blob: ad7813d73ea79879008eb570e790a101a4d7c3f7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins6922c0c2011-08-03 16:21:25 -07009 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -070011 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 *
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 *
Matt Mackall853ac432009-01-06 14:40:20 -080018 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * This file is released under the GPL.
22 */
23
Matt Mackall853ac432009-01-06 14:40:20 -080024#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
Andrew Morton250297e2013-04-29 15:06:12 -070028#include <linux/ramfs.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070029#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080030#include <linux/file.h>
31#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040032#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080033#include <linux/swap.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080034#include <linux/uio.h>
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -070035#include <linux/khugepaged.h>
Matt Mackall853ac432009-01-06 14:40:20 -080036
37static struct vfsmount *shm_mnt;
38
39#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
44 */
45
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070046#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070047#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010048#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080049#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/string.h>
52#include <linux/slab.h>
53#include <linux/backing-dev.h>
54#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070057#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070058#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070059#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070060#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <linux/security.h>
62#include <linux/swapops.h>
63#include <linux/mempolicy.h>
64#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000065#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070066#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070067#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080068#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040069#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070070#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070071#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070072#include <uapi/linux/memfd.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070073
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <asm/pgtable.h>
76
Mel Gormandd56b042015-11-06 16:28:43 -080077#include "internal.h"
78
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030079#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Linus Torvalds1da177e2005-04-16 15:20:36 -070082/* Pretend that each entry is of this size in directory's i_size */
83#define BOGO_DIRENT_SIZE 20
84
Hugh Dickins69f07ec2011-08-03 16:21:26 -070085/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86#define SHORT_SYMLINK_LEN 128
87
Hugh Dickins1aac1402012-05-29 15:06:42 -070088/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070089 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070092 */
93struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -070094 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -070095 pgoff_t start; /* start of range currently being fallocated */
96 pgoff_t next; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
99};
100
Andrew Mortonb76db732008-02-08 04:21:49 -0800101#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800102static unsigned long shmem_default_max_blocks(void)
103{
104 return totalram_pages / 2;
105}
106
107static unsigned long shmem_default_max_inodes(void)
108{
109 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
110}
Andrew Mortonb76db732008-02-08 04:21:49 -0800111#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800112
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700113static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700116static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700117 struct page **pagep, enum sgp_type sgp,
118 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700119
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700120int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700121 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700122{
123 return shmem_getpage_gfp(inode, index, pagep, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700124 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700125}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
128{
129 return sb->s_fs_info;
130}
131
132/*
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
137 */
138static inline int shmem_acct_size(unsigned long flags, loff_t size)
139{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000140 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000141 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
144static inline void shmem_unacct_size(unsigned long flags, loff_t size)
145{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000146 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 vm_unacct_memory(VM_ACCT(size));
148}
149
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700150static inline int shmem_reacct_size(unsigned long flags,
151 loff_t oldsize, loff_t newsize)
152{
153 if (!(flags & VM_NORESERVE)) {
154 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
155 return security_vm_enough_memory_mm(current->mm,
156 VM_ACCT(newsize) - VM_ACCT(oldsize));
157 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
158 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
159 }
160 return 0;
161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/*
164 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700165 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
168 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700169static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700171 if (!(flags & VM_NORESERVE))
172 return 0;
173
174 return security_vm_enough_memory_mm(current->mm,
175 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
178static inline void shmem_unacct_blocks(unsigned long flags, long pages)
179{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000180 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300181 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
Hugh Dickins759b9772007-03-05 00:30:28 -0800184static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700185static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800186static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800187static const struct inode_operations shmem_inode_operations;
188static const struct inode_operations shmem_dir_inode_operations;
189static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400190static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700191static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800194static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800196static int shmem_reserve_inode(struct super_block *sb)
197{
198 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
199 if (sbinfo->max_inodes) {
200 spin_lock(&sbinfo->stat_lock);
201 if (!sbinfo->free_inodes) {
202 spin_unlock(&sbinfo->stat_lock);
203 return -ENOSPC;
204 }
205 sbinfo->free_inodes--;
206 spin_unlock(&sbinfo->stat_lock);
207 }
208 return 0;
209}
210
211static void shmem_free_inode(struct super_block *sb)
212{
213 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
214 if (sbinfo->max_inodes) {
215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_inodes++;
217 spin_unlock(&sbinfo->stat_lock);
218 }
219}
220
Randy Dunlap46711812008-03-19 17:00:41 -0700221/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700222 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 * @inode: inode to recalc
224 *
225 * We have to calculate the free blocks since the mm can drop
226 * undirtied hole pages behind our back.
227 *
228 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
229 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
230 *
231 * It has to be called with the spinlock held.
232 */
233static void shmem_recalc_inode(struct inode *inode)
234{
235 struct shmem_inode_info *info = SHMEM_I(inode);
236 long freed;
237
238 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
239 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700240 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
241 if (sbinfo->max_blocks)
242 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700244 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247}
248
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700249bool shmem_charge(struct inode *inode, long pages)
250{
251 struct shmem_inode_info *info = SHMEM_I(inode);
252 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700253 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700254
255 if (shmem_acct_block(info->flags, pages))
256 return false;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700257 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700258 info->alloced += pages;
259 inode->i_blocks += pages * BLOCKS_PER_PAGE;
260 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700261 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700262 inode->i_mapping->nrpages += pages;
263
264 if (!sbinfo->max_blocks)
265 return true;
266 if (percpu_counter_compare(&sbinfo->used_blocks,
267 sbinfo->max_blocks - pages) > 0) {
268 inode->i_mapping->nrpages -= pages;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700269 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700270 info->alloced -= pages;
271 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700272 spin_unlock_irqrestore(&info->lock, flags);
Hugh Dickins71664662016-09-23 20:24:23 -0700273 shmem_unacct_blocks(info->flags, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700274 return false;
275 }
276 percpu_counter_add(&sbinfo->used_blocks, pages);
277 return true;
278}
279
280void shmem_uncharge(struct inode *inode, long pages)
281{
282 struct shmem_inode_info *info = SHMEM_I(inode);
283 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700284 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700285
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700286 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700287 info->alloced -= pages;
288 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
289 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700290 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700291
292 if (sbinfo->max_blocks)
293 percpu_counter_sub(&sbinfo->used_blocks, pages);
Hugh Dickins71664662016-09-23 20:24:23 -0700294 shmem_unacct_blocks(info->flags, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700295}
296
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700297/*
298 * Replace item expected in radix tree by a new item, while holding tree lock.
299 */
300static int shmem_radix_tree_replace(struct address_space *mapping,
301 pgoff_t index, void *expected, void *replacement)
302{
303 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700304 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700305
306 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700307 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700308 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700309 if (!pslot)
310 return -ENOENT;
311 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700312 if (item != expected)
313 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700314 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700315 return 0;
316}
317
318/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700319 * Sometimes, before we decide whether to proceed or to fail, we must check
320 * that an entry was not already brought back from swap by a racing thread.
321 *
322 * Checking page is not enough: by the time a SwapCache page is locked, it
323 * might be reused, and again be SwapCache, using the same swap as before.
324 */
325static bool shmem_confirm_swap(struct address_space *mapping,
326 pgoff_t index, swp_entry_t swap)
327{
328 void *item;
329
330 rcu_read_lock();
331 item = radix_tree_lookup(&mapping->page_tree, index);
332 rcu_read_unlock();
333 return item == swp_to_radix_entry(swap);
334}
335
336/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700337 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
338 *
339 * SHMEM_HUGE_NEVER:
340 * disables huge pages for the mount;
341 * SHMEM_HUGE_ALWAYS:
342 * enables huge pages for the mount;
343 * SHMEM_HUGE_WITHIN_SIZE:
344 * only allocate huge pages if the page will be fully within i_size,
345 * also respect fadvise()/madvise() hints;
346 * SHMEM_HUGE_ADVISE:
347 * only allocate huge pages if requested with fadvise()/madvise();
348 */
349
350#define SHMEM_HUGE_NEVER 0
351#define SHMEM_HUGE_ALWAYS 1
352#define SHMEM_HUGE_WITHIN_SIZE 2
353#define SHMEM_HUGE_ADVISE 3
354
355/*
356 * Special values.
357 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
358 *
359 * SHMEM_HUGE_DENY:
360 * disables huge on shm_mnt and all mounts, for emergency use;
361 * SHMEM_HUGE_FORCE:
362 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
363 *
364 */
365#define SHMEM_HUGE_DENY (-1)
366#define SHMEM_HUGE_FORCE (-2)
367
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700368#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700369/* ifdef here to avoid bloating shmem.o when not necessary */
370
371int shmem_huge __read_mostly;
372
373static int shmem_parse_huge(const char *str)
374{
375 if (!strcmp(str, "never"))
376 return SHMEM_HUGE_NEVER;
377 if (!strcmp(str, "always"))
378 return SHMEM_HUGE_ALWAYS;
379 if (!strcmp(str, "within_size"))
380 return SHMEM_HUGE_WITHIN_SIZE;
381 if (!strcmp(str, "advise"))
382 return SHMEM_HUGE_ADVISE;
383 if (!strcmp(str, "deny"))
384 return SHMEM_HUGE_DENY;
385 if (!strcmp(str, "force"))
386 return SHMEM_HUGE_FORCE;
387 return -EINVAL;
388}
389
390static const char *shmem_format_huge(int huge)
391{
392 switch (huge) {
393 case SHMEM_HUGE_NEVER:
394 return "never";
395 case SHMEM_HUGE_ALWAYS:
396 return "always";
397 case SHMEM_HUGE_WITHIN_SIZE:
398 return "within_size";
399 case SHMEM_HUGE_ADVISE:
400 return "advise";
401 case SHMEM_HUGE_DENY:
402 return "deny";
403 case SHMEM_HUGE_FORCE:
404 return "force";
405 default:
406 VM_BUG_ON(1);
407 return "bad_val";
408 }
409}
410
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700411static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
412 struct shrink_control *sc, unsigned long nr_to_split)
413{
414 LIST_HEAD(list), *pos, *next;
415 struct inode *inode;
416 struct shmem_inode_info *info;
417 struct page *page;
418 unsigned long batch = sc ? sc->nr_to_scan : 128;
419 int removed = 0, split = 0;
420
421 if (list_empty(&sbinfo->shrinklist))
422 return SHRINK_STOP;
423
424 spin_lock(&sbinfo->shrinklist_lock);
425 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
426 info = list_entry(pos, struct shmem_inode_info, shrinklist);
427
428 /* pin the inode */
429 inode = igrab(&info->vfs_inode);
430
431 /* inode is about to be evicted */
432 if (!inode) {
433 list_del_init(&info->shrinklist);
434 removed++;
435 goto next;
436 }
437
438 /* Check if there's anything to gain */
439 if (round_up(inode->i_size, PAGE_SIZE) ==
440 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
441 list_del_init(&info->shrinklist);
442 removed++;
443 iput(inode);
444 goto next;
445 }
446
447 list_move(&info->shrinklist, &list);
448next:
449 if (!--batch)
450 break;
451 }
452 spin_unlock(&sbinfo->shrinklist_lock);
453
454 list_for_each_safe(pos, next, &list) {
455 int ret;
456
457 info = list_entry(pos, struct shmem_inode_info, shrinklist);
458 inode = &info->vfs_inode;
459
460 if (nr_to_split && split >= nr_to_split) {
461 iput(inode);
462 continue;
463 }
464
465 page = find_lock_page(inode->i_mapping,
466 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
467 if (!page)
468 goto drop;
469
470 if (!PageTransHuge(page)) {
471 unlock_page(page);
472 put_page(page);
473 goto drop;
474 }
475
476 ret = split_huge_page(page);
477 unlock_page(page);
478 put_page(page);
479
480 if (ret) {
481 /* split failed: leave it on the list */
482 iput(inode);
483 continue;
484 }
485
486 split++;
487drop:
488 list_del_init(&info->shrinklist);
489 removed++;
490 iput(inode);
491 }
492
493 spin_lock(&sbinfo->shrinklist_lock);
494 list_splice_tail(&list, &sbinfo->shrinklist);
495 sbinfo->shrinklist_len -= removed;
496 spin_unlock(&sbinfo->shrinklist_lock);
497
498 return split;
499}
500
501static long shmem_unused_huge_scan(struct super_block *sb,
502 struct shrink_control *sc)
503{
504 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
505
506 if (!READ_ONCE(sbinfo->shrinklist_len))
507 return SHRINK_STOP;
508
509 return shmem_unused_huge_shrink(sbinfo, sc, 0);
510}
511
512static long shmem_unused_huge_count(struct super_block *sb,
513 struct shrink_control *sc)
514{
515 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
516 return READ_ONCE(sbinfo->shrinklist_len);
517}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700518#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700519
520#define shmem_huge SHMEM_HUGE_DENY
521
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700522static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
523 struct shrink_control *sc, unsigned long nr_to_split)
524{
525 return 0;
526}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700527#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700528
529/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700530 * Like add_to_page_cache_locked, but error if expected item has gone.
