blob: ed9047d1afa19087695b4f577bdd0137a3f83110 [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
Mike Rapoport9815b0f2017-09-06 16:22:59 -0700184static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
185{
186 struct shmem_inode_info *info = SHMEM_I(inode);
187 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
188
189 if (shmem_acct_block(info->flags, pages))
190 return false;
191
192 if (sbinfo->max_blocks) {
193 if (percpu_counter_compare(&sbinfo->used_blocks,
194 sbinfo->max_blocks - pages) > 0)
195 goto unacct;
196 percpu_counter_add(&sbinfo->used_blocks, pages);
197 }
198
199 return true;
200
201unacct:
202 shmem_unacct_blocks(info->flags, pages);
203 return false;
204}
205
206static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
207{
208 struct shmem_inode_info *info = SHMEM_I(inode);
209 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
210
211 if (sbinfo->max_blocks)
212 percpu_counter_sub(&sbinfo->used_blocks, pages);
213 shmem_unacct_blocks(info->flags, pages);
214}
215
Hugh Dickins759b9772007-03-05 00:30:28 -0800216static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700217static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800218static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800219static const struct inode_operations shmem_inode_operations;
220static const struct inode_operations shmem_dir_inode_operations;
221static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400222static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700223static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800226static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800228static int shmem_reserve_inode(struct super_block *sb)
229{
230 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
231 if (sbinfo->max_inodes) {
232 spin_lock(&sbinfo->stat_lock);
233 if (!sbinfo->free_inodes) {
234 spin_unlock(&sbinfo->stat_lock);
235 return -ENOSPC;
236 }
237 sbinfo->free_inodes--;
238 spin_unlock(&sbinfo->stat_lock);
239 }
240 return 0;
241}
242
243static void shmem_free_inode(struct super_block *sb)
244{
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 sbinfo->free_inodes++;
249 spin_unlock(&sbinfo->stat_lock);
250 }
251}
252
Randy Dunlap46711812008-03-19 17:00:41 -0700253/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700254 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 * @inode: inode to recalc
256 *
257 * We have to calculate the free blocks since the mm can drop
258 * undirtied hole pages behind our back.
259 *
260 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
261 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
262 *
263 * It has to be called with the spinlock held.
264 */
265static void shmem_recalc_inode(struct inode *inode)
266{
267 struct shmem_inode_info *info = SHMEM_I(inode);
268 long freed;
269
270 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
271 if (freed > 0) {
272 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700273 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Mike Rapoport9815b0f2017-09-06 16:22:59 -0700274 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 }
276}
277
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700278bool shmem_charge(struct inode *inode, long pages)
279{
280 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700281 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700282
Mike Rapoport9815b0f2017-09-06 16:22:59 -0700283 if (!shmem_inode_acct_block(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700284 return false;
Mike Rapoportcae7ed22017-09-06 16:22:56 -0700285
Hugh Dickins0dba3e52018-11-30 14:10:29 -0800286 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
287 inode->i_mapping->nrpages += pages;
288
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700289 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700290 info->alloced += pages;
291 inode->i_blocks += pages * BLOCKS_PER_PAGE;
292 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700293 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700294
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700295 return true;
296}
297
298void shmem_uncharge(struct inode *inode, long pages)
299{
300 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700301 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700302
Hugh Dickins0dba3e52018-11-30 14:10:29 -0800303 /* nrpages adjustment done by __delete_from_page_cache() or caller */
304
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700305 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700306 info->alloced -= pages;
307 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
308 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700309 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700310
Mike Rapoport9815b0f2017-09-06 16:22:59 -0700311 shmem_inode_unacct_blocks(inode, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700312}
313
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700314/*
315 * Replace item expected in radix tree by a new item, while holding tree lock.
316 */
317static int shmem_radix_tree_replace(struct address_space *mapping,
318 pgoff_t index, void *expected, void *replacement)
319{
320 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700321 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700322
323 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700324 VM_BUG_ON(!replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700325 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700326 if (!pslot)
327 return -ENOENT;
328 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700329 if (item != expected)
330 return -ENOENT;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700331 radix_tree_replace_slot(pslot, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700332 return 0;
333}
334
335/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700336 * Sometimes, before we decide whether to proceed or to fail, we must check
337 * that an entry was not already brought back from swap by a racing thread.
338 *
339 * Checking page is not enough: by the time a SwapCache page is locked, it
340 * might be reused, and again be SwapCache, using the same swap as before.
341 */
342static bool shmem_confirm_swap(struct address_space *mapping,
343 pgoff_t index, swp_entry_t swap)
344{
345 void *item;
346
347 rcu_read_lock();
348 item = radix_tree_lookup(&mapping->page_tree, index);
349 rcu_read_unlock();
350 return item == swp_to_radix_entry(swap);
351}
352
353/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700354 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
355 *
356 * SHMEM_HUGE_NEVER:
357 * disables huge pages for the mount;
358 * SHMEM_HUGE_ALWAYS:
359 * enables huge pages for the mount;
360 * SHMEM_HUGE_WITHIN_SIZE:
361 * only allocate huge pages if the page will be fully within i_size,
362 * also respect fadvise()/madvise() hints;
363 * SHMEM_HUGE_ADVISE:
364 * only allocate huge pages if requested with fadvise()/madvise();
365 */
366
367#define SHMEM_HUGE_NEVER 0
368#define SHMEM_HUGE_ALWAYS 1
369#define SHMEM_HUGE_WITHIN_SIZE 2
370#define SHMEM_HUGE_ADVISE 3
371
372/*
373 * Special values.
374 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
375 *
376 * SHMEM_HUGE_DENY:
377 * disables huge on shm_mnt and all mounts, for emergency use;
378 * SHMEM_HUGE_FORCE:
379 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
380 *
381 */
382#define SHMEM_HUGE_DENY (-1)
383#define SHMEM_HUGE_FORCE (-2)
384
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700385#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700386/* ifdef here to avoid bloating shmem.o when not necessary */
387
388int shmem_huge __read_mostly;
389
Jérémy Lefaure8bec83b2016-12-12 16:43:23 -0800390#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700391static int shmem_parse_huge(const char *str)
392{
393 if (!strcmp(str, "never"))
394 return SHMEM_HUGE_NEVER;
395 if (!strcmp(str, "always"))
396 return SHMEM_HUGE_ALWAYS;
397 if (!strcmp(str, "within_size"))
398 return SHMEM_HUGE_WITHIN_SIZE;
399 if (!strcmp(str, "advise"))
400 return SHMEM_HUGE_ADVISE;
401 if (!strcmp(str, "deny"))
402 return SHMEM_HUGE_DENY;
403 if (!strcmp(str, "force"))
404 return SHMEM_HUGE_FORCE;
405 return -EINVAL;
406}
407
408static const char *shmem_format_huge(int huge)
409{
410 switch (huge) {
411 case SHMEM_HUGE_NEVER:
412 return "never";
413 case SHMEM_HUGE_ALWAYS:
414 return "always";
415 case SHMEM_HUGE_WITHIN_SIZE:
416 return "within_size";
417 case SHMEM_HUGE_ADVISE:
418 return "advise";
419 case SHMEM_HUGE_DENY:
420 return "deny";
421 case SHMEM_HUGE_FORCE:
422 return "force";
423 default:
424 VM_BUG_ON(1);
425 return "bad_val";
426 }
427}
Jérémy Lefaure8bec83b2016-12-12 16:43:23 -0800428#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700429
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700430static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
431 struct shrink_control *sc, unsigned long nr_to_split)
432{
433 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400434 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700435 struct inode *inode;
436 struct shmem_inode_info *info;
437 struct page *page;
438 unsigned long batch = sc ? sc->nr_to_scan : 128;
439 int removed = 0, split = 0;
440
441 if (list_empty(&sbinfo->shrinklist))
442 return SHRINK_STOP;
443
444 spin_lock(&sbinfo->shrinklist_lock);
445 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
446 info = list_entry(pos, struct shmem_inode_info, shrinklist);
447
448 /* pin the inode */
449 inode = igrab(&info->vfs_inode);
450
451 /* inode is about to be evicted */
452 if (!inode) {
453 list_del_init(&info->shrinklist);
454 removed++;
455 goto next;
456 }
457
458 /* Check if there's anything to gain */
459 if (round_up(inode->i_size, PAGE_SIZE) ==
460 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400461 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700462 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700463 goto next;
464 }
465
466 list_move(&info->shrinklist, &list);
467next:
468 if (!--batch)
469 break;
470 }
471 spin_unlock(&sbinfo->shrinklist_lock);
472
Kirill A. Shutemov8936b742017-05-23 21:53:56 -0400473 list_for_each_safe(pos, next, &to_remove) {
474 info = list_entry(pos, struct shmem_inode_info, shrinklist);
475 inode = &info->vfs_inode;
476 list_del_init(&info->shrinklist);
477 iput(inode);
478 }
479
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700480 list_for_each_safe(pos, next, &list) {
481 int ret;
482
483 info = list_entry(pos, struct shmem_inode_info, shrinklist);
484 inode = &info->vfs_inode;
485
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700486 if (nr_to_split && split >= nr_to_split)
487 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700488
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700489 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700490 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
491 if (!page)
492 goto drop;
493
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700494 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700495 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700496 put_page(page);
497 goto drop;
498 }
499
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700500 /*
501 * Leave the inode on the list if we failed to lock
502 * the page at this time.
503 *
504 * Waiting for the lock may lead to deadlock in the
505 * reclaim path.
506 */
507 if (!trylock_page(page)) {
508 put_page(page);
509 goto leave;
510 }
511
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700512 ret = split_huge_page(page);
513 unlock_page(page);
514 put_page(page);
515
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700516 /* If split failed leave the inode on the list */
517 if (ret)
518 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700519
520 split++;
521drop:
522 list_del_init(&info->shrinklist);
523 removed++;
Kirill A. Shutemov8ab89952018-03-22 16:17:35 -0700524leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700525 iput(inode);
526 }
527
528 spin_lock(&sbinfo->shrinklist_lock);
529 list_splice_tail(&list, &sbinfo->shrinklist);
530 sbinfo->shrinklist_len -= removed;
531 spin_unlock(&sbinfo->shrinklist_lock);
532
533 return split;
534}
535
536static long shmem_unused_huge_scan(struct super_block *sb,
537 struct shrink_control *sc)
538{
539 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
540
541 if (!READ_ONCE(sbinfo->shrinklist_len))
542 return SHRINK_STOP;
543
544 return shmem_unused_huge_shrink(sbinfo, sc, 0);
545}
546
547static long shmem_unused_huge_count(struct super_block *sb,
548 struct shrink_control *sc)
549{
550 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
551 return READ_ONCE(sbinfo->shrinklist_len);
552}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700553#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700554
555#define shmem_huge SHMEM_HUGE_DENY
556
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700557static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
558 struct shrink_control *sc, unsigned long nr_to_split)
559{
560 return 0;
561}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700562#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700563
564/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700565 * Like add_to_page_cache_locked, but error if expected item has gone.
