261
3262static const struct xattr_handler shmem_trusted_xattr_handler = {
3263 .prefix = XATTR_TRUSTED_PREFIX,
3264 .get = shmem_xattr_handler_get,
3265 .set = shmem_xattr_handler_set,
3266};
3267
Eric Parisb09e0fa2011-05-24 17:12:39 -07003268static const struct xattr_handler *shmem_xattr_handlers[] = {
3269#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003270 &posix_acl_access_xattr_handler,
3271 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003272#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003273 &shmem_security_xattr_handler,
3274 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003275 NULL
3276};
3277
Eric Parisb09e0fa2011-05-24 17:12:39 -07003278static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3279{
David Howells75c3cfa2015-03-17 22:26:12 +00003280 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003281 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003282}
3283#endif /* CONFIG_TMPFS_XATTR */
3284
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003285static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003287 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003288#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003289 .setxattr = generic_setxattr,
3290 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003291 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003292 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003293#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294};
3295
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003296static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003298 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003299#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003300 .setxattr = generic_setxattr,
3301 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003302 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003303 .removexattr = generic_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003304#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003305};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003306
David M. Grimes91828a42006-10-17 00:09:45 -07003307static struct dentry *shmem_get_parent(struct dentry *child)
3308{
3309 return ERR_PTR(-ESTALE);
3310}
3311
3312static int shmem_match(struct inode *ino, void *vfh)
3313{
3314 __u32 *fh = vfh;
3315 __u64 inum = fh[2];
3316 inum = (inum << 32) | fh[1];
3317 return ino->i_ino == inum && fh[0] == ino->i_generation;
3318}
3319
Christoph Hellwig480b1162007-10-21 16:42:13 -07003320static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3321 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003322{
David M. Grimes91828a42006-10-17 00:09:45 -07003323 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003324 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003325 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003326
Christoph Hellwig480b1162007-10-21 16:42:13 -07003327 if (fh_len < 3)
3328 return NULL;
3329
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003330 inum = fid->raw[2];
3331 inum = (inum << 32) | fid->raw[1];
3332
Christoph Hellwig480b1162007-10-21 16:42:13 -07003333 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3334 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003335 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003336 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003337 iput(inode);
3338 }
3339
Christoph Hellwig480b1162007-10-21 16:42:13 -07003340 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003341}
3342
Al Virob0b03822012-04-02 14:34:06 -04003343static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3344 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003345{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303346 if (*len < 3) {
3347 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003348 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303349 }
David M. Grimes91828a42006-10-17 00:09:45 -07003350
Al Viro1d3382cb2010-10-23 15:19:20 -04003351 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003352 /* Unfortunately insert_inode_hash is not idempotent,
3353 * so as we hash inodes here rather than at creation
3354 * time, we need a lock to ensure we only try
3355 * to do it once
3356 */
3357 static DEFINE_SPINLOCK(lock);
3358 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003359 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003360 __insert_inode_hash(inode,
3361 inode->i_ino + inode->i_generation);
3362 spin_unlock(&lock);
3363 }
3364
3365 fh[0] = inode->i_generation;
3366 fh[1] = inode->i_ino;
3367 fh[2] = ((__u64)inode->i_ino) >> 32;
3368
3369 *len = 3;
3370 return 1;
3371}
3372
Christoph Hellwig39655162007-10-21 16:42:17 -07003373static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003374 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003375 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003376 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003377};
3378
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003379static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3380 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381{
3382 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003383 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003384 uid_t uid;
3385 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003387 while (options != NULL) {
3388 this_char = options;
3389 for (;;) {
3390 /*
3391 * NUL-terminate this option: unfortunately,
3392 * mount options form a comma-separated list,
3393 * but mpol's nodelist may also contain commas.