531 */
532static int shmem_add_to_page_cache(struct page *page,
533 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700534 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700535{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700536 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700537
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700538 VM_BUG_ON_PAGE(PageTail(page), page);
539 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800540 VM_BUG_ON_PAGE(!PageLocked(page), page);
541 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700542 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700543
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700544 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700545 page->mapping = mapping;
546 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700547
Hugh Dickinsb065b432012-07-11 14:02:48 -0700548 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700549 if (PageTransHuge(page)) {
550 void __rcu **results;
551 pgoff_t idx;
552 int i;
553
554 error = 0;
555 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
556 &results, &idx, index, 1) &&
557 idx < index + HPAGE_PMD_NR) {
558 error = -EEXIST;
559 }
560
561 if (!error) {
562 for (i = 0; i < HPAGE_PMD_NR; i++) {
563 error = radix_tree_insert(&mapping->page_tree,
564 index + i, page + i);
565 VM_BUG_ON(error);
566 }
567 count_vm_event(THP_FILE_ALLOC);
568 }
569 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700570 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700571 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700572 error = shmem_radix_tree_replace(mapping, index, expected,
573 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700574 }
575
Hugh Dickinsb065b432012-07-11 14:02:48 -0700576 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700577 mapping->nrpages += nr;
578 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700579 __inc_node_page_state(page, NR_SHMEM_THPS);
580 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
581 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700582 spin_unlock_irq(&mapping->tree_lock);
583 } else {
584 page->mapping = NULL;
585 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700586 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700587 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700588 return error;
589}
590
591/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700592 * Like delete_from_page_cache, but substitutes swap for page.
593 */
594static void shmem_delete_from_page_cache(struct page *page, void *radswap)
595{
596 struct address_space *mapping = page->mapping;
597 int error;
598
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700599 VM_BUG_ON_PAGE(PageCompound(page), page);
600
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700601 spin_lock_irq(&mapping->tree_lock);
602 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
603 page->mapping = NULL;
604 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700605 __dec_node_page_state(page, NR_FILE_PAGES);
606 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700607 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300608 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700609 BUG_ON(error);
610}
611
612/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700613 * Remove swap entry from radix tree, free the swap and its page cache.
614 */
615static int shmem_free_swap(struct address_space *mapping,
616 pgoff_t index, void *radswap)
617{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700618 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700619
620 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700621 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700622 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700623 if (old != radswap)
624 return -ENOENT;
625 free_swap_and_cache(radix_to_swp_entry(radswap));
626 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700627}
628
629/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800630 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800631 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800632 *
633 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
634 * as long as the inode doesn't go away and racy results are not a problem.
635 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800636unsigned long shmem_partial_swap_usage(struct address_space *mapping,
637 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800638{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800639 struct radix_tree_iter iter;
640 void **slot;
641 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800642 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800643
644 rcu_read_lock();
645
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800646 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
647 if (iter.index >= end)
648 break;
649
650 page = radix_tree_deref_slot(slot);
651
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700652 if (radix_tree_deref_retry(page)) {
653 slot = radix_tree_iter_retry(&iter);
654 continue;
655 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800656
657 if (radix_tree_exceptional_entry(page))
658 swapped++;
659
660 if (need_resched()) {
661 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700662 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800663 }
664 }
665
666 rcu_read_unlock();
667
668 return swapped << PAGE_SHIFT;
669}
670
671/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800672 * Determine (in bytes) how many of the shmem object's pages mapped by the
673 * given vma is swapped out.
674 *
675 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
676 * as long as the inode doesn't go away and racy results are not a problem.
677 */
678unsigned long shmem_swap_usage(struct vm_area_struct *vma)
679{
680 struct inode *inode = file_inode(vma->vm_file);
681 struct shmem_inode_info *info = SHMEM_I(inode);
682 struct address_space *mapping = inode->i_mapping;
683 unsigned long swapped;
684
685 /* Be careful as we don't hold info->lock */
686 swapped = READ_ONCE(info->swapped);
687
688 /*
689 * The easier cases are when the shmem object has nothing in swap, or
690 * the vma maps it whole. Then we can simply use the stats that we
691 * already track.
692 */
693 if (!swapped)
694 return 0;
695
696 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
697 return swapped << PAGE_SHIFT;
698
699 /* Here comes the more involved part */
700 return shmem_partial_swap_usage(mapping,
701 linear_page_index(vma, vma->vm_start),
702 linear_page_index(vma, vma->vm_end));
703}
704
705/*
Hugh Dickins24513262012-01-20 14:34:21 -0800706 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
707 */
708void shmem_unlock_mapping(struct address_space *mapping)
709{
710 struct pagevec pvec;
711 pgoff_t indices[PAGEVEC_SIZE];
712 pgoff_t index = 0;
713
714 pagevec_init(&pvec, 0);
715 /*
716 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
717 */
718 while (!mapping_unevictable(mapping)) {
719 /*
720 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
721 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
722 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700723 pvec.nr = find_get_entries(mapping, index,
724 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800725 if (!pvec.nr)
726 break;
727 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700728 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800729 check_move_unevictable_pages(pvec.pages, pvec.nr);
730 pagevec_release(&pvec);
731 cond_resched();
732 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700733}
734
735/*
736 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700737 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700738 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700739static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
740 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700742 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300744 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
745 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
746 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
747 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700748 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700749 pgoff_t indices[PAGEVEC_SIZE];
750 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700751 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700752 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700754 if (lend == -1)
755 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700756
757 pagevec_init(&pvec, 0);
758 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700759 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700760 pvec.nr = find_get_entries(mapping, index,
761 min(end - index, (pgoff_t)PAGEVEC_SIZE),
762 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700763 if (!pvec.nr)
764 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700765 for (i = 0; i < pagevec_count(&pvec); i++) {
766 struct page *page = pvec.pages[i];
767
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700768 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700769 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700770 break;
771
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700772 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700773 if (unfalloc)
774 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700775 nr_swaps_freed += !shmem_free_swap(mapping,
776 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700777 continue;
778 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700779
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700780 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
781
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700782 if (!trylock_page(page))
783 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700784
785 if (PageTransTail(page)) {
786 /* Middle of THP: zero out the page */
787 clear_highpage(page);
788 unlock_page(page);
789 continue;
790 } else if (PageTransHuge(page)) {
791 if (index == round_down(end, HPAGE_PMD_NR)) {
792 /*
793 * Range ends in the middle of THP:
794 * zero out the page
795 */
796 clear_highpage(page);
797 unlock_page(page);
798 continue;
799 }
800 index += HPAGE_PMD_NR - 1;
801 i += HPAGE_PMD_NR - 1;
802 }
803
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700804 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700805 VM_BUG_ON_PAGE(PageTail(page), page);
806 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800807 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700808 truncate_inode_page(mapping, page);
809 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700810 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700811 unlock_page(page);
812 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700813 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800814 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700815 cond_resched();
816 index++;
817 }
818
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700819 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700820 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700821 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700822 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300823 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700824 if (start > end) {
825 top = partial_end;
826 partial_end = 0;
827 }
828 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700829 set_page_dirty(page);
830 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300831 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700832 }
833 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700834 if (partial_end) {
835 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700836 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700837 if (page) {
838 zero_user_segment(page, 0, partial_end);
839 set_page_dirty(page);
840 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300841 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700842 }
843 }
844 if (start >= end)
845 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700846
847 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700848 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700849 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700850
851 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700852 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700853 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700854 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700855 /* If all gone or hole-punch or unfalloc, we're done */
856 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700857 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700858 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700859 index = start;
860 continue;
861 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700862 for (i = 0; i < pagevec_count(&pvec); i++) {
863 struct page *page = pvec.pages[i];
864
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700865 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700866 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700867 break;
868
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700869 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700870 if (unfalloc)
871 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700872 if (shmem_free_swap(mapping, index, page)) {
873 /* Swap was replaced by page: retry */
874 index--;
875 break;
876 }
877 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700878 continue;
879 }
880
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700881 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700882
883 if (PageTransTail(page)) {
884 /* Middle of THP: zero out the page */
885 clear_highpage(page);
886 unlock_page(page);
887 /*
888 * Partial thp truncate due 'start' in middle
889 * of THP: don't need to look on these pages
890 * again on !pvec.nr restart.
891 */
892 if (index != round_down(end, HPAGE_PMD_NR))
893 start++;
894 continue;
895 } else if (PageTransHuge(page)) {
896 if (index == round_down(end, HPAGE_PMD_NR)) {
897 /*
898 * Range ends in the middle of THP:
899 * zero out the page
900 */
901 clear_highpage(page);
902 unlock_page(page);
903 continue;
904 }
905 index += HPAGE_PMD_NR - 1;
906 i += HPAGE_PMD_NR - 1;
907 }
908
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700909 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700910 VM_BUG_ON_PAGE(PageTail(page), page);
911 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800912 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700913 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700914 } else {
915 /* Page was replaced by swap: retry */
916 unlock_page(page);
917 index--;
918 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700919 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700920 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700921 unlock_page(page);
922 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700923 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800924 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700925 index++;
926 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700927
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700928 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700929 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700931 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700932}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700934void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
935{
936 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700937 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700939EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Yu Zhao44a30222015-09-08 15:03:33 -0700941static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
942 struct kstat *stat)
943{
944 struct inode *inode = dentry->d_inode;
945 struct shmem_inode_info *info = SHMEM_I(inode);
946
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800947 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700948 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800949 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700950 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800951 }
Yu Zhao44a30222015-09-08 15:03:33 -0700952 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700953 return 0;
954}
955
Hugh Dickins94c1e622011-06-27 16:18:03 -0700956static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957{
David Howells75c3cfa2015-03-17 22:26:12 +0000958 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700959 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700960 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 int error;
962
Jan Kara31051c82016-05-26 16:55:18 +0200963 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200964 if (error)
965 return error;
966
Hugh Dickins94c1e622011-06-27 16:18:03 -0700967 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
968 loff_t oldsize = inode->i_size;
969 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000970
David Herrmann40e041a2014-08-08 14:25:27 -0700971 /* protected by i_mutex */
972 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
973 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
974 return -EPERM;
975
Hugh Dickins94c1e622011-06-27 16:18:03 -0700976 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700977 error = shmem_reacct_size(SHMEM_I(inode)->flags,
978 oldsize, newsize);
979 if (error)
980 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -0700981 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700982 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700983 }
Josef Bacikafa2db22015-06-24 16:58:45 -0700984 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -0700985 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800986 if (oldsize > holebegin)
987 unmap_mapping_range(inode->i_mapping,
988 holebegin, 0, 1);
989 if (info->alloced)
990 shmem_truncate_range(inode,
991 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700992 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800993 if (oldsize > holebegin)
994 unmap_mapping_range(inode->i_mapping,
995 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700996
997 /*
998 * Part of the huge page can be beyond i_size: subject
999 * to shrink under memory pressure.
1000 */
1001 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1002 spin_lock(&sbinfo->shrinklist_lock);
1003 if (list_empty(&info->shrinklist)) {
1004 list_add_tail(&info->shrinklist,
1005 &sbinfo->shrinklist);
1006 sbinfo->shrinklist_len++;
1007 }
1008 spin_unlock(&sbinfo->shrinklist_lock);
1009 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001010 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
1012
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001013 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001014 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001015 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 return error;
1017}
1018
Al Viro1f895f72010-06-05 19:10:41 -04001019static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001022 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001024 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 shmem_unacct_size(info->flags, inode->i_size);
1026 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001027 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001028 if (!list_empty(&info->shrinklist)) {
1029 spin_lock(&sbinfo->shrinklist_lock);
1030 if (!list_empty(&info->shrinklist)) {
1031 list_del_init(&info->shrinklist);
1032 sbinfo->shrinklist_len--;
1033 }
1034 spin_unlock(&sbinfo->shrinklist_lock);
1035 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001037 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001039 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
Al Viro3ed47db2016-01-22 18:08:52 -05001041 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001042
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001043 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001044 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001045 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001046 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047}
1048
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001049/*
1050 * If swap found in inode, free it and move page from swapcache to filecache.