566 */
567static int shmem_add_to_page_cache(struct page *page,
568 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700569 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700570{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700571 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700572
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700573 VM_BUG_ON_PAGE(PageTail(page), page);
574 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800575 VM_BUG_ON_PAGE(!PageLocked(page), page);
576 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700577 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700578
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700579 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700580 page->mapping = mapping;
581 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700582
Hugh Dickinsb065b432012-07-11 14:02:48 -0700583 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700584 if (PageTransHuge(page)) {
585 void __rcu **results;
586 pgoff_t idx;
587 int i;
588
589 error = 0;
590 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
591 &results, &idx, index, 1) &&
592 idx < index + HPAGE_PMD_NR) {
593 error = -EEXIST;
594 }
595
596 if (!error) {
597 for (i = 0; i < HPAGE_PMD_NR; i++) {
598 error = radix_tree_insert(&mapping->page_tree,
599 index + i, page + i);
600 VM_BUG_ON(error);
601 }
602 count_vm_event(THP_FILE_ALLOC);
603 }
604 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700605 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700606 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700607 error = shmem_radix_tree_replace(mapping, index, expected,
608 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700609 }
610
Hugh Dickinsb065b432012-07-11 14:02:48 -0700611 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700612 mapping->nrpages += nr;
613 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700614 __inc_node_page_state(page, NR_SHMEM_THPS);
615 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
616 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700617 spin_unlock_irq(&mapping->tree_lock);
618 } else {
619 page->mapping = NULL;
620 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700621 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700622 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700623 return error;
624}
625
626/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700627 * Like delete_from_page_cache, but substitutes swap for page.
628 */
629static void shmem_delete_from_page_cache(struct page *page, void *radswap)
630{
631 struct address_space *mapping = page->mapping;
632 int error;
633
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700634 VM_BUG_ON_PAGE(PageCompound(page), page);
635
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700636 spin_lock_irq(&mapping->tree_lock);
637 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
638 page->mapping = NULL;
639 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700640 __dec_node_page_state(page, NR_FILE_PAGES);
641 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700642 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300643 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700644 BUG_ON(error);
645}
646
647/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700648 * Remove swap entry from radix tree, free the swap and its page cache.
649 */
650static int shmem_free_swap(struct address_space *mapping,
651 pgoff_t index, void *radswap)
652{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700653 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700654
655 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700656 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700657 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700658 if (old != radswap)
659 return -ENOENT;
660 free_swap_and_cache(radix_to_swp_entry(radswap));
661 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700662}
663
664/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800665 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800666 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800667 *
668 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
669 * as long as the inode doesn't go away and racy results are not a problem.
670 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800671unsigned long shmem_partial_swap_usage(struct address_space *mapping,
672 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800673{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800674 struct radix_tree_iter iter;
675 void **slot;
676 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800677 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800678
679 rcu_read_lock();
680
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800681 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
682 if (iter.index >= end)
683 break;
684
685 page = radix_tree_deref_slot(slot);
686
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700687 if (radix_tree_deref_retry(page)) {
688 slot = radix_tree_iter_retry(&iter);
689 continue;
690 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800691
692 if (radix_tree_exceptional_entry(page))
693 swapped++;
694
695 if (need_resched()) {
696 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -0700697 slot = radix_tree_iter_next(&iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800698 }
699 }
700
701 rcu_read_unlock();
702
703 return swapped << PAGE_SHIFT;
704}
705
706/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800707 * Determine (in bytes) how many of the shmem object's pages mapped by the
708 * given vma is swapped out.
709 *
710 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
711 * as long as the inode doesn't go away and racy results are not a problem.
712 */
713unsigned long shmem_swap_usage(struct vm_area_struct *vma)
714{
715 struct inode *inode = file_inode(vma->vm_file);
716 struct shmem_inode_info *info = SHMEM_I(inode);
717 struct address_space *mapping = inode->i_mapping;
718 unsigned long swapped;
719
720 /* Be careful as we don't hold info->lock */
721 swapped = READ_ONCE(info->swapped);
722
723 /*
724 * The easier cases are when the shmem object has nothing in swap, or
725 * the vma maps it whole. Then we can simply use the stats that we
726 * already track.
727 */
728 if (!swapped)
729 return 0;
730
731 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
732 return swapped << PAGE_SHIFT;
733
734 /* Here comes the more involved part */
735 return shmem_partial_swap_usage(mapping,
736 linear_page_index(vma, vma->vm_start),
737 linear_page_index(vma, vma->vm_end));
738}
739
740/*
Hugh Dickins24513262012-01-20 14:34:21 -0800741 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
742 */
743void shmem_unlock_mapping(struct address_space *mapping)
744{
745 struct pagevec pvec;
746 pgoff_t indices[PAGEVEC_SIZE];
747 pgoff_t index = 0;
748
749 pagevec_init(&pvec, 0);
750 /*
751 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
752 */
753 while (!mapping_unevictable(mapping)) {
754 /*
755 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
756 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
757 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700758 pvec.nr = find_get_entries(mapping, index,
759 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800760 if (!pvec.nr)
761 break;
762 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700763 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800764 check_move_unevictable_pages(pvec.pages, pvec.nr);
765 pagevec_release(&pvec);
766 cond_resched();
767 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700768}
769
770/*
771 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700772 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700773 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700774static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
775 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700777 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300779 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
780 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
781 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
782 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700783 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700784 pgoff_t indices[PAGEVEC_SIZE];
785 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700786 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700787 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700789 if (lend == -1)
790 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700791
792 pagevec_init(&pvec, 0);
793 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700794 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700795 pvec.nr = find_get_entries(mapping, index,
796 min(end - index, (pgoff_t)PAGEVEC_SIZE),
797 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700798 if (!pvec.nr)
799 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700800 for (i = 0; i < pagevec_count(&pvec); i++) {
801 struct page *page = pvec.pages[i];
802
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700803 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700804 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700805 break;
806
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700807 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700808 if (unfalloc)
809 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700810 nr_swaps_freed += !shmem_free_swap(mapping,
811 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700812 continue;
813 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700814
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700815 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
816
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700817 if (!trylock_page(page))
818 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700819
820 if (PageTransTail(page)) {
821 /* Middle of THP: zero out the page */
822 clear_highpage(page);
823 unlock_page(page);
824 continue;
825 } else if (PageTransHuge(page)) {
826 if (index == round_down(end, HPAGE_PMD_NR)) {
827 /*
828 * Range ends in the middle of THP:
829 * zero out the page
830 */
831 clear_highpage(page);
832 unlock_page(page);
833 continue;
834 }
835 index += HPAGE_PMD_NR - 1;
836 i += HPAGE_PMD_NR - 1;
837 }
838
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700839 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700840 VM_BUG_ON_PAGE(PageTail(page), page);
841 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800842 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700843 truncate_inode_page(mapping, page);
844 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700845 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700846 unlock_page(page);
847 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700848 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800849 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700850 cond_resched();
851 index++;
852 }
853
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700854 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700855 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700856 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700857 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300858 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700859 if (start > end) {
860 top = partial_end;
861 partial_end = 0;
862 }
863 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700864 set_page_dirty(page);
865 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300866 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700867 }
868 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700869 if (partial_end) {
870 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700871 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700872 if (page) {
873 zero_user_segment(page, 0, partial_end);
874 set_page_dirty(page);
875 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300876 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700877 }
878 }
879 if (start >= end)
880 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700881
882 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700883 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700884 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700885
886 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700887 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700888 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700889 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700890 /* If all gone or hole-punch or unfalloc, we're done */
891 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700892 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700893 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700894 index = start;
895 continue;
896 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700897 for (i = 0; i < pagevec_count(&pvec); i++) {
898 struct page *page = pvec.pages[i];
899
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700900 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700901 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700902 break;
903
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700904 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700905 if (unfalloc)
906 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700907 if (shmem_free_swap(mapping, index, page)) {
908 /* Swap was replaced by page: retry */
909 index--;
910 break;
911 }
912 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700913 continue;
914 }
915
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700916 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700917
918 if (PageTransTail(page)) {
919 /* Middle of THP: zero out the page */
920 clear_highpage(page);
921 unlock_page(page);
922 /*
923 * Partial thp truncate due 'start' in middle
924 * of THP: don't need to look on these pages
925 * again on !pvec.nr restart.
926 */
927 if (index != round_down(end, HPAGE_PMD_NR))
928 start++;
929 continue;
930 } else if (PageTransHuge(page)) {
931 if (index == round_down(end, HPAGE_PMD_NR)) {
932 /*
933 * Range ends in the middle of THP:
934 * zero out the page
935 */
936 clear_highpage(page);
937 unlock_page(page);
938 continue;
939 }
940 index += HPAGE_PMD_NR - 1;
941 i += HPAGE_PMD_NR - 1;
942 }
943
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700944 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700945 VM_BUG_ON_PAGE(PageTail(page), page);
946 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800947 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700948 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700949 } else {
950 /* Page was replaced by swap: retry */
951 unlock_page(page);
952 index--;
953 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700954 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700955 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700956 unlock_page(page);
957 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700958 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800959 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700960 index++;
961 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700962
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700963 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700964 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700966 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700967}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700969void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
970{
971 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700972 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700974EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Yu Zhao44a30222015-09-08 15:03:33 -0700976static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
977 struct kstat *stat)
978{
979 struct inode *inode = dentry->d_inode;
980 struct shmem_inode_info *info = SHMEM_I(inode);
981
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800982 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700983 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800984 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700985 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800986 }
Yu Zhao44a30222015-09-08 15:03:33 -0700987 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700988 return 0;
989}
990
Hugh Dickins94c1e622011-06-27 16:18:03 -0700991static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992{
David Howells75c3cfa2015-03-17 22:26:12 +0000993 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700994 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700995 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 int error;
997
Jan Kara31051c82016-05-26 16:55:18 +0200998 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200999 if (error)
1000 return error;
1001
Hugh Dickins94c1e622011-06-27 16:18:03 -07001002 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1003 loff_t oldsize = inode->i_size;
1004 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001005
David Herrmann40e041a2014-08-08 14:25:27 -07001006 /* protected by i_mutex */
1007 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1008 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1009 return -EPERM;
1010
Hugh Dickins94c1e622011-06-27 16:18:03 -07001011 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001012 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1013 oldsize, newsize);
1014 if (error)
1015 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001016 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001017 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001018 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001019 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001020 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001021 if (oldsize > holebegin)
1022 unmap_mapping_range(inode->i_mapping,
1023 holebegin, 0, 1);
1024 if (info->alloced)
1025 shmem_truncate_range(inode,
1026 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001027 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001028 if (oldsize > holebegin)
1029 unmap_mapping_range(inode->i_mapping,
1030 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001031
1032 /*
1033 * Part of the huge page can be beyond i_size: subject
1034 * to shrink under memory pressure.