3394 */
3395 options = strchr(options, ',');
3396 if (options == NULL)
3397 break;
3398 options++;
3399 if (!isdigit(*options)) {
3400 options[-1] = '\0';
3401 break;
3402 }
3403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 if (!*this_char)
3405 continue;
3406 if ((value = strchr(this_char,'=')) != NULL) {
3407 *value++ = 0;
3408 } else {
Joe Perches11705322016-03-17 14:19:50 -07003409 pr_err("tmpfs: No value for mount option '%s'\n",
3410 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003411 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412 }
3413
3414 if (!strcmp(this_char,"size")) {
3415 unsigned long long size;
3416 size = memparse(value,&rest);
3417 if (*rest == '%') {
3418 size <<= PAGE_SHIFT;
3419 size *= totalram_pages;
3420 do_div(size, 100);
3421 rest++;
3422 }
3423 if (*rest)
3424 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003425 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003426 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003428 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 if (*rest)
3430 goto bad_val;
3431 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003432 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433 if (*rest)
3434 goto bad_val;
3435 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003436 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003438 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439 if (*rest)
3440 goto bad_val;
3441 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003442 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003444 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 if (*rest)
3446 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003447 sbinfo->uid = make_kuid(current_user_ns(), uid);
3448 if (!uid_valid(sbinfo->uid))
3449 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003451 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003453 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 if (*rest)
3455 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003456 sbinfo->gid = make_kgid(current_user_ns(), gid);
3457 if (!gid_valid(sbinfo->gid))
3458 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003459#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003460 } else if (!strcmp(this_char, "huge")) {
3461 int huge;
3462 huge = shmem_parse_huge(value);
3463 if (huge < 0)
3464 goto bad_val;
3465 if (!has_transparent_hugepage() &&
3466 huge != SHMEM_HUGE_NEVER)
3467 goto bad_val;
3468 sbinfo->huge = huge;
3469#endif
3470#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003471 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003472 mpol_put(mpol);
3473 mpol = NULL;
3474 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003475 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003476#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 } else {
Joe Perches11705322016-03-17 14:19:50 -07003478 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003479 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 }
3481 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003482 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483 return 0;
3484
3485bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003486 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003488error:
3489 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 return 1;
3491
3492}
3493
3494static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3495{
3496 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003497 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003498 unsigned long inodes;
3499 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500
Greg Thelen5f001102013-02-22 16:36:01 -08003501 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003502 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003503 return error;
3504
3505 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003506 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003507 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003508 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003509 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003510 goto out;
3511 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003512 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003513 * but we must separately disallow unlimited->limited, because
3514 * in that case we have no record of how much is already in use.
3515 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003516 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003517 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003518 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003519 goto out;
3520
3521 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003522 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003523 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003524 sbinfo->max_inodes = config.max_inodes;
3525 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003526
Greg Thelen5f001102013-02-22 16:36:01 -08003527 /*
3528 * Preserve previous mempolicy unless mpol remount option was specified.
3529 */
3530 if (config.mpol) {
3531 mpol_put(sbinfo->mpol);
3532 sbinfo->mpol = config.mpol; /* transfers initial ref */
3533 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003534out:
3535 spin_unlock(&sbinfo->stat_lock);
3536 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003538
Al Viro34c80b12011-12-08 21:32:45 -05003539static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003540{
Al Viro34c80b12011-12-08 21:32:45 -05003541 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003542
3543 if (sbinfo->max_blocks != shmem_default_max_blocks())
3544 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003545 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003546 if (sbinfo->max_inodes != shmem_default_max_inodes())
3547 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3548 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003549 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003550 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3551 seq_printf(seq, ",uid=%u",
3552 from_kuid_munged(&init_user_ns, sbinfo->uid));
3553 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3554 seq_printf(seq, ",gid=%u",
3555 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003556#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003557 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3558 if (sbinfo->huge)
3559 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3560#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003561 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003562 return 0;