1051 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001052static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001053 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001055 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001056 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001057 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001058 gfp_t gfp;
1059 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001061 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001062 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001063 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001064 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001065
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001066 /*
1067 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001068 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001069 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001070 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001071 */
1072 if (shmem_swaplist.next != &info->swaplist)
1073 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001074
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001075 gfp = mapping_gfp_mask(mapping);
1076 if (shmem_should_replace_page(*pagep, gfp)) {
1077 mutex_unlock(&shmem_swaplist_mutex);
1078 error = shmem_replace_page(pagep, gfp, info, index);
1079 mutex_lock(&shmem_swaplist_mutex);
1080 /*
1081 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001082 * allocation, but the inode might have been freed while we
1083 * dropped it: although a racing shmem_evict_inode() cannot
1084 * complete without emptying the radix_tree, our page lock
1085 * on this swapcache page is not enough to prevent that -
1086 * free_swap_and_cache() of our swap entry will only
1087 * trylock_page(), removing swap from radix_tree whatever.
1088 *
1089 * We must not proceed to shmem_add_to_page_cache() if the
1090 * inode has been freed, but of course we cannot rely on
1091 * inode or mapping or info to check that. However, we can
1092 * safely check if our swap entry is still in use (and here
1093 * it can't have got reused for another page): if it's still
1094 * in use, then the inode cannot have been freed yet, and we
1095 * can safely proceed (if it's no longer in use, that tells
1096 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001097 */
1098 if (!page_swapcount(*pagep))
1099 error = -ENOENT;
1100 }
1101
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001102 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001103 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1104 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1105 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001106 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001107 if (!error)
1108 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001109 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001110 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001111 /*
1112 * Truncation and eviction use free_swap_and_cache(), which
1113 * only does trylock page: if we raced, best clean up here.
1114 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001115 delete_from_swap_cache(*pagep);
1116 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001117 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001118 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001119 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001120 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001121 swap_free(swap);
1122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001124 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
1127/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001128 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001130int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001132 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001134 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001135 int error = 0;
1136
1137 /*
1138 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001139 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001140 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001141 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001142 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001143
1144 /*
1145 * Charge page using GFP_KERNEL while we can wait, before taking
1146 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1147 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001148 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001149 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1150 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001151 if (error)
1152 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001153 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001154 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001156 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001157 list_for_each_safe(this, next, &shmem_swaplist) {
1158 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001159 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001160 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001161 else
1162 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001163 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001164 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001165 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001166 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001168 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001169
Johannes Weiner00501b52014-08-08 14:19:20 -07001170 if (error) {
1171 if (error != -ENOMEM)
1172 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001173 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001174 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001175 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001176out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001177 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001178 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001179 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182/*
1183 * Move the page from the page cache to the swap cache.
1184 */
1185static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1186{
1187 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001190 swp_entry_t swap;
1191 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001193 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 mapping = page->mapping;
1196 index = page->index;
1197 inode = mapping->host;
1198 info = SHMEM_I(inode);
1199 if (info->flags & VM_LOCKED)
1200 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001201 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 goto redirty;
1203
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001204 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001205 * Our capabilities prevent regular writeback or sync from ever calling
1206 * shmem_writepage; but a stacking filesystem might use ->writepage of
1207 * its underlying filesystem, in which case tmpfs should write out to
1208 * swap only in response to memory pressure, and not for the writeback
1209 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001210 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001211 if (!wbc->for_reclaim) {
1212 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1213 goto redirty;
1214 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001215
1216 /*
1217 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1218 * value into swapfile.c, the only way we can correctly account for a
1219 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001220 *
1221 * That's okay for a page already fallocated earlier, but if we have
1222 * not yet completed the fallocation, then (a) we want to keep track
1223 * of this page in case we have to undo it, and (b) it may not be a
1224 * good idea to continue anyway, once we're pushing into swap. So
1225 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001226 */
1227 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001228 if (inode->i_private) {
1229 struct shmem_falloc *shmem_falloc;
1230 spin_lock(&inode->i_lock);
1231 shmem_falloc = inode->i_private;
1232 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001233 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001234 index >= shmem_falloc->start &&
1235 index < shmem_falloc->next)
1236 shmem_falloc->nr_unswapped++;
1237 else
1238 shmem_falloc = NULL;
1239 spin_unlock(&inode->i_lock);
1240 if (shmem_falloc)
1241 goto redirty;
1242 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001243 clear_highpage(page);
1244 flush_dcache_page(page);
1245 SetPageUptodate(page);
1246 }
1247
Hugh Dickins48f170f2011-07-25 17:12:37 -07001248 swap = get_swap_page();
1249 if (!swap.val)
1250 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001251
Vladimir Davydov37e84352016-01-20 15:02:56 -08001252 if (mem_cgroup_try_charge_swap(page, swap))
1253 goto free_swap;
1254
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001255 /*
1256 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001257 * if it's not already there. Do it now before the page is
1258 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001259 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001260 * we've incremented swapped, because shmem_unuse_inode() will
1261 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001262 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001263 mutex_lock(&shmem_swaplist_mutex);
1264 if (list_empty(&info->swaplist))
1265 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001266
Hugh Dickins48f170f2011-07-25 17:12:37 -07001267 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001268 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001269 shmem_recalc_inode(inode);
1270 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001271 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001272
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001273 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001274 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1275
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001276 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001277 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001278 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 return 0;
1280 }
1281
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001282 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001283free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001284 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285redirty:
1286 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001287 if (wbc->for_reclaim)
1288 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1289 unlock_page(page);
1290 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291}
1292
Hugh Dickins75edd342016-05-19 17:12:44 -07001293#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001294static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001295{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001296 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001297
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001298 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001299 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001300
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001301 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001302
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001303 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001304}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001305
1306static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1307{
1308 struct mempolicy *mpol = NULL;
1309 if (sbinfo->mpol) {
1310 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1311 mpol = sbinfo->mpol;
1312 mpol_get(mpol);
1313 spin_unlock(&sbinfo->stat_lock);
1314 }
1315 return mpol;
1316}
Hugh Dickins75edd342016-05-19 17:12:44 -07001317#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1318static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1319{
1320}
1321static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1322{
1323 return NULL;
1324}
1325#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1326#ifndef CONFIG_NUMA
1327#define vm_policy vm_private_data
1328#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001329
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001330static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1331 struct shmem_inode_info *info, pgoff_t index)
1332{
1333 /* Create a pseudo vma that just contains the policy */
1334 vma->vm_start = 0;
1335 /* Bias interleave by inode number to distribute better across nodes */
1336 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1337 vma->vm_ops = NULL;
1338 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1339}
1340
1341static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1342{
1343 /* Drop reference taken by mpol_shared_policy_lookup() */
1344 mpol_cond_put(vma->vm_policy);
1345}
1346
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001347static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1348 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001351 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001353 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001354 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001355 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001356
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001357 return page;
1358}
Mel Gorman18a2f372012-12-05 14:01:41 -08001359
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001360static struct page *shmem_alloc_hugepage(gfp_t gfp,
1361 struct shmem_inode_info *info, pgoff_t index)
1362{
1363 struct vm_area_struct pvma;
1364 struct inode *inode = &info->vfs_inode;
1365 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001366 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001367 void __rcu **results;
1368 struct page *page;
1369
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001370 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001371 return NULL;
1372
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001373 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001374 rcu_read_lock();
1375 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1376 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1377 rcu_read_unlock();
1378 return NULL;
1379 }
1380 rcu_read_unlock();
1381
1382 shmem_pseudo_vma_init(&pvma, info, hindex);
1383 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1384 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1385 shmem_pseudo_vma_destroy(&pvma);
1386 if (page)
1387 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001388 return page;
1389}
1390
1391static struct page *shmem_alloc_page(gfp_t gfp,
1392 struct shmem_inode_info *info, pgoff_t index)
1393{
1394 struct vm_area_struct pvma;
1395 struct page *page;
1396
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001397 shmem_pseudo_vma_init(&pvma, info, index);
1398 page = alloc_page_vma(gfp, &pvma, 0);
1399 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001400
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001401 return page;
1402}
1403
1404static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1405 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1406 pgoff_t index, bool huge)
1407{
1408 struct page *page;
1409 int nr;
1410 int err = -ENOSPC;
1411
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001412 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001413 huge = false;
1414 nr = huge ? HPAGE_PMD_NR : 1;
1415
1416 if (shmem_acct_block(info->flags, nr))
1417 goto failed;
1418 if (sbinfo->max_blocks) {
1419 if (percpu_counter_compare(&sbinfo->used_blocks,
1420 sbinfo->max_blocks - nr) > 0)
1421 goto unacct;
1422 percpu_counter_add(&sbinfo->used_blocks, nr);
1423 }
1424
1425 if (huge)
1426 page = shmem_alloc_hugepage(gfp, info, index);
1427 else
1428 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001429 if (page) {
1430 __SetPageLocked(page);
1431 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001432 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001433 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001434
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001435 err = -ENOMEM;
1436 if (sbinfo->max_blocks)
1437 percpu_counter_add(&sbinfo->used_blocks, -nr);
1438unacct:
1439 shmem_unacct_blocks(info->flags, nr);
1440failed:
1441 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001445 * When a page is moved from swapcache to shmem filecache (either by the
1446 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1447 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1448 * ignorance of the mapping it belongs to. If that mapping has special
1449 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1450 * we may need to copy to a suitable page before moving to filecache.
1451 *
1452 * In a future release, this may well be extended to respect cpuset and
1453 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1454 * but for now it is a simple matter of zone.
1455 */
1456static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1457{
1458 return page_zonenum(page) > gfp_zone(gfp);
1459}
1460
1461static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1462 struct shmem_inode_info *info, pgoff_t index)
1463{
1464 struct page *oldpage, *newpage;
1465 struct address_space *swap_mapping;
1466 pgoff_t swap_index;
1467 int error;
1468
1469 oldpage = *pagep;
1470 swap_index = page_private(oldpage);
1471 swap_mapping = page_mapping(oldpage);
1472
1473 /*
1474 * We have arrived here because our zones are constrained, so don't
1475 * limit chance of success by further cpuset and node constraints.
1476 */
1477 gfp &= ~GFP_CONSTRAINT_MASK;
1478 newpage = shmem_alloc_page(gfp, info, index);
1479 if (!newpage)
1480 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001481
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001482 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001483 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001484 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001485
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001486 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001487 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001488 SetPageSwapCache(newpage);
1489
1490 /*
1491 * Our caller will very soon move newpage out of swapcache, but it's
1492 * a nice clean interface for us to replace oldpage by newpage there.
1493 */
1494 spin_lock_irq(&swap_mapping->tree_lock);
1495 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1496 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001497 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001498 __inc_node_page_state(newpage, NR_FILE_PAGES);
1499 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001500 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001501 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001502
Hugh Dickins0142ef62012-06-07 14:21:09 -07001503 if (unlikely(error)) {
1504 /*
1505 * Is this possible? I think not, now that our callers check
1506 * both PageSwapCache and page_private after getting page lock;
1507 * but be defensive. Reverse old to newpage for clear and free.
1508 */
1509 oldpage = newpage;
1510 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001511 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001512 lru_cache_add_anon(newpage);
1513 *pagep = newpage;
1514 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001515
1516 ClearPageSwapCache(oldpage);
1517 set_page_private(oldpage, 0);
1518
1519 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001520 put_page(oldpage);
1521 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001522 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001523}
1524
1525/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001526 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 *
1528 * If we allocate a new one we do not mark it dirty. That's up to the
1529 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001530 * entry since a page cannot live in both the swap and page cache.
1531 *
1532 * fault_mm and fault_type are only supplied by shmem_fault:
1533 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001535static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001536 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1537 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538{
1539 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001540 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001542 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001543 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001544 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001546 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001547 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001549 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001550 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001552 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001554 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1555 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001557 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001558 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001559 if (radix_tree_exceptional_entry(page)) {
1560 swap = radix_to_swp_entry(page);
1561 page = NULL;
1562 }
1563
Hugh Dickins75edd342016-05-19 17:12:44 -07001564 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001565 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001566 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001567 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001568 }
1569
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001570 if (page && sgp == SGP_WRITE)
1571 mark_page_accessed(page);
1572
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001573 /* fallocated page? */
1574 if (page && !PageUptodate(page)) {
1575 if (sgp != SGP_READ)
1576 goto clear;
1577 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001578 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001579 page = NULL;
1580 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001581 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001582 *pagep = page;
1583 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001587 * Fast cache lookup did not find it:
1588 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001590 info = SHMEM_I(inode);
1591 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001592 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (swap.val) {
1595 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001596 page = lookup_swap_cache(swap);
1597 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001598 /* Or update major stats only when swapin succeeds?? */
1599 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001600 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001601 count_vm_event(PGMAJFAULT);
1602 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1603 }
1604 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001605 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001606 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001607 error = -ENOMEM;
1608 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 }
1611
1612 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001613 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001614 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001615 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001616 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001617 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001618 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001619 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001621 goto failed;
1622 }
1623 wait_on_page_writeback(page);
1624
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001625 if (shmem_should_replace_page(page, gfp)) {
1626 error = shmem_replace_page(&page, gfp, info, index);
1627 if (error)
1628 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 }
1630
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001631 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001632 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001633 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001634 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001635 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001636 /*
1637 * We already confirmed swap under page lock, and make
1638 * no memory allocation here, so usually no possibility
1639 * of error; but free_swap_and_cache() only trylocks a
1640 * page, so it is just possible that the entry has been
1641 * truncated or holepunched since swap was confirmed.