1035 */
1036 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1037 spin_lock(&sbinfo->shrinklist_lock);
Cong Wange2286912017-08-10 15:24:24 -07001038 /*
1039 * _careful to defend against unlocked access to
1040 * ->shrink_list in shmem_unused_huge_shrink()
1041 */
1042 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001043 list_add_tail(&info->shrinklist,
1044 &sbinfo->shrinklist);
1045 sbinfo->shrinklist_len++;
1046 }
1047 spin_unlock(&sbinfo->shrinklist_lock);
1048 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
1051
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001052 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001053 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001054 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 return error;
1056}
1057
Al Viro1f895f72010-06-05 19:10:41 -04001058static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001061 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001063 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 shmem_unacct_size(info->flags, inode->i_size);
1065 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001066 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001067 if (!list_empty(&info->shrinklist)) {
1068 spin_lock(&sbinfo->shrinklist_lock);
1069 if (!list_empty(&info->shrinklist)) {
1070 list_del_init(&info->shrinklist);
1071 sbinfo->shrinklist_len--;
1072 }
1073 spin_unlock(&sbinfo->shrinklist_lock);
1074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001076 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001078 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 }
Al Viro3ed47db2016-01-22 18:08:52 -05001080 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001081
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001082 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001083 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001084 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001085 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086}
1087
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001088/*
1089 * If swap found in inode, free it and move page from swapcache to filecache.
1090 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001091static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001092 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001094 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001095 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001096 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001097 gfp_t gfp;
1098 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001100 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001101 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001102 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001103 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001104
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001105 /*
1106 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001107 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001108 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001109 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001110 */
1111 if (shmem_swaplist.next != &info->swaplist)
1112 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001113
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001114 gfp = mapping_gfp_mask(mapping);
1115 if (shmem_should_replace_page(*pagep, gfp)) {
1116 mutex_unlock(&shmem_swaplist_mutex);
1117 error = shmem_replace_page(pagep, gfp, info, index);
1118 mutex_lock(&shmem_swaplist_mutex);
1119 /*
1120 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001121 * allocation, but the inode might have been freed while we
1122 * dropped it: although a racing shmem_evict_inode() cannot
1123 * complete without emptying the radix_tree, our page lock
1124 * on this swapcache page is not enough to prevent that -
1125 * free_swap_and_cache() of our swap entry will only
1126 * trylock_page(), removing swap from radix_tree whatever.
1127 *
1128 * We must not proceed to shmem_add_to_page_cache() if the
1129 * inode has been freed, but of course we cannot rely on
1130 * inode or mapping or info to check that. However, we can
1131 * safely check if our swap entry is still in use (and here
1132 * it can't have got reused for another page): if it's still
1133 * in use, then the inode cannot have been freed yet, and we
1134 * can safely proceed (if it's no longer in use, that tells
1135 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001136 */
1137 if (!page_swapcount(*pagep))
1138 error = -ENOENT;
1139 }
1140
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001141 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001142 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1143 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1144 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001145 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001146 if (!error)
1147 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001148 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001149 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001150 /*
1151 * Truncation and eviction use free_swap_and_cache(), which
1152 * only does trylock page: if we raced, best clean up here.
1153 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001154 delete_from_swap_cache(*pagep);
1155 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001156 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001157 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001158 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001159 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001160 swap_free(swap);
1161 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001163 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}
1165
1166/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001167 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001169int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001171 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001173 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001174 int error = 0;
1175
1176 /*
1177 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001178 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001179 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001180 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001181 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001182
1183 /*
1184 * Charge page using GFP_KERNEL while we can wait, before taking
1185 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1186 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001187 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001188 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1189 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001190 if (error)
1191 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001192 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001193 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001195 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001196 list_for_each_safe(this, next, &shmem_swaplist) {
1197 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001198 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001199 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001200 else
1201 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001202 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001203 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001204 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001205 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001207 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001208
Johannes Weiner00501b52014-08-08 14:19:20 -07001209 if (error) {
1210 if (error != -ENOMEM)
1211 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001212 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001213 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001214 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001215out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001216 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001217 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001218 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
1221/*
1222 * Move the page from the page cache to the swap cache.
1223 */
1224static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1225{
1226 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001229 swp_entry_t swap;
1230 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001232 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 mapping = page->mapping;
1235 index = page->index;
1236 inode = mapping->host;
1237 info = SHMEM_I(inode);
1238 if (info->flags & VM_LOCKED)
1239 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001240 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 goto redirty;
1242
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001243 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001244 * Our capabilities prevent regular writeback or sync from ever calling
1245 * shmem_writepage; but a stacking filesystem might use ->writepage of
1246 * its underlying filesystem, in which case tmpfs should write out to
1247 * swap only in response to memory pressure, and not for the writeback
1248 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001249 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001250 if (!wbc->for_reclaim) {
1251 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1252 goto redirty;
1253 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001254
1255 /*
1256 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1257 * value into swapfile.c, the only way we can correctly account for a
1258 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001259 *
1260 * That's okay for a page already fallocated earlier, but if we have
1261 * not yet completed the fallocation, then (a) we want to keep track
1262 * of this page in case we have to undo it, and (b) it may not be a
1263 * good idea to continue anyway, once we're pushing into swap. So
1264 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001265 */
1266 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001267 if (inode->i_private) {
1268 struct shmem_falloc *shmem_falloc;
1269 spin_lock(&inode->i_lock);
1270 shmem_falloc = inode->i_private;
1271 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001272 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001273 index >= shmem_falloc->start &&
1274 index < shmem_falloc->next)
1275 shmem_falloc->nr_unswapped++;
1276 else
1277 shmem_falloc = NULL;
1278 spin_unlock(&inode->i_lock);
1279 if (shmem_falloc)
1280 goto redirty;
1281 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001282 clear_highpage(page);
1283 flush_dcache_page(page);
1284 SetPageUptodate(page);
1285 }
1286
Hugh Dickins48f170f2011-07-25 17:12:37 -07001287 swap = get_swap_page();
1288 if (!swap.val)
1289 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001290
Vladimir Davydov37e84352016-01-20 15:02:56 -08001291 if (mem_cgroup_try_charge_swap(page, swap))
1292 goto free_swap;
1293
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001294 /*
1295 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001296 * if it's not already there. Do it now before the page is
1297 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001298 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001299 * we've incremented swapped, because shmem_unuse_inode() will
1300 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001301 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001302 mutex_lock(&shmem_swaplist_mutex);
1303 if (list_empty(&info->swaplist))
1304 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001305
Hugh Dickins48f170f2011-07-25 17:12:37 -07001306 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001307 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001308 shmem_recalc_inode(inode);
1309 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001310 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001311
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001312 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001313 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1314
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001315 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001316 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001317 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 return 0;
1319 }
1320
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001321 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001322free_swap:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001323 swapcache_free(swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324redirty:
1325 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001326 if (wbc->for_reclaim)
1327 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1328 unlock_page(page);
1329 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330}
1331
Hugh Dickins75edd342016-05-19 17:12:44 -07001332#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001333static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001334{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001335 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001336
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001337 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001338 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001339
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001340 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001341
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001342 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001343}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001344
1345static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1346{
1347 struct mempolicy *mpol = NULL;
1348 if (sbinfo->mpol) {
1349 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1350 mpol = sbinfo->mpol;
1351 mpol_get(mpol);
1352 spin_unlock(&sbinfo->stat_lock);
1353 }
1354 return mpol;
1355}
Hugh Dickins75edd342016-05-19 17:12:44 -07001356#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1357static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1358{
1359}
1360static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1361{
1362 return NULL;
1363}
1364#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1365#ifndef CONFIG_NUMA
1366#define vm_policy vm_private_data
1367#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001368
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001369static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1370 struct shmem_inode_info *info, pgoff_t index)
1371{
1372 /* Create a pseudo vma that just contains the policy */
1373 vma->vm_start = 0;
1374 /* Bias interleave by inode number to distribute better across nodes */
1375 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1376 vma->vm_ops = NULL;
1377 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1378}
1379
1380static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1381{
1382 /* Drop reference taken by mpol_shared_policy_lookup() */
1383 mpol_cond_put(vma->vm_policy);
1384}
1385
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001386static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1387 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001390 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001392 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001393 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001394 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001395
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001396 return page;
1397}
Mel Gorman18a2f372012-12-05 14:01:41 -08001398
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001399static struct page *shmem_alloc_hugepage(gfp_t gfp,
1400 struct shmem_inode_info *info, pgoff_t index)
1401{
1402 struct vm_area_struct pvma;
1403 struct inode *inode = &info->vfs_inode;
1404 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001405 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001406 void __rcu **results;
1407 struct page *page;
1408
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001409 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001410 return NULL;
1411
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001412 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001413 rcu_read_lock();
1414 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1415 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1416 rcu_read_unlock();
1417 return NULL;
1418 }
1419 rcu_read_unlock();
1420
1421 shmem_pseudo_vma_init(&pvma, info, hindex);
1422 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1423 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1424 shmem_pseudo_vma_destroy(&pvma);
1425 if (page)
1426 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001427 return page;
1428}
1429
1430static struct page *shmem_alloc_page(gfp_t gfp,
1431 struct shmem_inode_info *info, pgoff_t index)
1432{
1433 struct vm_area_struct pvma;
1434 struct page *page;
1435
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001436 shmem_pseudo_vma_init(&pvma, info, index);
1437 page = alloc_page_vma(gfp, &pvma, 0);
1438 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001439
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001440 return page;
1441}
1442
1443static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001444 struct inode *inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001445 pgoff_t index, bool huge)
1446{
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001447 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001448 struct page *page;
1449 int nr;
1450 int err = -ENOSPC;
1451
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001452 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001453 huge = false;
1454 nr = huge ? HPAGE_PMD_NR : 1;
1455
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001456 if (!shmem_inode_acct_block(inode, nr))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001457 goto failed;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001458
1459 if (huge)
1460 page = shmem_alloc_hugepage(gfp, info, index);
1461 else
1462 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001463 if (page) {
1464 __SetPageLocked(page);
1465 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001466 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001467 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001468
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001469 err = -ENOMEM;
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001470 shmem_inode_unacct_blocks(inode, nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001471failed:
1472 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001476 * When a page is moved from swapcache to shmem filecache (either by the
1477 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1478 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1479 * ignorance of the mapping it belongs to. If that mapping has special
1480 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1481 * we may need to copy to a suitable page before moving to filecache.
1482 *
1483 * In a future release, this may well be extended to respect cpuset and
1484 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1485 * but for now it is a simple matter of zone.
1486 */
1487static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1488{
1489 return page_zonenum(page) > gfp_zone(gfp);
1490}
1491
1492static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1493 struct shmem_inode_info *info, pgoff_t index)
1494{
1495 struct page *oldpage, *newpage;
1496 struct address_space *swap_mapping;
Yu Zhao3c470632018-11-30 14:09:03 -08001497 swp_entry_t entry;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001498 pgoff_t swap_index;
1499 int error;
1500
1501 oldpage = *pagep;
Yu Zhao3c470632018-11-30 14:09:03 -08001502 entry.val = page_private(oldpage);
1503 swap_index = swp_offset(entry);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001504 swap_mapping = page_mapping(oldpage);
1505
1506 /*
1507 * We have arrived here because our zones are constrained, so don't
1508 * limit chance of success by further cpuset and node constraints.