3563}
David Herrmann9183df22014-08-08 14:25:29 -07003564
3565#define MFD_NAME_PREFIX "memfd:"
3566#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3567#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3568
3569#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3570
3571SYSCALL_DEFINE2(memfd_create,
3572 const char __user *, uname,
3573 unsigned int, flags)
3574{
3575 struct shmem_inode_info *info;
3576 struct file *file;
3577 int fd, error;
3578 char *name;
3579 long len;
3580
3581 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3582 return -EINVAL;
3583
3584 /* length includes terminating zero */
3585 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3586 if (len <= 0)
3587 return -EFAULT;
3588 if (len > MFD_NAME_MAX_LEN + 1)
3589 return -EINVAL;
3590
3591 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3592 if (!name)
3593 return -ENOMEM;
3594
3595 strcpy(name, MFD_NAME_PREFIX);
3596 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3597 error = -EFAULT;
3598 goto err_name;
3599 }
3600
3601 /* terminating-zero may have changed after strnlen_user() returned */
3602 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3603 error = -EFAULT;
3604 goto err_name;
3605 }
3606
3607 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3608 if (fd < 0) {
3609 error = fd;
3610 goto err_name;
3611 }
3612
3613 file = shmem_file_setup(name, 0, VM_NORESERVE);
3614 if (IS_ERR(file)) {
3615 error = PTR_ERR(file);
3616 goto err_fd;
3617 }
3618 info = SHMEM_I(file_inode(file));
3619 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3620 file->f_flags |= O_RDWR | O_LARGEFILE;
3621 if (flags & MFD_ALLOW_SEALING)
3622 info->seals &= ~F_SEAL_SEAL;
3623
3624 fd_install(fd, file);
3625 kfree(name);
3626 return fd;
3627
3628err_fd:
3629 put_unused_fd(fd);
3630err_name:
3631 kfree(name);
3632 return error;
3633}
3634
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003635#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636
3637static void shmem_put_super(struct super_block *sb)
3638{
Hugh Dickins602586a2010-08-17 15:23:56 -07003639 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3640
3641 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003642 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003643 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 sb->s_fs_info = NULL;
3645}
3646
Kay Sievers2b2af542009-04-30 15:23:42 +02003647int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648{
3649 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003650 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003651 int err = -ENOMEM;
3652
3653 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003654 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003655 L1_CACHE_BYTES), GFP_KERNEL);
3656 if (!sbinfo)
3657 return -ENOMEM;
3658
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003659 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003660 sbinfo->uid = current_fsuid();
3661 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003662 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003664#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 /*
3666 * Per default we only allow half of the physical ram per
3667 * tmpfs instance, limiting inodes to one per page of lowmem;
3668 * but the internal instance is left unlimited.
3669 */
Al Viroca4e0512013-08-31 12:57:10 -04003670 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003671 sbinfo->max_blocks = shmem_default_max_blocks();
3672 sbinfo->max_inodes = shmem_default_max_inodes();
3673 if (shmem_parse_options(data, sbinfo, false)) {
3674 err = -EINVAL;
3675 goto failed;
3676 }
Al Viroca4e0512013-08-31 12:57:10 -04003677 } else {
3678 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 }
David M. Grimes91828a42006-10-17 00:09:45 -07003680 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003681 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682#else
3683 sb->s_flags |= MS_NOUSER;
3684#endif
3685
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003686 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003687 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003688 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003689 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003690 spin_lock_init(&sbinfo->shrinklist_lock);
3691 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003692
Hugh Dickins285b2c42011-08-03 16:21:20 -07003693 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003694 sb->s_blocksize = PAGE_SIZE;
3695 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 sb->s_magic = TMPFS_MAGIC;
3697 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003698 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003699#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003700 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003701#endif
3702#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003703 sb->s_flags |= MS_POSIXACL;
3704#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003705
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003706 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 if (!inode)
3708 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003709 inode->i_uid = sbinfo->uid;
3710 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003711 sb->s_root = d_make_root(inode);
3712 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003713 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714 return 0;
3715
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716failed:
3717 shmem_put_super(sb);
3718 return err;
3719}
3720
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003721static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722
3723static struct inode *shmem_alloc_inode(struct super_block *sb)
3724{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003725 struct shmem_inode_info *info;
3726 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3727 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003729 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730}
3731
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003732static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003733{
3734 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003735 if (S_ISLNK(inode->i_mode))
3736 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003737 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3738}
3739
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740static void shmem_destroy_inode(struct inode *inode)
3741{
Al Viro09208d12011-07-26 03:15:03 -04003742 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003744 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745}