1642 * shmem_undo_range() will have done some of the
1643 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001644 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001645 * Reset swap.val? No, leave it so "failed" goes back to
1646 * "repeat": reading a hole and writing should succeed.
1647 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001648 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001649 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001650 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001651 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001652 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001653 if (error)
1654 goto failed;
1655
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001656 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001657
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001658 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001659 info->swapped--;
1660 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001661 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001662
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001663 if (sgp == SGP_WRITE)
1664 mark_page_accessed(page);
1665
Hugh Dickins27ab7002011-07-25 17:12:36 -07001666 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001667 set_page_dirty(page);
1668 swap_free(swap);
1669
Hugh Dickins54af60422011-08-03 16:21:24 -07001670 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001671 /* shmem_symlink() */
1672 if (mapping->a_ops != &shmem_aops)
1673 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001674 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001675 goto alloc_nohuge;
1676 if (shmem_huge == SHMEM_HUGE_FORCE)
1677 goto alloc_huge;
1678 switch (sbinfo->huge) {
1679 loff_t i_size;
1680 pgoff_t off;
1681 case SHMEM_HUGE_NEVER:
1682 goto alloc_nohuge;
1683 case SHMEM_HUGE_WITHIN_SIZE:
1684 off = round_up(index, HPAGE_PMD_NR);
1685 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1686 if (i_size >= HPAGE_PMD_SIZE &&
1687 i_size >> PAGE_SHIFT >= off)
1688 goto alloc_huge;
1689 /* fallthrough */
1690 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001691 if (sgp_huge == SGP_HUGE)
1692 goto alloc_huge;
1693 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001694 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001697alloc_huge:
1698 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1699 index, true);
1700 if (IS_ERR(page)) {
1701alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1702 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001703 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001704 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001705 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001706 error = PTR_ERR(page);
1707 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001708 if (error != -ENOSPC)
1709 goto failed;
1710 /*
1711 * Try to reclaim some spece by splitting a huge page
1712 * beyond i_size on the filesystem.
1713 */
1714 while (retry--) {
1715 int ret;
1716 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1717 if (ret == SHRINK_STOP)
1718 break;
1719 if (ret)
1720 goto alloc_nohuge;
1721 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001722 goto failed;
1723 }
1724
1725 if (PageTransHuge(page))
1726 hindex = round_down(index, HPAGE_PMD_NR);
1727 else
1728 hindex = index;
1729
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001730 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001731 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001732
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001733 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001734 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001735 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001736 goto unacct;
1737 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1738 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001739 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001740 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001741 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001742 radix_tree_preload_end();
1743 }
1744 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001745 mem_cgroup_cancel_charge(page, memcg,
1746 PageTransHuge(page));
1747 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001748 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001749 mem_cgroup_commit_charge(page, memcg, false,
1750 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001751 lru_cache_add_anon(page);
1752
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001753 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001754 info->alloced += 1 << compound_order(page);
1755 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001756 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001757 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001758 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001759
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001760 if (PageTransHuge(page) &&
1761 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1762 hindex + HPAGE_PMD_NR - 1) {
1763 /*
1764 * Part of the huge page is beyond i_size: subject
1765 * to shrink under memory pressure.
1766 */
1767 spin_lock(&sbinfo->shrinklist_lock);
1768 if (list_empty(&info->shrinklist)) {
1769 list_add_tail(&info->shrinklist,
1770 &sbinfo->shrinklist);
1771 sbinfo->shrinklist_len++;
1772 }
1773 spin_unlock(&sbinfo->shrinklist_lock);
1774 }
1775
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001776 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001777 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1778 */
1779 if (sgp == SGP_FALLOC)
1780 sgp = SGP_WRITE;
1781clear:
1782 /*
1783 * Let SGP_WRITE caller clear ends if write does not fill page;
1784 * but SGP_FALLOC on a page fallocated earlier must initialize
1785 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001786 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001787 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1788 struct page *head = compound_head(page);
1789 int i;
1790
1791 for (i = 0; i < (1 << compound_order(head)); i++) {
1792 clear_highpage(head + i);
1793 flush_dcache_page(head + i);
1794 }
1795 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001796 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001797 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001798
Hugh Dickins54af60422011-08-03 16:21:24 -07001799 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001800 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001801 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001802 if (alloced) {
1803 ClearPageDirty(page);
1804 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001805 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001806 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001807 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001808 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001809 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001810 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001811 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001812 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001813 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001814
Nick Piggind0217ac2007-07-19 01:47:03 -07001815 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001816 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001818unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001819 if (sbinfo->max_blocks)
1820 percpu_counter_sub(&sbinfo->used_blocks,
1821 1 << compound_order(page));
1822 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1823
1824 if (PageTransHuge(page)) {
1825 unlock_page(page);
1826 put_page(page);
1827 goto alloc_nohuge;
1828 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001829failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001830 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001831 error = -EEXIST;
1832unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001833 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001834 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001835 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001836 }
1837 if (error == -ENOSPC && !once++) {
1838 info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001839 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001840 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001841 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001842 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001843 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001844 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001845 goto repeat;
1846 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847}
1848
1849static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1850{
Al Viro496ad9a2013-01-23 17:07:38 -05001851 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001852 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001853 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001855 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001857 /*
1858 * Trinity finds that probing a hole which tmpfs is punching can
1859 * prevent the hole-punch from ever completing: which in turn
1860 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001861 * faulting pages into the hole while it's being punched. Although
1862 * shmem_undo_range() does remove the additions, it may be unable to
1863 * keep up, as each new page needs its own unmap_mapping_range() call,
1864 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1865 *
1866 * It does not matter if we sometimes reach this check just before the
1867 * hole-punch begins, so that one fault then races with the punch:
1868 * we just need to make racing faults a rare case.
1869 *
1870 * The implementation below would be much simpler if we just used a
1871 * standard mutex or completion: but we cannot take i_mutex in fault,
1872 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001873 */
1874 if (unlikely(inode->i_private)) {
1875 struct shmem_falloc *shmem_falloc;
1876
1877 spin_lock(&inode->i_lock);
1878 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001879 if (shmem_falloc &&
1880 shmem_falloc->waitq &&
1881 vmf->pgoff >= shmem_falloc->start &&
1882 vmf->pgoff < shmem_falloc->next) {
1883 wait_queue_head_t *shmem_falloc_waitq;
1884 DEFINE_WAIT(shmem_fault_wait);
1885
1886 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001887 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1888 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001889 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001890 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001891 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001892 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001893
1894 shmem_falloc_waitq = shmem_falloc->waitq;
1895 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1896 TASK_UNINTERRUPTIBLE);
1897 spin_unlock(&inode->i_lock);
1898 schedule();
1899
1900 /*
1901 * shmem_falloc_waitq points into the shmem_fallocate()
1902 * stack of the hole-punching task: shmem_falloc_waitq
1903 * is usually invalid by the time we reach here, but
1904 * finish_wait() does not dereference it in that case;
1905 * though i_lock needed lest racing with wake_up_all().
1906 */
1907 spin_lock(&inode->i_lock);
1908 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1909 spin_unlock(&inode->i_lock);
1910 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001911 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001912 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001913 }
1914
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001915 sgp = SGP_CACHE;
1916 if (vma->vm_flags & VM_HUGEPAGE)
1917 sgp = SGP_HUGE;
1918 else if (vma->vm_flags & VM_NOHUGEPAGE)
1919 sgp = SGP_NOHUGE;
1920
1921 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001922 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 if (error)
1924 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001925 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926}
1927
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001928unsigned long shmem_get_unmapped_area(struct file *file,
1929 unsigned long uaddr, unsigned long len,
1930 unsigned long pgoff, unsigned long flags)
1931{
1932 unsigned long (*get_area)(struct file *,
1933 unsigned long, unsigned long, unsigned long, unsigned long);
1934 unsigned long addr;
1935 unsigned long offset;
1936 unsigned long inflated_len;
1937 unsigned long inflated_addr;
1938 unsigned long inflated_offset;
1939
1940 if (len > TASK_SIZE)
1941 return -ENOMEM;
1942
1943 get_area = current->mm->get_unmapped_area;
1944 addr = get_area(file, uaddr, len, pgoff, flags);
1945
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001946 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001947 return addr;
1948 if (IS_ERR_VALUE(addr))
1949 return addr;
1950 if (addr & ~PAGE_MASK)
1951 return addr;
1952 if (addr > TASK_SIZE - len)
1953 return addr;
1954
1955 if (shmem_huge == SHMEM_HUGE_DENY)
1956 return addr;
1957 if (len < HPAGE_PMD_SIZE)
1958 return addr;
1959 if (flags & MAP_FIXED)
1960 return addr;
1961 /*
1962 * Our priority is to support MAP_SHARED mapped hugely;
1963 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1964 * But if caller specified an address hint, respect that as before.
1965 */
1966 if (uaddr)
1967 return addr;
1968
1969 if (shmem_huge != SHMEM_HUGE_FORCE) {
1970 struct super_block *sb;
1971
1972 if (file) {
1973 VM_BUG_ON(file->f_op != &shmem_file_operations);
1974 sb = file_inode(file)->i_sb;
1975 } else {
1976 /*
1977 * Called directly from mm/mmap.c, or drivers/char/mem.c
1978 * for "/dev/zero", to create a shared anonymous object.
1979 */
1980 if (IS_ERR(shm_mnt))
1981 return addr;
1982 sb = shm_mnt->mnt_sb;
1983 }
Toshi Kani3089bf62016-09-23 20:21:56 -07001984 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001985 return addr;
1986 }
1987
1988 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
1989 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
1990 return addr;
1991 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
1992 return addr;
1993
1994 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
1995 if (inflated_len > TASK_SIZE)
1996 return addr;
1997 if (inflated_len < len)
1998 return addr;
1999
2000 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2001 if (IS_ERR_VALUE(inflated_addr))
2002 return addr;
2003 if (inflated_addr & ~PAGE_MASK)
2004 return addr;
2005
2006 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2007 inflated_addr += offset - inflated_offset;
2008 if (inflated_offset > offset)
2009 inflated_addr += HPAGE_PMD_SIZE;
2010
2011 if (inflated_addr > TASK_SIZE - len)
2012 return addr;
2013 return inflated_addr;
2014}
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002017static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018{
Al Viro496ad9a2013-01-23 17:07:38 -05002019 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002020 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021}
2022
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002023static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2024 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
Al Viro496ad9a2013-01-23 17:07:38 -05002026 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002027 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002029 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2030 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031}
2032#endif
2033
2034int shmem_lock(struct file *file, int lock, struct user_struct *user)
2035{
Al Viro496ad9a2013-01-23 17:07:38 -05002036 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 struct shmem_inode_info *info = SHMEM_I(inode);
2038 int retval = -ENOMEM;
2039
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002040 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 if (lock && !(info->flags & VM_LOCKED)) {
2042 if (!user_shm_lock(inode->i_size, user))
2043 goto out_nomem;
2044 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002045 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 }
2047 if (!lock && (info->flags & VM_LOCKED) && user) {
2048 user_shm_unlock(inode->i_size, user);
2049 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002050 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
2052 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002055 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 return retval;
2057}
2058
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002059static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060{
2061 file_accessed(file);
2062 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002063 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002064 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2065 (vma->vm_end & HPAGE_PMD_MASK)) {
2066 khugepaged_enter(vma, vma->vm_flags);
2067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 return 0;
2069}
2070
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002071static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002072 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073{
2074 struct inode *inode;
2075 struct shmem_inode_info *info;
2076 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2077
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002078 if (shmem_reserve_inode(sb))
2079 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
2081 inode = new_inode(sb);
2082 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002083 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002084 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002086 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002087 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 info = SHMEM_I(inode);
2089 memset(info, 0, (char *)inode - (char *)info);
2090 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002091 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002092 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002093 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002095 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002096 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
2098 switch (mode & S_IFMT) {
2099 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002100 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 init_special_inode(inode, mode, dev);
2102 break;
2103 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002104 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 inode->i_op = &shmem_inode_operations;
2106 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002107 mpol_shared_policy_init(&info->policy,
2108 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 break;
2110 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002111 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 /* Some things misbehave if size == 0 on a directory */
2113 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2114 inode->i_op = &shmem_dir_inode_operations;
2115 inode->i_fop = &simple_dir_operations;
2116 break;
2117 case S_IFLNK:
2118 /*
2119 * Must not load anything in the rbtree,
2120 * mpol_free_shared_policy will not be called.