1509 */
1510 gfp &= ~GFP_CONSTRAINT_MASK;
1511 newpage = shmem_alloc_page(gfp, info, index);
1512 if (!newpage)
1513 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001514
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001515 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001516 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001517 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001518
Hugh Dickins9956edf2016-11-10 10:46:11 -08001519 __SetPageLocked(newpage);
1520 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001521 SetPageUptodate(newpage);
Yu Zhao3c470632018-11-30 14:09:03 -08001522 set_page_private(newpage, entry.val);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001523 SetPageSwapCache(newpage);
1524
1525 /*
1526 * Our caller will very soon move newpage out of swapcache, but it's
1527 * a nice clean interface for us to replace oldpage by newpage there.
1528 */
1529 spin_lock_irq(&swap_mapping->tree_lock);
1530 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1531 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001532 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001533 __inc_node_page_state(newpage, NR_FILE_PAGES);
1534 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001535 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001536 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001537
Hugh Dickins0142ef62012-06-07 14:21:09 -07001538 if (unlikely(error)) {
1539 /*
1540 * Is this possible? I think not, now that our callers check
1541 * both PageSwapCache and page_private after getting page lock;
1542 * but be defensive. Reverse old to newpage for clear and free.
1543 */
1544 oldpage = newpage;
1545 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001546 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001547 lru_cache_add_anon(newpage);
1548 *pagep = newpage;
1549 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001550
1551 ClearPageSwapCache(oldpage);
1552 set_page_private(oldpage, 0);
1553
1554 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001555 put_page(oldpage);
1556 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001557 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001558}
1559
1560/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001561 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 *
1563 * If we allocate a new one we do not mark it dirty. That's up to the
1564 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001565 * entry since a page cannot live in both the swap and page cache.
1566 *
1567 * fault_mm and fault_type are only supplied by shmem_fault:
1568 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001570static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001571 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1572 struct mm_struct *fault_mm, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
1574 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann1dc68392016-12-12 16:42:28 -08001575 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001577 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001578 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001579 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001581 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001582 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001584 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001585 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001587 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001589 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1590 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001592 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001593 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001594 if (radix_tree_exceptional_entry(page)) {
1595 swap = radix_to_swp_entry(page);
1596 page = NULL;
1597 }
1598
Hugh Dickins75edd342016-05-19 17:12:44 -07001599 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001600 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001601 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001602 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001603 }
1604
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001605 if (page && sgp == SGP_WRITE)
1606 mark_page_accessed(page);
1607
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001608 /* fallocated page? */
1609 if (page && !PageUptodate(page)) {
1610 if (sgp != SGP_READ)
1611 goto clear;
1612 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001613 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001614 page = NULL;
1615 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001616 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001617 *pagep = page;
1618 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
1621 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001622 * Fast cache lookup did not find it:
1623 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001625 sbinfo = SHMEM_SB(inode->i_sb);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001626 charge_mm = fault_mm ? : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001627
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 if (swap.val) {
1629 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001630 page = lookup_swap_cache(swap);
1631 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001632 /* Or update major stats only when swapin succeeds?? */
1633 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001634 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001635 count_vm_event(PGMAJFAULT);
1636 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1637 }
1638 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001639 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001640 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001641 error = -ENOMEM;
1642 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 }
1645
1646 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001647 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001648 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001649 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001650 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001651 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001652 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001653 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001655 goto failed;
1656 }
1657 wait_on_page_writeback(page);
1658
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001659 if (shmem_should_replace_page(page, gfp)) {
1660 error = shmem_replace_page(&page, gfp, info, index);
1661 if (error)
1662 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
1664
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001665 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001666 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001667 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001668 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001669 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001670 /*
1671 * We already confirmed swap under page lock, and make
1672 * no memory allocation here, so usually no possibility
1673 * of error; but free_swap_and_cache() only trylocks a
1674 * page, so it is just possible that the entry has been
1675 * truncated or holepunched since swap was confirmed.
1676 * shmem_undo_range() will have done some of the
1677 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001678 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001679 * Reset swap.val? No, leave it so "failed" goes back to
1680 * "repeat": reading a hole and writing should succeed.
1681 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001682 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001683 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001684 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001685 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001686 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001687 if (error)
1688 goto failed;
1689
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001690 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001691
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001692 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001693 info->swapped--;
1694 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001695 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001696
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001697 if (sgp == SGP_WRITE)
1698 mark_page_accessed(page);
1699
Hugh Dickins27ab7002011-07-25 17:12:36 -07001700 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001701 set_page_dirty(page);
1702 swap_free(swap);
1703
Hugh Dickins54af60422011-08-03 16:21:24 -07001704 } else {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001705 /* shmem_symlink() */
1706 if (mapping->a_ops != &shmem_aops)
1707 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001708 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001709 goto alloc_nohuge;
1710 if (shmem_huge == SHMEM_HUGE_FORCE)
1711 goto alloc_huge;
1712 switch (sbinfo->huge) {
1713 loff_t i_size;
1714 pgoff_t off;
1715 case SHMEM_HUGE_NEVER:
1716 goto alloc_nohuge;
1717 case SHMEM_HUGE_WITHIN_SIZE:
1718 off = round_up(index, HPAGE_PMD_NR);
1719 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1720 if (i_size >= HPAGE_PMD_SIZE &&
1721 i_size >> PAGE_SHIFT >= off)
1722 goto alloc_huge;
1723 /* fallthrough */
1724 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001725 if (sgp_huge == SGP_HUGE)
1726 goto alloc_huge;
1727 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001728 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001731alloc_huge:
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001732 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001733 if (IS_ERR(page)) {
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001734alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001735 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001736 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001737 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001738 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001739 error = PTR_ERR(page);
1740 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001741 if (error != -ENOSPC)
1742 goto failed;
1743 /*
1744 * Try to reclaim some spece by splitting a huge page
1745 * beyond i_size on the filesystem.
1746 */
1747 while (retry--) {
1748 int ret;
1749 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1750 if (ret == SHRINK_STOP)
1751 break;
1752 if (ret)
1753 goto alloc_nohuge;
1754 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001755 goto failed;
1756 }
1757
1758 if (PageTransHuge(page))
1759 hindex = round_down(index, HPAGE_PMD_NR);
1760 else
1761 hindex = index;
1762
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001763 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001764 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001765
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001766 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001767 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001768 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001769 goto unacct;
1770 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1771 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001772 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001773 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001774 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001775 radix_tree_preload_end();
1776 }
1777 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001778 mem_cgroup_cancel_charge(page, memcg,
1779 PageTransHuge(page));
1780 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001781 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001782 mem_cgroup_commit_charge(page, memcg, false,
1783 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001784 lru_cache_add_anon(page);
1785
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001786 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001787 info->alloced += 1 << compound_order(page);
1788 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001789 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001790 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001791 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001792
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001793 if (PageTransHuge(page) &&
1794 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1795 hindex + HPAGE_PMD_NR - 1) {
1796 /*
1797 * Part of the huge page is beyond i_size: subject
1798 * to shrink under memory pressure.
1799 */
1800 spin_lock(&sbinfo->shrinklist_lock);
Cong Wange2286912017-08-10 15:24:24 -07001801 /*
1802 * _careful to defend against unlocked access to
1803 * ->shrink_list in shmem_unused_huge_shrink()
1804 */
1805 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001806 list_add_tail(&info->shrinklist,
1807 &sbinfo->shrinklist);
1808 sbinfo->shrinklist_len++;
1809 }
1810 spin_unlock(&sbinfo->shrinklist_lock);
1811 }
1812
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001813 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001814 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1815 */
1816 if (sgp == SGP_FALLOC)
1817 sgp = SGP_WRITE;
1818clear:
1819 /*
1820 * Let SGP_WRITE caller clear ends if write does not fill page;
1821 * but SGP_FALLOC on a page fallocated earlier must initialize
1822 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001823 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001824 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1825 struct page *head = compound_head(page);
1826 int i;
1827
1828 for (i = 0; i < (1 << compound_order(head)); i++) {
1829 clear_highpage(head + i);
1830 flush_dcache_page(head + i);
1831 }
1832 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001833 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001834 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001835
Hugh Dickins54af60422011-08-03 16:21:24 -07001836 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001837 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001838 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001839 if (alloced) {
1840 ClearPageDirty(page);
1841 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001842 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001843 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001844 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001845 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001846 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001847 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001848 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001849 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001850 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001851
Nick Piggind0217ac2007-07-19 01:47:03 -07001852 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001853 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001855unacct:
Mike Rapoport9815b0f2017-09-06 16:22:59 -07001856 shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001857
1858 if (PageTransHuge(page)) {
1859 unlock_page(page);
1860 put_page(page);
1861 goto alloc_nohuge;
1862 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001863failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001864 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001865 error = -EEXIST;
1866unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001867 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001868 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001869 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001870 }
1871 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001872 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001873 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001874 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001875 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001876 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001877 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001878 goto repeat;
1879 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
1881
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001882/*
1883 * This is like autoremove_wake_function, but it removes the wait queue
1884 * entry unconditionally - even if something else had already woken the
1885 * target.
1886 */
1887static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1888{
1889 int ret = default_wake_function(wait, mode, sync, key);
1890 list_del_init(&wait->task_list);
1891 return ret;
1892}
1893
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895{
Al Viro496ad9a2013-01-23 17:07:38 -05001896 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001897 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001898 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001900 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001902 /*
1903 * Trinity finds that probing a hole which tmpfs is punching can
1904 * prevent the hole-punch from ever completing: which in turn
1905 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001906 * faulting pages into the hole while it's being punched. Although
1907 * shmem_undo_range() does remove the additions, it may be unable to
1908 * keep up, as each new page needs its own unmap_mapping_range() call,
1909 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1910 *
1911 * It does not matter if we sometimes reach this check just before the
1912 * hole-punch begins, so that one fault then races with the punch:
1913 * we just need to make racing faults a rare case.
1914 *
1915 * The implementation below would be much simpler if we just used a
1916 * standard mutex or completion: but we cannot take i_mutex in fault,
1917 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001918 */
1919 if (unlikely(inode->i_private)) {
1920 struct shmem_falloc *shmem_falloc;
1921
1922 spin_lock(&inode->i_lock);
1923 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001924 if (shmem_falloc &&
1925 shmem_falloc->waitq &&
1926 vmf->pgoff >= shmem_falloc->start &&
1927 vmf->pgoff < shmem_falloc->next) {
1928 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001929 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001930
1931 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001932 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1933 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001934 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001935 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001936 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001937 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001938
1939 shmem_falloc_waitq = shmem_falloc->waitq;
1940 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1941 TASK_UNINTERRUPTIBLE);
1942 spin_unlock(&inode->i_lock);
1943 schedule();
1944
1945 /*
1946 * shmem_falloc_waitq points into the shmem_fallocate()
1947 * stack of the hole-punching task: shmem_falloc_waitq
1948 * is usually invalid by the time we reach here, but
1949 * finish_wait() does not dereference it in that case;
1950 * though i_lock needed lest racing with wake_up_all().