3746
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003747static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003749 struct shmem_inode_info *info = foo;
3750 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751}
3752
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003753static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754{
3755 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3756 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003757 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 return 0;
3759}
3760
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003761static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003763 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764}
3765
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003766static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003768 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003770 .write_begin = shmem_write_begin,
3771 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003773#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003774 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003775#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003776 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777};
3778
Helge Deller15ad7cd2006-12-06 20:40:36 -08003779static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003781 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003783 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003784 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003785 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003786 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07003787 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003788 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003789 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790#endif
3791};
3792
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003793static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003794 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003795 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003796#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003797 .setxattr = generic_setxattr,
3798 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003799 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003800 .removexattr = generic_removexattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003801 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003802#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803};
3804
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003805static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806#ifdef CONFIG_TMPFS
3807 .create = shmem_create,
3808 .lookup = simple_lookup,
3809 .link = shmem_link,
3810 .unlink = shmem_unlink,
3811 .symlink = shmem_symlink,
3812 .mkdir = shmem_mkdir,
3813 .rmdir = shmem_rmdir,
3814 .mknod = shmem_mknod,
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003815 .rename2 = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003816 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003818#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003819 .setxattr = generic_setxattr,
3820 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003821 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003822 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003823#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003824#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003825 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003826 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003827#endif
3828};
3829
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003830static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003831#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003832 .setxattr = generic_setxattr,
3833 .getxattr = generic_getxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003834 .listxattr = shmem_listxattr,
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003835 .removexattr = generic_removexattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003836#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003837#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003838 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003839 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003840#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841};
3842
Hugh Dickins759b9772007-03-05 00:30:28 -08003843static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844 .alloc_inode = shmem_alloc_inode,
3845 .destroy_inode = shmem_destroy_inode,
3846#ifdef CONFIG_TMPFS
3847 .statfs = shmem_statfs,
3848 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003849 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850#endif
Al Viro1f895f72010-06-05 19:10:41 -04003851 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852 .drop_inode = generic_delete_inode,
3853 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003854#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3855 .nr_cached_objects = shmem_unused_huge_count,
3856 .free_cached_objects = shmem_unused_huge_scan,
3857#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858};
3859
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003860static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003861 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003862 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863#ifdef CONFIG_NUMA
3864 .set_policy = shmem_set_policy,
3865 .get_policy = shmem_get_policy,
3866#endif
3867};
3868
Al Viro3c26ff62010-07-25 11:46:36 +04003869static struct dentry *shmem_mount(struct file_system_type *fs_type,
3870 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871{
Al Viro3c26ff62010-07-25 11:46:36 +04003872 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873}
3874
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003875static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 .owner = THIS_MODULE,
3877 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003878 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003880 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003883int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884{
3885 int error;
3886
Rob Landley16203a72013-09-11 14:26:12 -07003887 /* If rootfs called this, don't re-init */
3888 if (shmem_inode_cachep)
3889 return 0;
3890
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003891 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 if (error)