2121 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002122 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 break;
2124 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002125 } else
2126 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 return inode;
2128}
2129
Johannes Weiner0cd61442014-04-03 14:47:46 -07002130bool shmem_mapping(struct address_space *mapping)
2131{
Sasha Levinf0774d82015-02-23 05:38:00 -05002132 if (!mapping->host)
2133 return false;
2134
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002135 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002136}
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002139static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002140static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002142#ifdef CONFIG_TMPFS_XATTR
2143static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2144#else
2145#define shmem_initxattrs NULL
2146#endif
2147
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002149shmem_write_begin(struct file *file, struct address_space *mapping,
2150 loff_t pos, unsigned len, unsigned flags,
2151 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
Nick Piggin800d15a2007-10-16 01:25:03 -07002153 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002154 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002155 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002156
2157 /* i_mutex is held by caller */
2158 if (unlikely(info->seals)) {
2159 if (info->seals & F_SEAL_WRITE)
2160 return -EPERM;
2161 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2162 return -EPERM;
2163 }
2164
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002165 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002166}
2167
2168static int
2169shmem_write_end(struct file *file, struct address_space *mapping,
2170 loff_t pos, unsigned len, unsigned copied,
2171 struct page *page, void *fsdata)
2172{
2173 struct inode *inode = mapping->host;
2174
Hugh Dickinsd3602442008-02-04 22:28:44 -08002175 if (pos + copied > inode->i_size)
2176 i_size_write(inode, pos + copied);
2177
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002178 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002179 struct page *head = compound_head(page);
2180 if (PageTransCompound(page)) {
2181 int i;
2182
2183 for (i = 0; i < HPAGE_PMD_NR; i++) {
2184 if (head + i == page)
2185 continue;
2186 clear_highpage(head + i);
2187 flush_dcache_page(head + i);
2188 }
2189 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002190 if (copied < PAGE_SIZE) {
2191 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002192 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002193 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002194 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002195 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002196 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002197 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002198 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002199 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002200
Nick Piggin800d15a2007-10-16 01:25:03 -07002201 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202}
2203
Al Viro2ba5bbe2014-04-02 20:00:02 -04002204static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
Al Viro6e58e792014-02-03 17:07:03 -05002206 struct file *file = iocb->ki_filp;
2207 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002209 pgoff_t index;
2210 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002211 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002212 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002213 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002214 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002215
2216 /*
2217 * Might this read be for a stacking filesystem? Then when reading
2218 * holes of a sparse file, we actually need to allocate those pages,
2219 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2220 */
Al Viro777eda22014-12-17 04:46:46 -05002221 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002222 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002224 index = *ppos >> PAGE_SHIFT;
2225 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226
2227 for (;;) {
2228 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002229 pgoff_t end_index;
2230 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 loff_t i_size = i_size_read(inode);
2232
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002233 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 if (index > end_index)
2235 break;
2236 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002237 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 if (nr <= offset)
2239 break;
2240 }
2241
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002242 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002243 if (error) {
2244 if (error == -EINVAL)
2245 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 break;
2247 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002248 if (page) {
2249 if (sgp == SGP_CACHE)
2250 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002251 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
2254 /*
2255 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002256 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002258 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002260 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002262 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 if (nr <= offset) {
2264 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002265 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 break;
2267 }
2268 }
2269 nr -= offset;
2270
2271 if (page) {
2272 /*
2273 * If users can be writing to this page using arbitrary
2274 * virtual addresses, take care about potential aliasing
2275 * before reading the page on the kernel side.
2276 */
2277 if (mapping_writably_mapped(mapping))
2278 flush_dcache_page(page);
2279 /*
2280 * Mark the page accessed if we read the beginning.
2281 */
2282 if (!offset)
2283 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002284 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002286 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288
2289 /*
2290 * Ok, we have the page, and it's up-to-date, so
2291 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002293 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002294 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002296 index += offset >> PAGE_SHIFT;
2297 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002299 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002300 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 break;
Al Viro6e58e792014-02-03 17:07:03 -05002302 if (ret < nr) {
2303 error = -EFAULT;
2304 break;
2305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 cond_resched();
2307 }
2308
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002309 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002310 file_accessed(file);
2311 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312}
2313
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002314/*
2315 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2316 */
2317static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002318 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002319{
2320 struct page *page;
2321 struct pagevec pvec;
2322 pgoff_t indices[PAGEVEC_SIZE];
2323 bool done = false;
2324 int i;
2325
2326 pagevec_init(&pvec, 0);
2327 pvec.nr = 1; /* start small: we may be there already */
2328 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002329 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002330 pvec.nr, pvec.pages, indices);
2331 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002332 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002333 index = end;
2334 break;
2335 }
2336 for (i = 0; i < pvec.nr; i++, index++) {
2337 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002338 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002339 done = true;
2340 break;
2341 }
2342 index = indices[i];
2343 }
2344 page = pvec.pages[i];
2345 if (page && !radix_tree_exceptional_entry(page)) {
2346 if (!PageUptodate(page))
2347 page = NULL;
2348 }
2349 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002350 (page && whence == SEEK_DATA) ||
2351 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002352 done = true;
2353 break;
2354 }
2355 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002356 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002357 pagevec_release(&pvec);
2358 pvec.nr = PAGEVEC_SIZE;
2359 cond_resched();
2360 }
2361 return index;
2362}
2363
Andrew Morton965c8e52012-12-17 15:59:39 -08002364static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002365{
2366 struct address_space *mapping = file->f_mapping;
2367 struct inode *inode = mapping->host;
2368 pgoff_t start, end;
2369 loff_t new_offset;
2370
Andrew Morton965c8e52012-12-17 15:59:39 -08002371 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2372 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002373 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002374 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002375 /* We're holding i_mutex so we can access i_size directly */
2376
2377 if (offset < 0)
2378 offset = -EINVAL;
2379 else if (offset >= inode->i_size)
2380 offset = -ENXIO;
2381 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002382 start = offset >> PAGE_SHIFT;
2383 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002384 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002385 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002386 if (new_offset > offset) {
2387 if (new_offset < inode->i_size)
2388 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002389 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002390 offset = -ENXIO;
2391 else
2392 offset = inode->i_size;
2393 }
2394 }
2395
Hugh Dickins387aae62013-08-04 11:30:25 -07002396 if (offset >= 0)
2397 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002398 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002399 return offset;
2400}
2401
David Herrmann05f65b52014-08-08 14:25:36 -07002402/*
2403 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2404 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2405 */
2406#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2407#define LAST_SCAN 4 /* about 150ms max */
2408
2409static void shmem_tag_pins(struct address_space *mapping)
2410{
2411 struct radix_tree_iter iter;
2412 void **slot;
2413 pgoff_t start;
2414 struct page *page;
2415
2416 lru_add_drain();
2417 start = 0;
2418 rcu_read_lock();
2419
David Herrmann05f65b52014-08-08 14:25:36 -07002420 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2421 page = radix_tree_deref_slot(slot);
2422 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002423 if (radix_tree_deref_retry(page)) {
2424 slot = radix_tree_iter_retry(&iter);
2425 continue;
2426 }
David Herrmann05f65b52014-08-08 14:25:36 -07002427 } else if (page_count(page) - page_mapcount(page) > 1) {
2428 spin_lock_irq(&mapping->tree_lock);
2429 radix_tree_tag_set(&mapping->page_tree, iter.index,
2430 SHMEM_TAG_PINNED);
2431 spin_unlock_irq(&mapping->tree_lock);
2432 }
2433
2434 if (need_resched()) {
2435 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002436 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002437 }
2438 }
2439 rcu_read_unlock();
2440}
2441
2442/*
2443 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2444 * via get_user_pages(), drivers might have some pending I/O without any active
2445 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2446 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2447 * them to be dropped.
2448 * The caller must guarantee that no new user will acquire writable references
2449 * to those pages to avoid races.
2450 */
David Herrmann40e041a2014-08-08 14:25:27 -07002451static int shmem_wait_for_pins(struct address_space *mapping)
2452{
David Herrmann05f65b52014-08-08 14:25:36 -07002453 struct radix_tree_iter iter;
2454 void **slot;
2455 pgoff_t start;
2456 struct page *page;
2457 int error, scan;
2458
2459 shmem_tag_pins(mapping);
2460
2461 error = 0;
2462 for (scan = 0; scan <= LAST_SCAN; scan++) {
2463 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2464 break;
2465
2466 if (!scan)
2467 lru_add_drain_all();
2468 else if (schedule_timeout_killable((HZ << scan) / 200))
2469 scan = LAST_SCAN;
2470
2471 start = 0;
2472 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002473 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2474 start, SHMEM_TAG_PINNED) {
2475
2476 page = radix_tree_deref_slot(slot);
2477 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002478 if (radix_tree_deref_retry(page)) {
2479 slot = radix_tree_iter_retry(&iter);
2480 continue;
2481 }
David Herrmann05f65b52014-08-08 14:25:36 -07002482
2483 page = NULL;
2484 }
2485
2486 if (page &&
2487 page_count(page) - page_mapcount(page) != 1) {
2488 if (scan < LAST_SCAN)
2489 goto continue_resched;
2490
2491 /*
2492 * On the last scan, we clean up all those tags
2493 * we inserted; but make a note that we still
2494 * found pages pinned.
2495 */
2496 error = -EBUSY;
2497 }
2498
2499 spin_lock_irq(&mapping->tree_lock);
2500 radix_tree_tag_clear(&mapping->page_tree,
2501 iter.index, SHMEM_TAG_PINNED);
2502 spin_unlock_irq(&mapping->tree_lock);
2503continue_resched:
2504 if (need_resched()) {
2505 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002506 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002507 }
2508 }
2509 rcu_read_unlock();
2510 }
2511
2512 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002513}
2514
2515#define F_ALL_SEALS (F_SEAL_SEAL | \
2516 F_SEAL_SHRINK | \
2517 F_SEAL_GROW | \
2518 F_SEAL_WRITE)
2519
2520int shmem_add_seals(struct file *file, unsigned int seals)
2521{
2522 struct inode *inode = file_inode(file);
2523 struct shmem_inode_info *info = SHMEM_I(inode);
2524 int error;
2525
2526 /*
2527 * SEALING
2528 * Sealing allows multiple parties to share a shmem-file but restrict
2529 * access to a specific subset of file operations. Seals can only be
2530 * added, but never removed. This way, mutually untrusted parties can
2531 * share common memory regions with a well-defined policy. A malicious
2532 * peer can thus never perform unwanted operations on a shared object.
2533 *
2534 * Seals are only supported on special shmem-files and always affect
2535 * the whole underlying inode. Once a seal is set, it may prevent some
2536 * kinds of access to the file. Currently, the following seals are
2537 * defined:
2538 * SEAL_SEAL: Prevent further seals from being set on this file
2539 * SEAL_SHRINK: Prevent the file from shrinking
2540 * SEAL_GROW: Prevent the file from growing
2541 * SEAL_WRITE: Prevent write access to the file
2542 *
2543 * As we don't require any trust relationship between two parties, we
2544 * must prevent seals from being removed. Therefore, sealing a file
2545 * only adds a given set of seals to the file, it never touches
2546 * existing seals. Furthermore, the "setting seals"-operation can be
2547 * sealed itself, which basically prevents any further seal from being
2548 * added.
2549 *
2550 * Semantics of sealing are only defined on volatile files. Only
2551 * anonymous shmem files support sealing. More importantly, seals are
2552 * never written to disk. Therefore, there's no plan to support it on
2553 * other file types.