1951 */
1952 spin_lock(&inode->i_lock);
1953 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1954 spin_unlock(&inode->i_lock);
1955 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001956 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001957 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001958 }
1959
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001960 sgp = SGP_CACHE;
1961 if (vma->vm_flags & VM_HUGEPAGE)
1962 sgp = SGP_HUGE;
1963 else if (vma->vm_flags & VM_NOHUGEPAGE)
1964 sgp = SGP_NOHUGE;
1965
1966 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001967 gfp, vma->vm_mm, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (error)
1969 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001970 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}
1972
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001973unsigned long shmem_get_unmapped_area(struct file *file,
1974 unsigned long uaddr, unsigned long len,
1975 unsigned long pgoff, unsigned long flags)
1976{
1977 unsigned long (*get_area)(struct file *,
1978 unsigned long, unsigned long, unsigned long, unsigned long);
1979 unsigned long addr;
1980 unsigned long offset;
1981 unsigned long inflated_len;
1982 unsigned long inflated_addr;
1983 unsigned long inflated_offset;
1984
1985 if (len > TASK_SIZE)
1986 return -ENOMEM;
1987
1988 get_area = current->mm->get_unmapped_area;
1989 addr = get_area(file, uaddr, len, pgoff, flags);
1990
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001991 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001992 return addr;
1993 if (IS_ERR_VALUE(addr))
1994 return addr;
1995 if (addr & ~PAGE_MASK)
1996 return addr;
1997 if (addr > TASK_SIZE - len)
1998 return addr;
1999
2000 if (shmem_huge == SHMEM_HUGE_DENY)
2001 return addr;
2002 if (len < HPAGE_PMD_SIZE)
2003 return addr;
2004 if (flags & MAP_FIXED)
2005 return addr;
2006 /*
2007 * Our priority is to support MAP_SHARED mapped hugely;
2008 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2009 * But if caller specified an address hint, respect that as before.
2010 */
2011 if (uaddr)
2012 return addr;
2013
2014 if (shmem_huge != SHMEM_HUGE_FORCE) {
2015 struct super_block *sb;
2016
2017 if (file) {
2018 VM_BUG_ON(file->f_op != &shmem_file_operations);
2019 sb = file_inode(file)->i_sb;
2020 } else {
2021 /*
2022 * Called directly from mm/mmap.c, or drivers/char/mem.c
2023 * for "/dev/zero", to create a shared anonymous object.
2024 */
2025 if (IS_ERR(shm_mnt))
2026 return addr;
2027 sb = shm_mnt->mnt_sb;
2028 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002029 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002030 return addr;
2031 }
2032
2033 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2034 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2035 return addr;
2036 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2037 return addr;
2038
2039 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2040 if (inflated_len > TASK_SIZE)
2041 return addr;
2042 if (inflated_len < len)
2043 return addr;
2044
2045 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2046 if (IS_ERR_VALUE(inflated_addr))
2047 return addr;
2048 if (inflated_addr & ~PAGE_MASK)
2049 return addr;
2050
2051 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2052 inflated_addr += offset - inflated_offset;
2053 if (inflated_offset > offset)
2054 inflated_addr += HPAGE_PMD_SIZE;
2055
2056 if (inflated_addr > TASK_SIZE - len)
2057 return addr;
2058 return inflated_addr;
2059}
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002062static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063{
Al Viro496ad9a2013-01-23 17:07:38 -05002064 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002065 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066}
2067
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002068static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2069 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
Al Viro496ad9a2013-01-23 17:07:38 -05002071 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002072 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002074 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2075 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076}
2077#endif
2078
2079int shmem_lock(struct file *file, int lock, struct user_struct *user)
2080{
Al Viro496ad9a2013-01-23 17:07:38 -05002081 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 struct shmem_inode_info *info = SHMEM_I(inode);
2083 int retval = -ENOMEM;
2084
Hugh Dickinsfccb7bb2020-04-20 18:14:14 -07002085 /*
2086 * What serializes the accesses to info->flags?
2087 * ipc_lock_object() when called from shmctl_do_lock(),
2088 * no serialization needed when called from shm_destroy().
2089 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 if (lock && !(info->flags & VM_LOCKED)) {
2091 if (!user_shm_lock(inode->i_size, user))
2092 goto out_nomem;
2093 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002094 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 }
2096 if (!lock && (info->flags & VM_LOCKED) && user) {
2097 user_shm_unlock(inode->i_size, user);
2098 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002099 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 }
2101 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002102
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103out_nomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 return retval;
2105}
2106
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002107static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108{
2109 file_accessed(file);
2110 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002111 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002112 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2113 (vma->vm_end & HPAGE_PMD_MASK)) {
2114 khugepaged_enter(vma, vma->vm_flags);
2115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 return 0;
2117}
2118
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002119static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002120 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121{
2122 struct inode *inode;
2123 struct shmem_inode_info *info;
2124 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2125
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002126 if (shmem_reserve_inode(sb))
2127 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
2129 inode = new_inode(sb);
2130 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002131 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002132 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002134 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002135 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 info = SHMEM_I(inode);
2137 memset(info, 0, (char *)inode - (char *)info);
2138 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002139 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002140 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002141 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002143 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002144 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
2146 switch (mode & S_IFMT) {
2147 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002148 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 init_special_inode(inode, mode, dev);
2150 break;
2151 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002152 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 inode->i_op = &shmem_inode_operations;
2154 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002155 mpol_shared_policy_init(&info->policy,
2156 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 break;
2158 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002159 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 /* Some things misbehave if size == 0 on a directory */
2161 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2162 inode->i_op = &shmem_dir_inode_operations;
2163 inode->i_fop = &simple_dir_operations;
2164 break;
2165 case S_IFLNK:
2166 /*
2167 * Must not load anything in the rbtree,
2168 * mpol_free_shared_policy will not be called.
2169 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002170 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 break;
2172 }
Joel Fernandes (Google)6b1bd5e2018-09-20 12:22:39 -07002173
2174 lockdep_annotate_inode_mutex_key(inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002175 } else
2176 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 return inode;
2178}
2179
Johannes Weiner0cd61442014-04-03 14:47:46 -07002180bool shmem_mapping(struct address_space *mapping)
2181{
Sasha Levinf0774d82015-02-23 05:38:00 -05002182 if (!mapping->host)
2183 return false;
2184
Christoph Hellwig97b713b2015-01-14 10:42:31 +01002185 return mapping->host->i_sb->s_op == &shmem_ops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002186}
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002189static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002190static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002192#ifdef CONFIG_TMPFS_XATTR
2193static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2194#else
2195#define shmem_initxattrs NULL
2196#endif
2197
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002199shmem_write_begin(struct file *file, struct address_space *mapping,
2200 loff_t pos, unsigned len, unsigned flags,
2201 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
Nick Piggin800d15a2007-10-16 01:25:03 -07002203 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002204 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002205 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002206
2207 /* i_mutex is held by caller */
2208 if (unlikely(info->seals)) {
2209 if (info->seals & F_SEAL_WRITE)
2210 return -EPERM;
2211 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2212 return -EPERM;
2213 }
2214
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002215 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002216}
2217
2218static int
2219shmem_write_end(struct file *file, struct address_space *mapping,
2220 loff_t pos, unsigned len, unsigned copied,
2221 struct page *page, void *fsdata)
2222{
2223 struct inode *inode = mapping->host;
2224
Hugh Dickinsd3602442008-02-04 22:28:44 -08002225 if (pos + copied > inode->i_size)
2226 i_size_write(inode, pos + copied);
2227
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002228 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002229 struct page *head = compound_head(page);
2230 if (PageTransCompound(page)) {
2231 int i;
2232
2233 for (i = 0; i < HPAGE_PMD_NR; i++) {
2234 if (head + i == page)
2235 continue;
2236 clear_highpage(head + i);
2237 flush_dcache_page(head + i);
2238 }
2239 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002240 if (copied < PAGE_SIZE) {
2241 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002242 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002243 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002244 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002245 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002246 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002247 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002248 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002249 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002250
Nick Piggin800d15a2007-10-16 01:25:03 -07002251 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252}
2253
Al Viro2ba5bbe2014-04-02 20:00:02 -04002254static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255{
Al Viro6e58e792014-02-03 17:07:03 -05002256 struct file *file = iocb->ki_filp;
2257 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002259 pgoff_t index;
2260 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002261 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002262 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002263 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002264 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002265
2266 /*
2267 * Might this read be for a stacking filesystem? Then when reading
2268 * holes of a sparse file, we actually need to allocate those pages,
2269 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2270 */
Al Viro777eda22014-12-17 04:46:46 -05002271 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002272 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002274 index = *ppos >> PAGE_SHIFT;
2275 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
2277 for (;;) {
2278 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002279 pgoff_t end_index;
2280 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 loff_t i_size = i_size_read(inode);
2282
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002283 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 if (index > end_index)
2285 break;
2286 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002287 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (nr <= offset)
2289 break;
2290 }
2291
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002292 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002293 if (error) {
2294 if (error == -EINVAL)
2295 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 break;
2297 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002298 if (page) {
2299 if (sgp == SGP_CACHE)
2300 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002301 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
2304 /*
2305 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002306 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002308 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002310 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002312 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 if (nr <= offset) {
2314 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002315 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 break;
2317 }
2318 }
2319 nr -= offset;
2320
2321 if (page) {
2322 /*
2323 * If users can be writing to this page using arbitrary
2324 * virtual addresses, take care about potential aliasing
2325 * before reading the page on the kernel side.
2326 */
2327 if (mapping_writably_mapped(mapping))
2328 flush_dcache_page(page);
2329 /*
2330 * Mark the page accessed if we read the beginning.