3893 goto out3;
3894
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003895 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003897 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 goto out2;
3899 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003900
Al Viroca4e0512013-08-31 12:57:10 -04003901 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 if (IS_ERR(shm_mnt)) {
3903 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003904 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 goto out1;
3906 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003907
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003908#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003909 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3910 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3911 else
3912 shmem_huge = 0; /* just in case it was patched */
3913#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 return 0;
3915
3916out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003917 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003919 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920out3:
3921 shm_mnt = ERR_PTR(error);
3922 return error;
3923}
Matt Mackall853ac432009-01-06 14:40:20 -08003924
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003925#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003926static ssize_t shmem_enabled_show(struct kobject *kobj,
3927 struct kobj_attribute *attr, char *buf)
3928{
3929 int values[] = {
3930 SHMEM_HUGE_ALWAYS,
3931 SHMEM_HUGE_WITHIN_SIZE,
3932 SHMEM_HUGE_ADVISE,
3933 SHMEM_HUGE_NEVER,
3934 SHMEM_HUGE_DENY,
3935 SHMEM_HUGE_FORCE,
3936 };
3937 int i, count;
3938
3939 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3940 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3941
3942 count += sprintf(buf + count, fmt,
3943 shmem_format_huge(values[i]));
3944 }
3945 buf[count - 1] = '\n';
3946 return count;
3947}
3948
3949static ssize_t shmem_enabled_store(struct kobject *kobj,
3950 struct kobj_attribute *attr, const char *buf, size_t count)
3951{
3952 char tmp[16];
3953 int huge;
3954
3955 if (count + 1 > sizeof(tmp))
3956 return -EINVAL;
3957 memcpy(tmp, buf, count);
3958 tmp[count] = '\0';
3959 if (count && tmp[count - 1] == '\n')
3960 tmp[count - 1] = '\0';
3961
3962 huge = shmem_parse_huge(tmp);
3963 if (huge == -EINVAL)
3964 return -EINVAL;
3965 if (!has_transparent_hugepage() &&
3966 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3967 return -EINVAL;
3968
3969 shmem_huge = huge;
3970 if (shmem_huge < SHMEM_HUGE_DENY)
3971 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3972 return count;
3973}
3974
3975struct kobj_attribute shmem_enabled_attr =
3976 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003977
3978bool shmem_huge_enabled(struct vm_area_struct *vma)
3979{
3980 struct inode *inode = file_inode(vma->vm_file);
3981 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3982 loff_t i_size;
3983 pgoff_t off;
3984
3985 if (shmem_huge == SHMEM_HUGE_FORCE)
3986 return true;
3987 if (shmem_huge == SHMEM_HUGE_DENY)
3988 return false;
3989 switch (sbinfo->huge) {
3990 case SHMEM_HUGE_NEVER:
3991 return false;
3992 case SHMEM_HUGE_ALWAYS:
3993 return true;
3994 case SHMEM_HUGE_WITHIN_SIZE:
3995 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3996 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3997 if (i_size >= HPAGE_PMD_SIZE &&
3998 i_size >> PAGE_SHIFT >= off)
3999 return true;
4000 case SHMEM_HUGE_ADVISE:
4001 /* TODO: implement fadvise() hints */
4002 return (vma->vm_flags & VM_HUGEPAGE);
4003 default:
4004 VM_BUG_ON(1);
4005 return false;
4006 }
4007}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004008#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004009
Matt Mackall853ac432009-01-06 14:40:20 -08004010#else /* !CONFIG_SHMEM */
4011
4012/*
4013 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4014 *
4015 * This is intended for small system where the benefits of the full
4016 * shmem code (swap-backed and resource-limited) are outweighed by
4017 * their complexity. On systems without swap this code should be
4018 * effectively equivalent, but much lighter weight.
4019 */
4020
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004021static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004022 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04004023 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08004024 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004025 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004026};
4027
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004028int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004029{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004030 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004031
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004032 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004033 BUG_ON(IS_ERR(shm_mnt));
4034
4035 return 0;
4036}
4037
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004038int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08004039{
4040 return 0;
4041}
4042
Hugh Dickins3f96b792009-09-21 17:03:37 -07004043int shmem_lock(struct file *file, int lock, struct user_struct *user)
4044{
4045 return 0;
4046}
4047
Hugh Dickins24513262012-01-20 14:34:21 -08004048void shmem_unlock_mapping(struct address_space *mapping)
4049{
4050}
4051
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004052#ifdef CONFIG_MMU
4053unsigned long shmem_get_unmapped_area(struct file *file,
4054 unsigned long addr, unsigned long len,
4055 unsigned long pgoff, unsigned long flags)
4056{
4057 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4058}
4059#endif
4060
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004061void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004062{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004063 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004064}
4065EXPORT_SYMBOL_GPL(shmem_truncate_range);
4066
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004067#define shmem_vm_ops generic_file_vm_ops
4068#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004069#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004070#define shmem_acct_size(flags, size) 0
4071#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004072
4073#endif /* CONFIG_SHMEM */
4074
4075/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076
Al Viro34515382013-02-14 22:38:02 -05004077static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04004078 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05004079};