2554 */
2555
2556 if (file->f_op != &shmem_file_operations)
2557 return -EINVAL;
2558 if (!(file->f_mode & FMODE_WRITE))
2559 return -EPERM;
2560 if (seals & ~(unsigned int)F_ALL_SEALS)
2561 return -EINVAL;
2562
Al Viro59551022016-01-22 15:40:57 -05002563 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002564
2565 if (info->seals & F_SEAL_SEAL) {
2566 error = -EPERM;
2567 goto unlock;
2568 }
2569
2570 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2571 error = mapping_deny_writable(file->f_mapping);
2572 if (error)
2573 goto unlock;
2574
2575 error = shmem_wait_for_pins(file->f_mapping);
2576 if (error) {
2577 mapping_allow_writable(file->f_mapping);
2578 goto unlock;
2579 }
2580 }
2581
2582 info->seals |= seals;
2583 error = 0;
2584
2585unlock:
Al Viro59551022016-01-22 15:40:57 -05002586 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002587 return error;
2588}
2589EXPORT_SYMBOL_GPL(shmem_add_seals);
2590
2591int shmem_get_seals(struct file *file)
2592{
2593 if (file->f_op != &shmem_file_operations)
2594 return -EINVAL;
2595
2596 return SHMEM_I(file_inode(file))->seals;
2597}
2598EXPORT_SYMBOL_GPL(shmem_get_seals);
2599
2600long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2601{
2602 long error;
2603
2604 switch (cmd) {
2605 case F_ADD_SEALS:
2606 /* disallow upper 32bit */
2607 if (arg > UINT_MAX)
2608 return -EINVAL;
2609
2610 error = shmem_add_seals(file, arg);
2611 break;
2612 case F_GET_SEALS:
2613 error = shmem_get_seals(file);
2614 break;
2615 default:
2616 error = -EINVAL;
2617 break;
2618 }
2619
2620 return error;
2621}
2622
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002623static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2624 loff_t len)
2625{
Al Viro496ad9a2013-01-23 17:07:38 -05002626 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002627 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002628 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002629 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002630 pgoff_t start, index, end;
2631 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002632
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002633 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2634 return -EOPNOTSUPP;
2635
Al Viro59551022016-01-22 15:40:57 -05002636 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002637
2638 if (mode & FALLOC_FL_PUNCH_HOLE) {
2639 struct address_space *mapping = file->f_mapping;
2640 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2641 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002642 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002643
David Herrmann40e041a2014-08-08 14:25:27 -07002644 /* protected by i_mutex */
2645 if (info->seals & F_SEAL_WRITE) {
2646 error = -EPERM;
2647 goto out;
2648 }
2649
Hugh Dickins8e205f72014-07-23 14:00:10 -07002650 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002651 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2652 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2653 spin_lock(&inode->i_lock);
2654 inode->i_private = &shmem_falloc;
2655 spin_unlock(&inode->i_lock);
2656
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002657 if ((u64)unmap_end > (u64)unmap_start)
2658 unmap_mapping_range(mapping, unmap_start,
2659 1 + unmap_end - unmap_start, 0);
2660 shmem_truncate_range(inode, offset, offset + len - 1);
2661 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002662
2663 spin_lock(&inode->i_lock);
2664 inode->i_private = NULL;
2665 wake_up_all(&shmem_falloc_waitq);
2666 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002667 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002668 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002669 }
2670
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002671 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2672 error = inode_newsize_ok(inode, offset + len);
2673 if (error)
2674 goto out;
2675
David Herrmann40e041a2014-08-08 14:25:27 -07002676 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2677 error = -EPERM;
2678 goto out;
2679 }
2680
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002681 start = offset >> PAGE_SHIFT;
2682 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002683 /* Try to avoid a swapstorm if len is impossible to satisfy */
2684 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2685 error = -ENOSPC;
2686 goto out;
2687 }
2688
Hugh Dickins8e205f72014-07-23 14:00:10 -07002689 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002690 shmem_falloc.start = start;
2691 shmem_falloc.next = start;
2692 shmem_falloc.nr_falloced = 0;
2693 shmem_falloc.nr_unswapped = 0;
2694 spin_lock(&inode->i_lock);
2695 inode->i_private = &shmem_falloc;
2696 spin_unlock(&inode->i_lock);
2697
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002698 for (index = start; index < end; index++) {
2699 struct page *page;
2700
2701 /*
2702 * Good, the fallocate(2) manpage permits EINTR: we may have
2703 * been interrupted because we are using up too much memory.
2704 */
2705 if (signal_pending(current))
2706 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002707 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2708 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002709 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002710 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002711 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002712 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002713 if (index > start) {
2714 shmem_undo_range(inode,
2715 (loff_t)start << PAGE_SHIFT,
2716 ((loff_t)index << PAGE_SHIFT) - 1, true);
2717 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002718 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002719 }
2720
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002721 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002722 * Inform shmem_writepage() how far we have reached.
2723 * No need for lock or barrier: we have the page lock.
2724 */
2725 shmem_falloc.next++;
2726 if (!PageUptodate(page))
2727 shmem_falloc.nr_falloced++;
2728
2729 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002730 * If !PageUptodate, leave it that way so that freeable pages
2731 * can be recognized if we need to rollback on error later.
2732 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002733 * than free the pages we are allocating (and SGP_CACHE pages
2734 * might still be clean: we now need to mark those dirty too).
2735 */
2736 set_page_dirty(page);
2737 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002738 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002739 cond_resched();
2740 }
2741
2742 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2743 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002744 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002745undone:
2746 spin_lock(&inode->i_lock);
2747 inode->i_private = NULL;
2748 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002749out:
Al Viro59551022016-01-22 15:40:57 -05002750 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002751 return error;
2752}
2753
David Howells726c3342006-06-23 02:02:58 -07002754static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755{
David Howells726c3342006-06-23 02:02:58 -07002756 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
2758 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002759 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002761 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002763 buf->f_bavail =
2764 buf->f_bfree = sbinfo->max_blocks -
2765 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002766 }
2767 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 buf->f_files = sbinfo->max_inodes;
2769 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 }
2771 /* else leave those fields 0 like simple_statfs */
2772 return 0;
2773}
2774
2775/*
2776 * File creation. Allocate an inode, and we're done..
2777 */
2778static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002779shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002781 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 int error = -ENOSPC;
2783
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002784 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002786 error = simple_acl_create(dir, inode);
2787 if (error)
2788 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002789 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002790 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002791 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002792 if (error && error != -EOPNOTSUPP)
2793 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002794
Al Viro718deb62009-12-16 19:35:36 -05002795 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002797 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 d_instantiate(dentry, inode);
2799 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 }
2801 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002802out_iput:
2803 iput(inode);
2804 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805}
2806
Al Viro60545d02013-06-07 01:20:27 -04002807static int
2808shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2809{
2810 struct inode *inode;
2811 int error = -ENOSPC;
2812
2813 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2814 if (inode) {
2815 error = security_inode_init_security(inode, dir,
2816 NULL,
2817 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002818 if (error && error != -EOPNOTSUPP)
2819 goto out_iput;
2820 error = simple_acl_create(dir, inode);
2821 if (error)
2822 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002823 d_tmpfile(dentry, inode);
2824 }
2825 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002826out_iput:
2827 iput(inode);
2828 return error;
Al Viro60545d02013-06-07 01:20:27 -04002829}
2830
Al Viro18bb1db2011-07-26 01:41:39 -04002831static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832{
2833 int error;
2834
2835 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2836 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002837 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 return 0;
2839}
2840
Al Viro4acdaf22011-07-26 01:42:34 -04002841static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002842 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843{
2844 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2845}
2846
2847/*
2848 * Link a file..
2849 */
2850static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2851{
David Howells75c3cfa2015-03-17 22:26:12 +00002852 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002853 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
2855 /*
2856 * No ordinary (disk based) filesystem counts links as inodes;
2857 * but each new link needs a new dentry, pinning lowmem, and
2858 * tmpfs dentries cannot be pruned until they are unlinked.
2859 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002860 ret = shmem_reserve_inode(inode->i_sb);
2861 if (ret)
2862 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863
2864 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002865 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002866 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002867 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 dget(dentry); /* Extra pinning count for the created dentry */
2869 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002870out:
2871 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872}
2873
2874static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2875{
David Howells75c3cfa2015-03-17 22:26:12 +00002876 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002878 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2879 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
2881 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002882 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002883 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 dput(dentry); /* Undo the count from "create" - this does all the work */
2885 return 0;
2886}
2887
2888static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2889{
2890 if (!simple_empty(dentry))
2891 return -ENOTEMPTY;
2892
David Howells75c3cfa2015-03-17 22:26:12 +00002893 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002894 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 return shmem_unlink(dir, dentry);
2896}
2897
Miklos Szeredi37456772014-07-23 15:15:34 +02002898static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2899{
David Howellse36cb0b2015-01-29 12:02:35 +00002900 bool old_is_dir = d_is_dir(old_dentry);
2901 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002902
2903 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2904 if (old_is_dir) {
2905 drop_nlink(old_dir);
2906 inc_nlink(new_dir);
2907 } else {
2908 drop_nlink(new_dir);
2909 inc_nlink(old_dir);
2910 }
2911 }
2912 old_dir->i_ctime = old_dir->i_mtime =
2913 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002914 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002915 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02002916
2917 return 0;
2918}
2919
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002920static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2921{
2922 struct dentry *whiteout;
2923 int error;
2924
2925 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2926 if (!whiteout)
2927 return -ENOMEM;
2928
2929 error = shmem_mknod(old_dir, whiteout,
2930 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2931 dput(whiteout);
2932 if (error)
2933 return error;
2934
2935 /*
2936 * Cheat and hash the whiteout while the old dentry is still in
2937 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2938 *
2939 * d_lookup() will consistently find one of them at this point,
2940 * not sure which one, but that isn't even important.
2941 */
2942 d_rehash(whiteout);
2943 return 0;
2944}
2945
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946/*
2947 * The VFS layer already does all the dentry stuff for rename,
2948 * we just have to decrement the usage count for the target if
2949 * it exists so that the VFS layer correctly free's it when it
2950 * gets overwritten.
2951 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002952static 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 -07002953{
David Howells75c3cfa2015-03-17 22:26:12 +00002954 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 int they_are_dirs = S_ISDIR(inode->i_mode);
2956
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002957 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02002958 return -EINVAL;
2959
Miklos Szeredi37456772014-07-23 15:15:34 +02002960 if (flags & RENAME_EXCHANGE)
2961 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2962
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 if (!simple_empty(new_dentry))
2964 return -ENOTEMPTY;
2965
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002966 if (flags & RENAME_WHITEOUT) {
2967 int error;
2968
2969 error = shmem_whiteout(old_dir, old_dentry);
2970 if (error)
2971 return error;
2972 }
2973
David Howells75c3cfa2015-03-17 22:26:12 +00002974 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02002976 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00002977 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002978 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02002979 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002981 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002982 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 }
2984
2985 old_dir->i_size -= BOGO_DIRENT_SIZE;
2986 new_dir->i_size += BOGO_DIRENT_SIZE;
2987 old_dir->i_ctime = old_dir->i_mtime =
2988 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002989 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 return 0;
2991}
2992
2993static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2994{
2995 int error;
2996 int len;
2997 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002998 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 struct shmem_inode_info *info;
3000
3001 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003002 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 return -ENAMETOOLONG;
3004
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003005 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 if (!inode)
3007 return -ENOSPC;
3008
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003009 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003010 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003011 if (error) {
3012 if (error != -EOPNOTSUPP) {
3013 iput(inode);
3014 return error;
3015 }
3016 error = 0;
3017 }
3018
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 info = SHMEM_I(inode);
3020 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003021 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003022 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3023 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003024 iput(inode);
3025 return -ENOMEM;
3026 }
3027 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003029 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003030 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 if (error) {
3032 iput(inode);
3033 return error;
3034 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003035 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003037 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003038 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003040 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003041 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003044 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 d_instantiate(dentry, inode);
3046 dget(dentry);
3047 return 0;
3048}
3049
Al Virofceef392015-12-29 15:58:39 -05003050static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051{
Al Virofceef392015-12-29 15:58:39 -05003052 mark_page_accessed(arg);
3053 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054}
3055
Al Viro6b255392015-11-17 10:20:54 -05003056static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003057 struct inode *inode,
3058 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003061 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003062 if (!dentry) {
3063 page = find_get_page(inode->i_mapping, 0);
3064 if (!page)
3065 return ERR_PTR(-ECHILD);
3066 if (!PageUptodate(page)) {
3067 put_page(page);
3068 return ERR_PTR(-ECHILD);
3069 }
3070 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003071 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003072 if (error)
3073 return ERR_PTR(error);
3074 unlock_page(page);
3075 }
Al Virofceef392015-12-29 15:58:39 -05003076 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003077 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078}
3079
Eric Parisb09e0fa2011-05-24 17:12:39 -07003080#ifdef CONFIG_TMPFS_XATTR
3081/*
3082 * Superblocks without xattr inode operations may get some security.* xattr
3083 * support from the LSM "for free". As soon as we have any other xattrs
3084 * like ACLs, we also need to implement the security.* handlers at
3085 * filesystem level, though.