2331 */
2332 if (!offset)
2333 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002334 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002336 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002337 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
2339 /*
2340 * Ok, we have the page, and it's up-to-date, so
2341 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002343 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002344 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002346 index += offset >> PAGE_SHIFT;
2347 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002349 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002350 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 break;
Al Viro6e58e792014-02-03 17:07:03 -05002352 if (ret < nr) {
2353 error = -EFAULT;
2354 break;
2355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 cond_resched();
2357 }
2358
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002359 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002360 file_accessed(file);
2361 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362}
2363
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002364/*
2365 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2366 */
2367static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002368 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002369{
2370 struct page *page;
2371 struct pagevec pvec;
2372 pgoff_t indices[PAGEVEC_SIZE];
2373 bool done = false;
2374 int i;
2375
2376 pagevec_init(&pvec, 0);
2377 pvec.nr = 1; /* start small: we may be there already */
2378 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002379 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002380 pvec.nr, pvec.pages, indices);
2381 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002382 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002383 index = end;
2384 break;
2385 }
2386 for (i = 0; i < pvec.nr; i++, index++) {
2387 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002388 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002389 done = true;
2390 break;
2391 }
2392 index = indices[i];
2393 }
2394 page = pvec.pages[i];
2395 if (page && !radix_tree_exceptional_entry(page)) {
2396 if (!PageUptodate(page))
2397 page = NULL;
2398 }
2399 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002400 (page && whence == SEEK_DATA) ||
2401 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002402 done = true;
2403 break;
2404 }
2405 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002406 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002407 pagevec_release(&pvec);
2408 pvec.nr = PAGEVEC_SIZE;
2409 cond_resched();
2410 }
2411 return index;
2412}
2413
Andrew Morton965c8e52012-12-17 15:59:39 -08002414static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002415{
2416 struct address_space *mapping = file->f_mapping;
2417 struct inode *inode = mapping->host;
2418 pgoff_t start, end;
2419 loff_t new_offset;
2420
Andrew Morton965c8e52012-12-17 15:59:39 -08002421 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2422 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002423 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002424 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002425 /* We're holding i_mutex so we can access i_size directly */
2426
Yufen Yud77eacd2018-11-16 15:08:39 -08002427 if (offset < 0 || offset >= inode->i_size)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002428 offset = -ENXIO;
2429 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002430 start = offset >> PAGE_SHIFT;
2431 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002432 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002433 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002434 if (new_offset > offset) {
2435 if (new_offset < inode->i_size)
2436 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002437 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002438 offset = -ENXIO;
2439 else
2440 offset = inode->i_size;
2441 }
2442 }
2443
Hugh Dickins387aae62013-08-04 11:30:25 -07002444 if (offset >= 0)
2445 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002446 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002447 return offset;
2448}
2449
David Herrmann05f65b52014-08-08 14:25:36 -07002450/*
2451 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2452 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2453 */
2454#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2455#define LAST_SCAN 4 /* about 150ms max */
2456
2457static void shmem_tag_pins(struct address_space *mapping)
2458{
2459 struct radix_tree_iter iter;
2460 void **slot;
2461 pgoff_t start;
2462 struct page *page;
Matthew Wilcox (Oracle)3ce6b462019-10-25 09:58:36 -07002463 unsigned int tagged = 0;
David Herrmann05f65b52014-08-08 14:25:36 -07002464
2465 lru_add_drain();
2466 start = 0;
David Herrmann05f65b52014-08-08 14:25:36 -07002467
Matthew Wilcox (Oracle)3ce6b462019-10-25 09:58:36 -07002468 spin_lock_irq(&mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002469 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
zhong jiange0088cf2019-11-18 11:26:10 +08002470 page = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002471 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002472 if (radix_tree_deref_retry(page)) {
2473 slot = radix_tree_iter_retry(&iter);
2474 continue;
2475 }
David Herrmann05f65b52014-08-08 14:25:36 -07002476 } else if (page_count(page) - page_mapcount(page) > 1) {
David Herrmann05f65b52014-08-08 14:25:36 -07002477 radix_tree_tag_set(&mapping->page_tree, iter.index,
2478 SHMEM_TAG_PINNED);
David Herrmann05f65b52014-08-08 14:25:36 -07002479 }
2480
Matthew Wilcox (Oracle)3ce6b462019-10-25 09:58:36 -07002481 if (++tagged % 1024)
2482 continue;
2483
2484 slot = radix_tree_iter_next(&iter);
2485 spin_unlock_irq(&mapping->tree_lock);
2486 cond_resched();
2487 spin_lock_irq(&mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002488 }
Matthew Wilcox (Oracle)3ce6b462019-10-25 09:58:36 -07002489 spin_unlock_irq(&mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002490}
2491
2492/*
2493 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2494 * via get_user_pages(), drivers might have some pending I/O without any active
2495 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2496 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2497 * them to be dropped.
2498 * The caller must guarantee that no new user will acquire writable references
2499 * to those pages to avoid races.
2500 */
David Herrmann40e041a2014-08-08 14:25:27 -07002501static int shmem_wait_for_pins(struct address_space *mapping)
2502{
David Herrmann05f65b52014-08-08 14:25:36 -07002503 struct radix_tree_iter iter;
2504 void **slot;
2505 pgoff_t start;
2506 struct page *page;
2507 int error, scan;
2508
2509 shmem_tag_pins(mapping);
2510
2511 error = 0;
2512 for (scan = 0; scan <= LAST_SCAN; scan++) {
2513 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2514 break;
2515
2516 if (!scan)
2517 lru_add_drain_all();
2518 else if (schedule_timeout_killable((HZ << scan) / 200))
2519 scan = LAST_SCAN;
2520
2521 start = 0;
2522 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002523 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2524 start, SHMEM_TAG_PINNED) {
2525
2526 page = radix_tree_deref_slot(slot);
2527 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002528 if (radix_tree_deref_retry(page)) {
2529 slot = radix_tree_iter_retry(&iter);
2530 continue;
2531 }
David Herrmann05f65b52014-08-08 14:25:36 -07002532
2533 page = NULL;
2534 }
2535
2536 if (page &&
2537 page_count(page) - page_mapcount(page) != 1) {
2538 if (scan < LAST_SCAN)
2539 goto continue_resched;
2540
2541 /*
2542 * On the last scan, we clean up all those tags
2543 * we inserted; but make a note that we still
2544 * found pages pinned.
2545 */
2546 error = -EBUSY;
2547 }
2548
2549 spin_lock_irq(&mapping->tree_lock);
2550 radix_tree_tag_clear(&mapping->page_tree,
2551 iter.index, SHMEM_TAG_PINNED);
2552 spin_unlock_irq(&mapping->tree_lock);
2553continue_resched:
2554 if (need_resched()) {
2555 cond_resched_rcu();
Matthew Wilcox71650922016-03-17 14:22:06 -07002556 slot = radix_tree_iter_next(&iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002557 }
2558 }
2559 rcu_read_unlock();
2560 }
2561
2562 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002563}
2564
2565#define F_ALL_SEALS (F_SEAL_SEAL | \
2566 F_SEAL_SHRINK | \
2567 F_SEAL_GROW | \
2568 F_SEAL_WRITE)
2569
2570int shmem_add_seals(struct file *file, unsigned int seals)
2571{
2572 struct inode *inode = file_inode(file);
2573 struct shmem_inode_info *info = SHMEM_I(inode);
2574 int error;
2575
2576 /*
2577 * SEALING
2578 * Sealing allows multiple parties to share a shmem-file but restrict
2579 * access to a specific subset of file operations. Seals can only be
2580 * added, but never removed. This way, mutually untrusted parties can
2581 * share common memory regions with a well-defined policy. A malicious
2582 * peer can thus never perform unwanted operations on a shared object.
2583 *
2584 * Seals are only supported on special shmem-files and always affect
2585 * the whole underlying inode. Once a seal is set, it may prevent some
2586 * kinds of access to the file. Currently, the following seals are
2587 * defined:
2588 * SEAL_SEAL: Prevent further seals from being set on this file
2589 * SEAL_SHRINK: Prevent the file from shrinking
2590 * SEAL_GROW: Prevent the file from growing
2591 * SEAL_WRITE: Prevent write access to the file
2592 *
2593 * As we don't require any trust relationship between two parties, we
2594 * must prevent seals from being removed. Therefore, sealing a file
2595 * only adds a given set of seals to the file, it never touches
2596 * existing seals. Furthermore, the "setting seals"-operation can be
2597 * sealed itself, which basically prevents any further seal from being
2598 * added.
2599 *
2600 * Semantics of sealing are only defined on volatile files. Only
2601 * anonymous shmem files support sealing. More importantly, seals are
2602 * never written to disk. Therefore, there's no plan to support it on
2603 * other file types.
2604 */
2605
2606 if (file->f_op != &shmem_file_operations)
2607 return -EINVAL;
2608 if (!(file->f_mode & FMODE_WRITE))
2609 return -EPERM;
2610 if (seals & ~(unsigned int)F_ALL_SEALS)
2611 return -EINVAL;
2612
Al Viro59551022016-01-22 15:40:57 -05002613 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002614
2615 if (info->seals & F_SEAL_SEAL) {
2616 error = -EPERM;
2617 goto unlock;
2618 }
2619
2620 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2621 error = mapping_deny_writable(file->f_mapping);
2622 if (error)
2623 goto unlock;
2624
2625 error = shmem_wait_for_pins(file->f_mapping);
2626 if (error) {
2627 mapping_allow_writable(file->f_mapping);
2628 goto unlock;
2629 }
2630 }
2631
2632 info->seals |= seals;
2633 error = 0;
2634
2635unlock:
Al Viro59551022016-01-22 15:40:57 -05002636 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002637 return error;
2638}
2639EXPORT_SYMBOL_GPL(shmem_add_seals);
2640
2641int shmem_get_seals(struct file *file)
2642{
2643 if (file->f_op != &shmem_file_operations)
2644 return -EINVAL;
2645
2646 return SHMEM_I(file_inode(file))->seals;
2647}
2648EXPORT_SYMBOL_GPL(shmem_get_seals);
2649
2650long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2651{
2652 long error;
2653
2654 switch (cmd) {
2655 case F_ADD_SEALS:
2656 /* disallow upper 32bit */
2657 if (arg > UINT_MAX)
2658 return -EINVAL;
2659
2660 error = shmem_add_seals(file, arg);
2661 break;
2662 case F_GET_SEALS:
2663 error = shmem_get_seals(file);
2664 break;
2665 default:
2666 error = -EINVAL;
2667 break;
2668 }
2669
2670 return error;
2671}
2672
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002673static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2674 loff_t len)
2675{
Al Viro496ad9a2013-01-23 17:07:38 -05002676 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002677 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002678 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002679 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002680 pgoff_t start, index, end;
2681 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002682
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002683 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2684 return -EOPNOTSUPP;
2685
Al Viro59551022016-01-22 15:40:57 -05002686 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002687
2688 if (mode & FALLOC_FL_PUNCH_HOLE) {
2689 struct address_space *mapping = file->f_mapping;
2690 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2691 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002692 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002693
David Herrmann40e041a2014-08-08 14:25:27 -07002694 /* protected by i_mutex */
2695 if (info->seals & F_SEAL_WRITE) {
2696 error = -EPERM;
2697 goto out;
2698 }
2699
Hugh Dickins8e205f72014-07-23 14:00:10 -07002700 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junf9318662019-11-30 17:58:11 -08002701 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002702 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2703 spin_lock(&inode->i_lock);
2704 inode->i_private = &shmem_falloc;
2705 spin_unlock(&inode->i_lock);
2706
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002707 if ((u64)unmap_end > (u64)unmap_start)
2708 unmap_mapping_range(mapping, unmap_start,
2709 1 + unmap_end - unmap_start, 0);
2710 shmem_truncate_range(inode, offset, offset + len - 1);
2711 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002712
2713 spin_lock(&inode->i_lock);
2714 inode->i_private = NULL;
2715 wake_up_all(&shmem_falloc_waitq);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002716 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002717 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002718 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002719 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002720 }
2721
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002722 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2723 error = inode_newsize_ok(inode, offset + len);
2724 if (error)
2725 goto out;
2726
David Herrmann40e041a2014-08-08 14:25:27 -07002727 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2728 error = -EPERM;
2729 goto out;
2730 }
2731
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002732 start = offset >> PAGE_SHIFT;
2733 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002734 /* Try to avoid a swapstorm if len is impossible to satisfy */
2735 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2736 error = -ENOSPC;
2737 goto out;
2738 }
2739
Hugh Dickins8e205f72014-07-23 14:00:10 -07002740 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002741 shmem_falloc.start = start;
2742 shmem_falloc.next = start;
2743 shmem_falloc.nr_falloced = 0;
2744 shmem_falloc.nr_unswapped = 0;
2745 spin_lock(&inode->i_lock);
2746 inode->i_private = &shmem_falloc;
2747 spin_unlock(&inode->i_lock);
2748
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002749 for (index = start; index < end; index++) {
2750 struct page *page;
2751
2752 /*
2753 * Good, the fallocate(2) manpage permits EINTR: we may have
2754 * been interrupted because we are using up too much memory.