4080
Eric Parisc7277092013-12-02 11:24:19 +00004081static struct file *__shmem_file_setup(const char *name, loff_t size,
4082 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083{
Al Viro6b4d0b22013-02-14 21:37:26 -05004084 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004085 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04004086 struct path path;
Al Viro34515382013-02-14 22:38:02 -05004087 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 struct qstr this;
4089
4090 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05004091 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092
Hugh Dickins285b2c42011-08-03 16:21:20 -07004093 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 return ERR_PTR(-EINVAL);
4095
4096 if (shmem_acct_size(flags, size))
4097 return ERR_PTR(-ENOMEM);
4098
Al Viro6b4d0b22013-02-14 21:37:26 -05004099 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100 this.name = name;
4101 this.len = strlen(name);
4102 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004103 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004104 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004105 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004106 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004108 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109
Al Viro6b4d0b22013-02-14 21:37:26 -05004110 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004111 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004113 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114
Eric Parisc7277092013-12-02 11:24:19 +00004115 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004116 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004118 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004119 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4120 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004121 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004122
Al Viro6b4d0b22013-02-14 21:37:26 -05004123 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004124 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004125 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004126 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004127
Al Viro6b4d0b22013-02-14 21:37:26 -05004128 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130put_memory:
4131 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004132put_path:
4133 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004134 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135}
Eric Parisc7277092013-12-02 11:24:19 +00004136
4137/**
4138 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4139 * kernel internal. There will be NO LSM permission checks against the
4140 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004141 * higher layer. The users are the big_key and shm implementations. LSM
4142 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004143 * @name: name for dentry (to be seen in /proc/<pid>/maps
4144 * @size: size to be set for the file
4145 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4146 */
4147struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4148{
4149 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4150}
4151
4152/**
4153 * shmem_file_setup - get an unlinked file living in tmpfs
4154 * @name: name for dentry (to be seen in /proc/<pid>/maps
4155 * @size: size to be set for the file
4156 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4157 */
4158struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4159{
4160 return __shmem_file_setup(name, size, flags, 0);
4161}
Keith Packard395e0dd2008-06-20 00:08:06 -07004162EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163
Randy Dunlap46711812008-03-19 17:00:41 -07004164/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004165 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4167 */
4168int shmem_zero_setup(struct vm_area_struct *vma)
4169{
4170 struct file *file;
4171 loff_t size = vma->vm_end - vma->vm_start;
4172
Hugh Dickins66fc1302015-06-14 09:48:09 -07004173 /*
4174 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4175 * between XFS directory reading and selinux: since this file is only
4176 * accessible to the user through its mapping, use S_PRIVATE flag to
4177 * bypass file security, in the same way as shmem_kernel_file_setup().
4178 */
4179 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180 if (IS_ERR(file))
4181 return PTR_ERR(file);
4182
4183 if (vma->vm_file)
4184 fput(vma->vm_file);
4185 vma->vm_file = file;
4186 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004187
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004188 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004189 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4190 (vma->vm_end & HPAGE_PMD_MASK)) {
4191 khugepaged_enter(vma, vma->vm_flags);
4192 }
4193
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194 return 0;
4195}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004196
4197/**
4198 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4199 * @mapping: the page's address_space
4200 * @index: the page index
4201 * @gfp: the page allocator flags to use if allocating
4202 *
4203 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4204 * with any new page allocations done using the specified allocation flags.
4205 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4206 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4207 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4208 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004209 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4210 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004211 */
4212struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4213 pgoff_t index, gfp_t gfp)
4214{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004215#ifdef CONFIG_SHMEM
4216 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004217 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004218 int error;
4219
4220 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004221 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4222 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004223 if (error)
4224 page = ERR_PTR(error);
4225 else
4226 unlock_page(page);
4227 return page;
4228#else
4229 /*
4230 * The tiny !SHMEM case uses ramfs without swap
4231 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004232 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004233#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004234}
4235EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);