3086 */
3087
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003088/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003089 * Callback for security_inode_init_security() for acquiring xattrs.
3090 */
3091static int shmem_initxattrs(struct inode *inode,
3092 const struct xattr *xattr_array,
3093 void *fs_info)
3094{
3095 struct shmem_inode_info *info = SHMEM_I(inode);
3096 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003097 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003098 size_t len;
3099
3100 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003101 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003102 if (!new_xattr)
3103 return -ENOMEM;
3104
3105 len = strlen(xattr->name) + 1;
3106 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3107 GFP_KERNEL);
3108 if (!new_xattr->name) {
3109 kfree(new_xattr);
3110 return -ENOMEM;
3111 }
3112
3113 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3114 XATTR_SECURITY_PREFIX_LEN);
3115 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3116 xattr->name, len);
3117
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003118 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003119 }
3120
3121 return 0;
3122}
3123
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003124static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003125 struct dentry *unused, struct inode *inode,
3126 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003127{
Al Virob2968212016-04-10 20:48:24 -04003128 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003129
3130 name = xattr_full_name(handler, name);
3131 return simple_xattr_get(&info->xattrs, name, buffer, size);
3132}
3133
3134static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003135 struct dentry *unused, struct inode *inode,
3136 const char *name, const void *value,
3137 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003138{
Al Viro59301222016-05-27 10:19:30 -04003139 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003140
3141 name = xattr_full_name(handler, name);
3142 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3143}
3144
3145static const struct xattr_handler shmem_security_xattr_handler = {
3146 .prefix = XATTR_SECURITY_PREFIX,
3147 .get = shmem_xattr_handler_get,
3148 .set = shmem_xattr_handler_set,
3149};
3150
3151static const struct xattr_handler shmem_trusted_xattr_handler = {
3152 .prefix = XATTR_TRUSTED_PREFIX,
3153 .get = shmem_xattr_handler_get,
3154 .set = shmem_xattr_handler_set,
3155};
3156
Eric Parisb09e0fa2011-05-24 17:12:39 -07003157static const struct xattr_handler *shmem_xattr_handlers[] = {
3158#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003159 &posix_acl_access_xattr_handler,
3160 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003161#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003162 &shmem_security_xattr_handler,
3163 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003164 NULL
3165};
3166
Eric Parisb09e0fa2011-05-24 17:12:39 -07003167static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3168{
David Howells75c3cfa2015-03-17 22:26:12 +00003169 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003170 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003171}
3172#endif /* CONFIG_TMPFS_XATTR */
3173
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003174static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003176 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003177#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003178 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003179#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180};
3181
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003182static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003184 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003185#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003186 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003187#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003188};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003189
David M. Grimes91828a42006-10-17 00:09:45 -07003190static struct dentry *shmem_get_parent(struct dentry *child)
3191{
3192 return ERR_PTR(-ESTALE);
3193}
3194
3195static int shmem_match(struct inode *ino, void *vfh)
3196{
3197 __u32 *fh = vfh;
3198 __u64 inum = fh[2];
3199 inum = (inum << 32) | fh[1];
3200 return ino->i_ino == inum && fh[0] == ino->i_generation;
3201}
3202
Christoph Hellwig480b1162007-10-21 16:42:13 -07003203static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3204 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003205{
David M. Grimes91828a42006-10-17 00:09:45 -07003206 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003207 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003208 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003209
Christoph Hellwig480b1162007-10-21 16:42:13 -07003210 if (fh_len < 3)
3211 return NULL;
3212
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003213 inum = fid->raw[2];
3214 inum = (inum << 32) | fid->raw[1];
3215
Christoph Hellwig480b1162007-10-21 16:42:13 -07003216 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3217 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003218 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003219 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003220 iput(inode);
3221 }
3222
Christoph Hellwig480b1162007-10-21 16:42:13 -07003223 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003224}
3225
Al Virob0b03822012-04-02 14:34:06 -04003226static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3227 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003228{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303229 if (*len < 3) {
3230 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003231 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303232 }
David M. Grimes91828a42006-10-17 00:09:45 -07003233
Al Viro1d3382cb2010-10-23 15:19:20 -04003234 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003235 /* Unfortunately insert_inode_hash is not idempotent,
3236 * so as we hash inodes here rather than at creation
3237 * time, we need a lock to ensure we only try
3238 * to do it once
3239 */
3240 static DEFINE_SPINLOCK(lock);
3241 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003242 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003243 __insert_inode_hash(inode,
3244 inode->i_ino + inode->i_generation);
3245 spin_unlock(&lock);
3246 }
3247
3248 fh[0] = inode->i_generation;
3249 fh[1] = inode->i_ino;
3250 fh[2] = ((__u64)inode->i_ino) >> 32;
3251
3252 *len = 3;
3253 return 1;
3254}
3255
Christoph Hellwig39655162007-10-21 16:42:17 -07003256static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003257 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003258 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003259 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003260};
3261
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003262static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3263 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264{
3265 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003266 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003267 uid_t uid;
3268 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003270 while (options != NULL) {
3271 this_char = options;
3272 for (;;) {
3273 /*
3274 * NUL-terminate this option: unfortunately,
3275 * mount options form a comma-separated list,
3276 * but mpol's nodelist may also contain commas.
3277 */
3278 options = strchr(options, ',');
3279 if (options == NULL)
3280 break;
3281 options++;
3282 if (!isdigit(*options)) {
3283 options[-1] = '\0';
3284 break;
3285 }
3286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 if (!*this_char)
3288 continue;
3289 if ((value = strchr(this_char,'=')) != NULL) {
3290 *value++ = 0;
3291 } else {
Joe Perches11705322016-03-17 14:19:50 -07003292 pr_err("tmpfs: No value for mount option '%s'\n",
3293 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003294 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 }
3296
3297 if (!strcmp(this_char,"size")) {
3298 unsigned long long size;
3299 size = memparse(value,&rest);
3300 if (*rest == '%') {
3301 size <<= PAGE_SHIFT;
3302 size *= totalram_pages;
3303 do_div(size, 100);
3304 rest++;
3305 }
3306 if (*rest)
3307 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003308 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003309 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003311 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 if (*rest)
3313 goto bad_val;
3314 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003315 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 if (*rest)
3317 goto bad_val;
3318 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003319 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003321 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 if (*rest)
3323 goto bad_val;
3324 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003325 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003327 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 if (*rest)
3329 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003330 sbinfo->uid = make_kuid(current_user_ns(), uid);
3331 if (!uid_valid(sbinfo->uid))
3332 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003334 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003336 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 if (*rest)
3338 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003339 sbinfo->gid = make_kgid(current_user_ns(), gid);
3340 if (!gid_valid(sbinfo->gid))
3341 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003342#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003343 } else if (!strcmp(this_char, "huge")) {
3344 int huge;
3345 huge = shmem_parse_huge(value);
3346 if (huge < 0)
3347 goto bad_val;
3348 if (!has_transparent_hugepage() &&
3349 huge != SHMEM_HUGE_NEVER)
3350 goto bad_val;
3351 sbinfo->huge = huge;
3352#endif
3353#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003354 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003355 mpol_put(mpol);
3356 mpol = NULL;
3357 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003358 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003359#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 } else {
Joe Perches11705322016-03-17 14:19:50 -07003361 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003362 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 }
3364 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003365 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 return 0;
3367
3368bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003369 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003371error:
3372 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 return 1;
3374
3375}
3376
3377static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3378{
3379 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003380 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003381 unsigned long inodes;
3382 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
Greg Thelen5f001102013-02-22 16:36:01 -08003384 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003385 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003386 return error;
3387
3388 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003389 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003390 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003391 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003392 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003393 goto out;
3394 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003395 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003396 * but we must separately disallow unlimited->limited, because
3397 * in that case we have no record of how much is already in use.
3398 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003399 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003400 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003401 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003402 goto out;
3403
3404 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003405 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003406 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003407 sbinfo->max_inodes = config.max_inodes;
3408 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003409
Greg Thelen5f001102013-02-22 16:36:01 -08003410 /*
3411 * Preserve previous mempolicy unless mpol remount option was specified.
3412 */
3413 if (config.mpol) {
3414 mpol_put(sbinfo->mpol);
3415 sbinfo->mpol = config.mpol; /* transfers initial ref */
3416 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003417out:
3418 spin_unlock(&sbinfo->stat_lock);
3419 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003421
Al Viro34c80b12011-12-08 21:32:45 -05003422static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003423{
Al Viro34c80b12011-12-08 21:32:45 -05003424 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003425
3426 if (sbinfo->max_blocks != shmem_default_max_blocks())
3427 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003428 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003429 if (sbinfo->max_inodes != shmem_default_max_inodes())
3430 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3431 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003432 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003433 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3434 seq_printf(seq, ",uid=%u",
3435 from_kuid_munged(&init_user_ns, sbinfo->uid));
3436 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3437 seq_printf(seq, ",gid=%u",
3438 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003439#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003440 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3441 if (sbinfo->huge)
3442 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3443#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003444 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003445 return 0;
3446}
David Herrmann9183df22014-08-08 14:25:29 -07003447
3448#define MFD_NAME_PREFIX "memfd:"
3449#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3450#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3451
3452#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3453
3454SYSCALL_DEFINE2(memfd_create,
3455 const char __user *, uname,
3456 unsigned int, flags)
3457{
3458 struct shmem_inode_info *info;
3459 struct file *file;
3460 int fd, error;
3461 char *name;
3462 long len;
3463
3464 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3465 return -EINVAL;
3466
3467 /* length includes terminating zero */
3468 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3469 if (len <= 0)
3470 return -EFAULT;
3471 if (len > MFD_NAME_MAX_LEN + 1)
3472 return -EINVAL;
3473
3474 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3475 if (!name)
3476 return -ENOMEM;
3477
3478 strcpy(name, MFD_NAME_PREFIX);
3479 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3480 error = -EFAULT;
3481 goto err_name;
3482 }
3483
3484 /* terminating-zero may have changed after strnlen_user() returned */
3485 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3486 error = -EFAULT;
3487 goto err_name;
3488 }
3489
3490 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3491 if (fd < 0) {
3492 error = fd;
3493 goto err_name;
3494 }
3495
3496 file = shmem_file_setup(name, 0, VM_NORESERVE);
3497 if (IS_ERR(file)) {
3498 error = PTR_ERR(file);
3499 goto err_fd;
3500 }
3501 info = SHMEM_I(file_inode(file));
3502 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3503 file->f_flags |= O_RDWR | O_LARGEFILE;
3504 if (flags & MFD_ALLOW_SEALING)
3505 info->seals &= ~F_SEAL_SEAL;
3506
3507 fd_install(fd, file);
3508 kfree(name);
3509 return fd;
3510
3511err_fd:
3512 put_unused_fd(fd);
3513err_name:
3514 kfree(name);
3515 return error;
3516}
3517
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003518#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
3520static void shmem_put_super(struct super_block *sb)
3521{
Hugh Dickins602586a2010-08-17 15:23:56 -07003522 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3523
3524 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003525 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003526 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 sb->s_fs_info = NULL;
3528}
3529
Kay Sievers2b2af542009-04-30 15:23:42 +02003530int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531{
3532 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003533 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003534 int err = -ENOMEM;
3535
3536 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003537 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003538 L1_CACHE_BYTES), GFP_KERNEL);
3539 if (!sbinfo)
3540 return -ENOMEM;
3541
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003542 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003543 sbinfo->uid = current_fsuid();
3544 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003545 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003547#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 /*
3549 * Per default we only allow half of the physical ram per
3550 * tmpfs instance, limiting inodes to one per page of lowmem;
3551 * but the internal instance is left unlimited.