2755 */
2756 if (signal_pending(current))
2757 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002758 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2759 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002760 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002761 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002762 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002763 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002764 if (index > start) {
2765 shmem_undo_range(inode,
2766 (loff_t)start << PAGE_SHIFT,
2767 ((loff_t)index << PAGE_SHIFT) - 1, true);
2768 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002769 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002770 }
2771
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002772 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002773 * Inform shmem_writepage() how far we have reached.
2774 * No need for lock or barrier: we have the page lock.
2775 */
2776 shmem_falloc.next++;
2777 if (!PageUptodate(page))
2778 shmem_falloc.nr_falloced++;
2779
2780 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002781 * If !PageUptodate, leave it that way so that freeable pages
2782 * can be recognized if we need to rollback on error later.
2783 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002784 * than free the pages we are allocating (and SGP_CACHE pages
2785 * might still be clean: we now need to mark those dirty too).
2786 */
2787 set_page_dirty(page);
2788 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002789 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002790 cond_resched();
2791 }
2792
2793 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2794 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002795 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002796undone:
2797 spin_lock(&inode->i_lock);
2798 inode->i_private = NULL;
2799 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002800out:
Al Viro59551022016-01-22 15:40:57 -05002801 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002802 return error;
2803}
2804
David Howells726c3342006-06-23 02:02:58 -07002805static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806{
David Howells726c3342006-06-23 02:02:58 -07002807 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
2809 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002810 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002812 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002814 buf->f_bavail =
2815 buf->f_bfree = sbinfo->max_blocks -
2816 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002817 }
2818 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 buf->f_files = sbinfo->max_inodes;
2820 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 }
2822 /* else leave those fields 0 like simple_statfs */
2823 return 0;
2824}
2825
2826/*
2827 * File creation. Allocate an inode, and we're done..
2828 */
2829static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002830shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002832 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 int error = -ENOSPC;
2834
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002835 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002837 error = simple_acl_create(dir, inode);
2838 if (error)
2839 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002840 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002841 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002842 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002843 if (error && error != -EOPNOTSUPP)
2844 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002845
Al Viro718deb62009-12-16 19:35:36 -05002846 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002848 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 d_instantiate(dentry, inode);
2850 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 }
2852 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002853out_iput:
2854 iput(inode);
2855 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856}
2857
Al Viro60545d02013-06-07 01:20:27 -04002858static int
2859shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2860{
2861 struct inode *inode;
2862 int error = -ENOSPC;
2863
2864 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2865 if (inode) {
2866 error = security_inode_init_security(inode, dir,
2867 NULL,
2868 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002869 if (error && error != -EOPNOTSUPP)
2870 goto out_iput;
2871 error = simple_acl_create(dir, inode);
2872 if (error)
2873 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002874 d_tmpfile(dentry, inode);
2875 }
2876 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002877out_iput:
2878 iput(inode);
2879 return error;
Al Viro60545d02013-06-07 01:20:27 -04002880}
2881
Al Viro18bb1db2011-07-26 01:41:39 -04002882static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883{
2884 int error;
2885
2886 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2887 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002888 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 return 0;
2890}
2891
Al Viro4acdaf22011-07-26 01:42:34 -04002892static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002893 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894{
2895 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2896}
2897
2898/*
2899 * Link a file..
2900 */
2901static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2902{
David Howells75c3cfa2015-03-17 22:26:12 +00002903 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong2484aab2019-02-22 22:35:32 -08002904 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905
2906 /*
2907 * No ordinary (disk based) filesystem counts links as inodes;
2908 * but each new link needs a new dentry, pinning lowmem, and
2909 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wongeb85c846ec2019-02-21 08:48:09 -08002910 * But if an O_TMPFILE file is linked into the tmpfs, the
2911 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 */
Darrick J. Wongeb85c846ec2019-02-21 08:48:09 -08002913 if (inode->i_nlink) {
2914 ret = shmem_reserve_inode(inode->i_sb);
2915 if (ret)
2916 goto out;
2917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918
2919 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002920 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002921 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002922 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 dget(dentry); /* Extra pinning count for the created dentry */
2924 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002925out:
2926 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927}
2928
2929static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2930{
David Howells75c3cfa2015-03-17 22:26:12 +00002931 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002933 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2934 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935
2936 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002937 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002938 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 dput(dentry); /* Undo the count from "create" - this does all the work */
2940 return 0;
2941}
2942
2943static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2944{
2945 if (!simple_empty(dentry))
2946 return -ENOTEMPTY;
2947
David Howells75c3cfa2015-03-17 22:26:12 +00002948 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002949 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 return shmem_unlink(dir, dentry);
2951}
2952
Miklos Szeredi37456772014-07-23 15:15:34 +02002953static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2954{
David Howellse36cb0b2015-01-29 12:02:35 +00002955 bool old_is_dir = d_is_dir(old_dentry);
2956 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02002957
2958 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2959 if (old_is_dir) {
2960 drop_nlink(old_dir);
2961 inc_nlink(new_dir);
2962 } else {
2963 drop_nlink(new_dir);
2964 inc_nlink(old_dir);
2965 }
2966 }
2967 old_dir->i_ctime = old_dir->i_mtime =
2968 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00002969 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07002970 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02002971
2972 return 0;
2973}
2974
Miklos Szeredi46fdb792014-10-24 00:14:37 +02002975static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2976{
2977 struct dentry *whiteout;
2978 int error;
2979
2980 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2981 if (!whiteout)
2982 return -ENOMEM;
2983
2984 error = shmem_mknod(old_dir, whiteout,
2985 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2986 dput(whiteout);
2987 if (error)
2988 return error;
2989
2990 /*
2991 * Cheat and hash the whiteout while the old dentry is still in
2992 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2993 *
2994 * d_lookup() will consistently find one of them at this point,
2995 * not sure which one, but that isn't even important.
2996 */
2997 d_rehash(whiteout);
2998 return 0;
2999}
3000
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001/*
3002 * The VFS layer already does all the dentry stuff for rename,
3003 * we just have to decrement the usage count for the target if
3004 * it exists so that the VFS layer correctly free's it when it
3005 * gets overwritten.
3006 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003007static 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 -07003008{
David Howells75c3cfa2015-03-17 22:26:12 +00003009 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 int they_are_dirs = S_ISDIR(inode->i_mode);
3011
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003012 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003013 return -EINVAL;
3014
Miklos Szeredi37456772014-07-23 15:15:34 +02003015 if (flags & RENAME_EXCHANGE)
3016 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3017
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 if (!simple_empty(new_dentry))
3019 return -ENOTEMPTY;
3020
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003021 if (flags & RENAME_WHITEOUT) {
3022 int error;
3023
3024 error = shmem_whiteout(old_dir, old_dentry);
3025 if (error)
3026 return error;
3027 }
3028
David Howells75c3cfa2015-03-17 22:26:12 +00003029 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003031 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003032 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003033 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003036 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003037 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 }
3039
3040 old_dir->i_size -= BOGO_DIRENT_SIZE;
3041 new_dir->i_size += BOGO_DIRENT_SIZE;
3042 old_dir->i_ctime = old_dir->i_mtime =
3043 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003044 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 return 0;
3046}
3047
3048static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3049{
3050 int error;
3051 int len;
3052 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003053 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 struct shmem_inode_info *info;
3055
3056 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003057 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 return -ENAMETOOLONG;
3059
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003060 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 if (!inode)
3062 return -ENOSPC;
3063
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003064 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003065 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003066 if (error) {
3067 if (error != -EOPNOTSUPP) {
3068 iput(inode);
3069 return error;
3070 }
3071 error = 0;
3072 }
3073
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 info = SHMEM_I(inode);
3075 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003076 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003077 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3078 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003079 iput(inode);
3080 return -ENOMEM;
3081 }
3082 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003084 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003085 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 if (error) {
3087 iput(inode);
3088 return error;
3089 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003090 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003092 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003093 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003095 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003096 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003099 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 d_instantiate(dentry, inode);
3101 dget(dentry);
3102 return 0;
3103}
3104
Al Virofceef392015-12-29 15:58:39 -05003105static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106{
Al Virofceef392015-12-29 15:58:39 -05003107 mark_page_accessed(arg);
3108 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109}
3110
Al Viro6b255392015-11-17 10:20:54 -05003111static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003112 struct inode *inode,
3113 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003116 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003117 if (!dentry) {
3118 page = find_get_page(inode->i_mapping, 0);
3119 if (!page)
3120 return ERR_PTR(-ECHILD);
3121 if (!PageUptodate(page)) {
3122 put_page(page);
3123 return ERR_PTR(-ECHILD);
3124 }
3125 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003126 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003127 if (error)
3128 return ERR_PTR(error);
3129 unlock_page(page);
3130 }
Al Virofceef392015-12-29 15:58:39 -05003131 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003132 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133}
3134
Eric Parisb09e0fa2011-05-24 17:12:39 -07003135#ifdef CONFIG_TMPFS_XATTR
3136/*
3137 * Superblocks without xattr inode operations may get some security.* xattr
3138 * support from the LSM "for free". As soon as we have any other xattrs
3139 * like ACLs, we also need to implement the security.* handlers at
3140 * filesystem level, though.
3141 */
3142
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003143/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003144 * Callback for security_inode_init_security() for acquiring xattrs.