3552 */
Al Viroca4e0512013-08-31 12:57:10 -04003553 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003554 sbinfo->max_blocks = shmem_default_max_blocks();
3555 sbinfo->max_inodes = shmem_default_max_inodes();
3556 if (shmem_parse_options(data, sbinfo, false)) {
3557 err = -EINVAL;
3558 goto failed;
3559 }
Al Viroca4e0512013-08-31 12:57:10 -04003560 } else {
3561 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 }
David M. Grimes91828a42006-10-17 00:09:45 -07003563 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003564 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565#else
3566 sb->s_flags |= MS_NOUSER;
3567#endif
3568
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003569 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003570 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003571 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003572 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003573 spin_lock_init(&sbinfo->shrinklist_lock);
3574 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003575
Hugh Dickins285b2c42011-08-03 16:21:20 -07003576 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003577 sb->s_blocksize = PAGE_SIZE;
3578 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 sb->s_magic = TMPFS_MAGIC;
3580 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003581 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003582#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003583 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003584#endif
3585#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003586 sb->s_flags |= MS_POSIXACL;
3587#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003588
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003589 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 if (!inode)
3591 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003592 inode->i_uid = sbinfo->uid;
3593 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003594 sb->s_root = d_make_root(inode);
3595 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003596 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597 return 0;
3598
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599failed:
3600 shmem_put_super(sb);
3601 return err;
3602}
3603
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003604static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605
3606static struct inode *shmem_alloc_inode(struct super_block *sb)
3607{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003608 struct shmem_inode_info *info;
3609 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3610 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003612 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613}
3614
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003615static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003616{
3617 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003618 if (S_ISLNK(inode->i_mode))
3619 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003620 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3621}
3622
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623static void shmem_destroy_inode(struct inode *inode)
3624{
Al Viro09208d12011-07-26 03:15:03 -04003625 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003627 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628}
3629
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003630static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003632 struct shmem_inode_info *info = foo;
3633 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634}
3635
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003636static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637{
3638 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3639 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003640 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 return 0;
3642}
3643
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003644static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003646 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647}
3648
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003649static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003651 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003653 .write_begin = shmem_write_begin,
3654 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003656#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003657 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003658#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003659 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660};
3661
Helge Deller15ad7cd2006-12-06 20:40:36 -08003662static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003664 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003666 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003667 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003668 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003669 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003670 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003671 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003672 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673#endif
3674};
3675
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003676static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003677 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003678 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003679#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003680 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003681 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003682#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683};
3684
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003685static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686#ifdef CONFIG_TMPFS
3687 .create = shmem_create,
3688 .lookup = simple_lookup,
3689 .link = shmem_link,
3690 .unlink = shmem_unlink,
3691 .symlink = shmem_symlink,
3692 .mkdir = shmem_mkdir,
3693 .rmdir = shmem_rmdir,
3694 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003695 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003696 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003698#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003699 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003700#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003701#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003702 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003703 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003704#endif
3705};
3706
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003707static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003708#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003709 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003710#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003711#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003712 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003713 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003714#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715};
3716
Hugh Dickins759b9772007-03-05 00:30:28 -08003717static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 .alloc_inode = shmem_alloc_inode,
3719 .destroy_inode = shmem_destroy_inode,
3720#ifdef CONFIG_TMPFS
3721 .statfs = shmem_statfs,
3722 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003723 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724#endif
Al Viro1f895f72010-06-05 19:10:41 -04003725 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 .drop_inode = generic_delete_inode,
3727 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003728#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3729 .nr_cached_objects = shmem_unused_huge_count,
3730 .free_cached_objects = shmem_unused_huge_scan,
3731#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732};
3733
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003734static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003735 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003736 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737#ifdef CONFIG_NUMA
3738 .set_policy = shmem_set_policy,
3739 .get_policy = shmem_get_policy,
3740#endif
3741};
3742
Al Viro3c26ff62010-07-25 11:46:36 +04003743static struct dentry *shmem_mount(struct file_system_type *fs_type,
3744 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745{
Al Viro3c26ff62010-07-25 11:46:36 +04003746 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747}
3748
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003749static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750 .owner = THIS_MODULE,
3751 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003752 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003754 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003757int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758{
3759 int error;
3760
Rob Landley16203a72013-09-11 14:26:12 -07003761 /* If rootfs called this, don't re-init */
3762 if (shmem_inode_cachep)
3763 return 0;
3764
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003765 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 if (error)
3767 goto out3;
3768
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003769 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003771 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772 goto out2;
3773 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003774
Al Viroca4e0512013-08-31 12:57:10 -04003775 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776 if (IS_ERR(shm_mnt)) {
3777 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003778 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 goto out1;
3780 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003781
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003782#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003783 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3784 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3785 else
3786 shmem_huge = 0; /* just in case it was patched */
3787#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 return 0;
3789
3790out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003791 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003793 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794out3:
3795 shm_mnt = ERR_PTR(error);
3796 return error;
3797}
Matt Mackall853ac432009-01-06 14:40:20 -08003798
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003799#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003800static ssize_t shmem_enabled_show(struct kobject *kobj,
3801 struct kobj_attribute *attr, char *buf)
3802{
3803 int values[] = {
3804 SHMEM_HUGE_ALWAYS,
3805 SHMEM_HUGE_WITHIN_SIZE,
3806 SHMEM_HUGE_ADVISE,
3807 SHMEM_HUGE_NEVER,
3808 SHMEM_HUGE_DENY,
3809 SHMEM_HUGE_FORCE,
3810 };
3811 int i, count;
3812
3813 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3814 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3815
3816 count += sprintf(buf + count, fmt,
3817 shmem_format_huge(values[i]));
3818 }
3819 buf[count - 1] = '\n';
3820 return count;
3821}
3822
3823static ssize_t shmem_enabled_store(struct kobject *kobj,
3824 struct kobj_attribute *attr, const char *buf, size_t count)
3825{
3826 char tmp[16];
3827 int huge;
3828
3829 if (count + 1 > sizeof(tmp))
3830 return -EINVAL;
3831 memcpy(tmp, buf, count);
3832 tmp[count] = '\0';
3833 if (count && tmp[count - 1] == '\n')
3834 tmp[count - 1] = '\0';
3835
3836 huge = shmem_parse_huge(tmp);
3837 if (huge == -EINVAL)
3838 return -EINVAL;
3839 if (!has_transparent_hugepage() &&
3840 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3841 return -EINVAL;
3842
3843 shmem_huge = huge;
3844 if (shmem_huge < SHMEM_HUGE_DENY)
3845 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3846 return count;
3847}
3848
3849struct kobj_attribute shmem_enabled_attr =
3850 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003851#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003852
Arnd Bergmann3b337192016-08-10 16:27:44 -07003853#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003854bool shmem_huge_enabled(struct vm_area_struct *vma)
3855{
3856 struct inode *inode = file_inode(vma->vm_file);
3857 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3858 loff_t i_size;
3859 pgoff_t off;
3860
3861 if (shmem_huge == SHMEM_HUGE_FORCE)
3862 return true;
3863 if (shmem_huge == SHMEM_HUGE_DENY)
3864 return false;
3865 switch (sbinfo->huge) {
3866 case SHMEM_HUGE_NEVER:
3867 return false;
3868 case SHMEM_HUGE_ALWAYS:
3869 return true;
3870 case SHMEM_HUGE_WITHIN_SIZE:
3871 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3872 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3873 if (i_size >= HPAGE_PMD_SIZE &&
3874 i_size >> PAGE_SHIFT >= off)
3875 return true;
3876 case SHMEM_HUGE_ADVISE:
3877 /* TODO: implement fadvise() hints */
3878 return (vma->vm_flags & VM_HUGEPAGE);
3879 default:
3880 VM_BUG_ON(1);
3881 return false;
3882 }
3883}
Arnd Bergmann3b337192016-08-10 16:27:44 -07003884#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003885
Matt Mackall853ac432009-01-06 14:40:20 -08003886#else /* !CONFIG_SHMEM */
3887
3888/*
3889 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3890 *
3891 * This is intended for small system where the benefits of the full
3892 * shmem code (swap-backed and resource-limited) are outweighed by
3893 * their complexity. On systems without swap this code should be
3894 * effectively equivalent, but much lighter weight.
3895 */
3896
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003897static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003898 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003899 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003900 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003901 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003902};
3903
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003904int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003905{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003906 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003907
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003908 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003909 BUG_ON(IS_ERR(shm_mnt));
3910
3911 return 0;
3912}
3913
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003914int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003915{
3916 return 0;
3917}
3918
Hugh Dickins3f96b792009-09-21 17:03:37 -07003919int shmem_lock(struct file *file, int lock, struct user_struct *user)
3920{
3921 return 0;
3922}
3923
Hugh Dickins24513262012-01-20 14:34:21 -08003924void shmem_unlock_mapping(struct address_space *mapping)
3925{
3926}
3927
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003928#ifdef CONFIG_MMU
3929unsigned long shmem_get_unmapped_area(struct file *file,
3930 unsigned long addr, unsigned long len,
3931 unsigned long pgoff, unsigned long flags)
3932{
3933 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3934}
3935#endif
3936
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003937void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003938{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003939 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003940}
3941EXPORT_SYMBOL_GPL(shmem_truncate_range);
3942
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003943#define shmem_vm_ops generic_file_vm_ops
3944#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003945#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003946#define shmem_acct_size(flags, size) 0
3947#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08003948
3949#endif /* CONFIG_SHMEM */
3950
3951/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952
Rasmus Villemoes19938e32016-10-07 17:01:01 -07003953static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04003954 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05003955};
3956
Eric Parisc7277092013-12-02 11:24:19 +00003957static struct file *__shmem_file_setup(const char *name, loff_t size,
3958 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959{
Al Viro6b4d0b22013-02-14 21:37:26 -05003960 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04003962 struct path path;
Al Viro34515382013-02-14 22:38:02 -05003963 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964 struct qstr this;
3965
3966 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05003967 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
Hugh Dickins285b2c42011-08-03 16:21:20 -07003969 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 return ERR_PTR(-EINVAL);
3971
3972 if (shmem_acct_size(flags, size))
3973 return ERR_PTR(-ENOMEM);
3974
Al Viro6b4d0b22013-02-14 21:37:26 -05003975 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 this.name = name;
3977 this.len = strlen(name);
3978 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05003979 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003980 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05003981 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04003982 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05003984 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985
Al Viro6b4d0b22013-02-14 21:37:26 -05003986 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05003987 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003988 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003989 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990
Eric Parisc7277092013-12-02 11:24:19 +00003991 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04003992 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003994 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05003995 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3996 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07003997 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04003998
Al Viro6b4d0b22013-02-14 21:37:26 -05003999 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004000 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004001 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004002 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004003
Al Viro6b4d0b22013-02-14 21:37:26 -05004004 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006put_memory:
4007 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004008put_path:
4009 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004010 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011}
Eric Parisc7277092013-12-02 11:24:19 +00004012
4013/**
4014 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4015 * kernel internal. There will be NO LSM permission checks against the
4016 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004017 * higher layer. The users are the big_key and shm implementations. LSM
4018 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004019 * @name: name for dentry (to be seen in /proc/<pid>/maps
4020 * @size: size to be set for the file
4021 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4022 */
4023struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4024{
4025 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4026}
4027
4028/**
4029 * shmem_file_setup - get an unlinked file living in tmpfs
4030 * @name: name for dentry (to be seen in /proc/<pid>/maps
4031 * @size: size to be set for the file
4032 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4033 */
4034struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4035{
4036 return __shmem_file_setup(name, size, flags, 0);
4037}
Keith Packard395e0dd2008-06-20 00:08:06 -07004038EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039
Randy Dunlap46711812008-03-19 17:00:41 -07004040/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4043 */
4044int shmem_zero_setup(struct vm_area_struct *vma)
4045{
4046 struct file *file;
4047 loff_t size = vma->vm_end - vma->vm_start;
4048
Hugh Dickins66fc1302015-06-14 09:48:09 -07004049 /*
4050 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4051 * between XFS directory reading and selinux: since this file is only
4052 * accessible to the user through its mapping, use S_PRIVATE flag to
4053 * bypass file security, in the same way as shmem_kernel_file_setup().
4054 */
4055 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 if (IS_ERR(file))
4057 return PTR_ERR(file);
4058
4059 if (vma->vm_file)
4060 fput(vma->vm_file);
4061 vma->vm_file = file;
4062 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004063
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004064 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004065 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4066 (vma->vm_end & HPAGE_PMD_MASK)) {
4067 khugepaged_enter(vma, vma->vm_flags);
4068 }
4069
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 return 0;
4071}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004072
4073/**
4074 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4075 * @mapping: the page's address_space
4076 * @index: the page index
4077 * @gfp: the page allocator flags to use if allocating
4078 *
4079 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4080 * with any new page allocations done using the specified allocation flags.
4081 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4082 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4083 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4084 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004085 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4086 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004087 */
4088struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4089 pgoff_t index, gfp_t gfp)
4090{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004091#ifdef CONFIG_SHMEM
4092 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004093 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004094 int error;
4095
4096 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004097 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4098 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004099 if (error)
4100 page = ERR_PTR(error);
4101 else
4102 unlock_page(page);
4103 return page;
4104#else
4105 /*
4106 * The tiny !SHMEM case uses ramfs without swap
4107 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004108 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004109#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004110}
4111EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);