3145 */
3146static int shmem_initxattrs(struct inode *inode,
3147 const struct xattr *xattr_array,
3148 void *fs_info)
3149{
3150 struct shmem_inode_info *info = SHMEM_I(inode);
3151 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003152 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003153 size_t len;
3154
3155 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003156 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003157 if (!new_xattr)
3158 return -ENOMEM;
3159
3160 len = strlen(xattr->name) + 1;
3161 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3162 GFP_KERNEL);
3163 if (!new_xattr->name) {
3164 kfree(new_xattr);
3165 return -ENOMEM;
3166 }
3167
3168 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3169 XATTR_SECURITY_PREFIX_LEN);
3170 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3171 xattr->name, len);
3172
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003173 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003174 }
3175
3176 return 0;
3177}
3178
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003179static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003180 struct dentry *unused, struct inode *inode,
3181 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003182{
Al Virob2968212016-04-10 20:48:24 -04003183 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003184
3185 name = xattr_full_name(handler, name);
3186 return simple_xattr_get(&info->xattrs, name, buffer, size);
3187}
3188
3189static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003190 struct dentry *unused, struct inode *inode,
3191 const char *name, const void *value,
3192 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003193{
Al Viro59301222016-05-27 10:19:30 -04003194 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003195
3196 name = xattr_full_name(handler, name);
3197 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3198}
3199
3200static const struct xattr_handler shmem_security_xattr_handler = {
3201 .prefix = XATTR_SECURITY_PREFIX,
3202 .get = shmem_xattr_handler_get,
3203 .set = shmem_xattr_handler_set,
3204};
3205
3206static const struct xattr_handler shmem_trusted_xattr_handler = {
3207 .prefix = XATTR_TRUSTED_PREFIX,
3208 .get = shmem_xattr_handler_get,
3209 .set = shmem_xattr_handler_set,
3210};
3211
Eric Parisb09e0fa2011-05-24 17:12:39 -07003212static const struct xattr_handler *shmem_xattr_handlers[] = {
3213#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003214 &posix_acl_access_xattr_handler,
3215 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003216#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003217 &shmem_security_xattr_handler,
3218 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003219 NULL
3220};
3221
Eric Parisb09e0fa2011-05-24 17:12:39 -07003222static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3223{
David Howells75c3cfa2015-03-17 22:26:12 +00003224 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003225 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003226}
3227#endif /* CONFIG_TMPFS_XATTR */
3228
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003229static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003231 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003232#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003233 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003234#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235};
3236
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003237static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -05003239 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003240#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003241 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003242#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003243};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003244
David M. Grimes91828a42006-10-17 00:09:45 -07003245static struct dentry *shmem_get_parent(struct dentry *child)
3246{
3247 return ERR_PTR(-ESTALE);
3248}
3249
3250static int shmem_match(struct inode *ino, void *vfh)
3251{
3252 __u32 *fh = vfh;
3253 __u64 inum = fh[2];
3254 inum = (inum << 32) | fh[1];
3255 return ino->i_ino == inum && fh[0] == ino->i_generation;
3256}
3257
Christoph Hellwig480b1162007-10-21 16:42:13 -07003258static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3259 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003260{
David M. Grimes91828a42006-10-17 00:09:45 -07003261 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003262 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003263 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003264
Christoph Hellwig480b1162007-10-21 16:42:13 -07003265 if (fh_len < 3)
3266 return NULL;
3267
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003268 inum = fid->raw[2];
3269 inum = (inum << 32) | fid->raw[1];
3270
Christoph Hellwig480b1162007-10-21 16:42:13 -07003271 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3272 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003273 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003274 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003275 iput(inode);
3276 }
3277
Christoph Hellwig480b1162007-10-21 16:42:13 -07003278 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003279}
3280
Al Virob0b03822012-04-02 14:34:06 -04003281static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3282 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003283{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303284 if (*len < 3) {
3285 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003286 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303287 }
David M. Grimes91828a42006-10-17 00:09:45 -07003288
Al Viro1d3382cb2010-10-23 15:19:20 -04003289 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003290 /* Unfortunately insert_inode_hash is not idempotent,
3291 * so as we hash inodes here rather than at creation
3292 * time, we need a lock to ensure we only try
3293 * to do it once
3294 */
3295 static DEFINE_SPINLOCK(lock);
3296 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003297 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003298 __insert_inode_hash(inode,
3299 inode->i_ino + inode->i_generation);
3300 spin_unlock(&lock);
3301 }
3302
3303 fh[0] = inode->i_generation;
3304 fh[1] = inode->i_ino;
3305 fh[2] = ((__u64)inode->i_ino) >> 32;
3306
3307 *len = 3;
3308 return 1;
3309}
3310
Christoph Hellwig39655162007-10-21 16:42:17 -07003311static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003312 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003313 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003314 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003315};
3316
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003317static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3318 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319{
3320 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003321 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003322 uid_t uid;
3323 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003325 while (options != NULL) {
3326 this_char = options;
3327 for (;;) {
3328 /*
3329 * NUL-terminate this option: unfortunately,
3330 * mount options form a comma-separated list,
3331 * but mpol's nodelist may also contain commas.
3332 */
3333 options = strchr(options, ',');
3334 if (options == NULL)
3335 break;
3336 options++;
3337 if (!isdigit(*options)) {
3338 options[-1] = '\0';
3339 break;
3340 }
3341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342 if (!*this_char)
3343 continue;
3344 if ((value = strchr(this_char,'=')) != NULL) {
3345 *value++ = 0;
3346 } else {
Joe Perches11705322016-03-17 14:19:50 -07003347 pr_err("tmpfs: No value for mount option '%s'\n",
3348 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003349 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 }
3351
3352 if (!strcmp(this_char,"size")) {
3353 unsigned long long size;
3354 size = memparse(value,&rest);
3355 if (*rest == '%') {
3356 size <<= PAGE_SHIFT;
3357 size *= totalram_pages;
3358 do_div(size, 100);
3359 rest++;
3360 }
3361 if (*rest)
3362 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003363 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003364 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003366 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 if (*rest)
3368 goto bad_val;
3369 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003370 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 if (*rest)
3372 goto bad_val;
3373 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003374 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003376 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 if (*rest)
3378 goto bad_val;
3379 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003380 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003382 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 if (*rest)
3384 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003385 sbinfo->uid = make_kuid(current_user_ns(), uid);
3386 if (!uid_valid(sbinfo->uid))
3387 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003389 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003391 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 if (*rest)
3393 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003394 sbinfo->gid = make_kgid(current_user_ns(), gid);
3395 if (!gid_valid(sbinfo->gid))
3396 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003397#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003398 } else if (!strcmp(this_char, "huge")) {
3399 int huge;
3400 huge = shmem_parse_huge(value);
3401 if (huge < 0)
3402 goto bad_val;
3403 if (!has_transparent_hugepage() &&
3404 huge != SHMEM_HUGE_NEVER)
3405 goto bad_val;
3406 sbinfo->huge = huge;
3407#endif
3408#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003409 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003410 mpol_put(mpol);
3411 mpol = NULL;
3412 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003413 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003414#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 } else {
Joe Perches11705322016-03-17 14:19:50 -07003416 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003417 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 }
3419 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003420 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 return 0;
3422
3423bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003424 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003426error:
3427 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 return 1;
3429
3430}
3431
3432static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3433{
3434 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003435 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003436 unsigned long inodes;
3437 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438
Greg Thelen5f001102013-02-22 16:36:01 -08003439 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003440 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003441 return error;
3442
3443 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003444 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003445 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003446 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003447 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003448 goto out;
3449 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003450 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003451 * but we must separately disallow unlimited->limited, because
3452 * in that case we have no record of how much is already in use.
3453 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003454 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003455 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003456 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003457 goto out;
3458
3459 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003460 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003461 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003462 sbinfo->max_inodes = config.max_inodes;
3463 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003464
Greg Thelen5f001102013-02-22 16:36:01 -08003465 /*
3466 * Preserve previous mempolicy unless mpol remount option was specified.
3467 */
3468 if (config.mpol) {
3469 mpol_put(sbinfo->mpol);
3470 sbinfo->mpol = config.mpol; /* transfers initial ref */
3471 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003472out:
3473 spin_unlock(&sbinfo->stat_lock);
3474 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003476
Al Viro34c80b12011-12-08 21:32:45 -05003477static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003478{
Al Viro34c80b12011-12-08 21:32:45 -05003479 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003480
3481 if (sbinfo->max_blocks != shmem_default_max_blocks())
3482 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003483 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003484 if (sbinfo->max_inodes != shmem_default_max_inodes())
3485 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3486 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003487 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003488 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3489 seq_printf(seq, ",uid=%u",
3490 from_kuid_munged(&init_user_ns, sbinfo->uid));
3491 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3492 seq_printf(seq, ",gid=%u",
3493 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003494#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003495 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3496 if (sbinfo->huge)
3497 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3498#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003499 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003500 return 0;
3501}
David Herrmann9183df22014-08-08 14:25:29 -07003502
3503#define MFD_NAME_PREFIX "memfd:"
3504#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3505#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3506
3507#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3508
3509SYSCALL_DEFINE2(memfd_create,
3510 const char __user *, uname,
3511 unsigned int, flags)
3512{
3513 struct shmem_inode_info *info;
3514 struct file *file;
3515 int fd, error;
3516 char *name;
3517 long len;
3518
3519 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3520 return -EINVAL;
3521
3522 /* length includes terminating zero */
3523 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3524 if (len <= 0)
3525 return -EFAULT;
3526 if (len > MFD_NAME_MAX_LEN + 1)
3527 return -EINVAL;
3528
3529 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3530 if (!name)
3531 return -ENOMEM;
3532
3533 strcpy(name, MFD_NAME_PREFIX);
3534 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3535 error = -EFAULT;
3536 goto err_name;
3537 }
3538
3539 /* terminating-zero may have changed after strnlen_user() returned */
3540 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3541 error = -EFAULT;
3542 goto err_name;
3543 }
3544
3545 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3546 if (fd < 0) {
3547 error = fd;
3548 goto err_name;
3549 }
3550
3551 file = shmem_file_setup(name, 0, VM_NORESERVE);
3552 if (IS_ERR(file)) {
3553 error = PTR_ERR(file);
3554 goto err_fd;
3555 }
3556 info = SHMEM_I(file_inode(file));
3557 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3558 file->f_flags |= O_RDWR | O_LARGEFILE;
3559 if (flags & MFD_ALLOW_SEALING)
3560 info->seals &= ~F_SEAL_SEAL;
3561
3562 fd_install(fd, file);
3563 kfree(name);
3564 return fd;
3565
3566err_fd:
3567 put_unused_fd(fd);
3568err_name:
3569 kfree(name);
3570 return error;
3571}
3572
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003573#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574
3575static void shmem_put_super(struct super_block *sb)
3576{
Hugh Dickins602586a2010-08-17 15:23:56 -07003577 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3578
3579 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003580 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003581 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582 sb->s_fs_info = NULL;
3583}
3584
Kay Sievers2b2af542009-04-30 15:23:42 +02003585int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586{
3587 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003588 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003589 int err = -ENOMEM;
3590
3591 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003592 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003593 L1_CACHE_BYTES), GFP_KERNEL);
3594 if (!sbinfo)
3595 return -ENOMEM;
3596
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003597 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003598 sbinfo->uid = current_fsuid();
3599 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003600 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003602#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 /*
3604 * Per default we only allow half of the physical ram per
3605 * tmpfs instance, limiting inodes to one per page of lowmem;
3606 * but the internal instance is left unlimited.
3607 */
Al Viroca4e0512013-08-31 12:57:10 -04003608 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003609 sbinfo->max_blocks = shmem_default_max_blocks();
3610 sbinfo->max_inodes = shmem_default_max_inodes();
3611 if (shmem_parse_options(data, sbinfo, false)) {
3612 err = -EINVAL;
3613 goto failed;
3614 }
Al Viroca4e0512013-08-31 12:57:10 -04003615 } else {
3616 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 }
David M. Grimes91828a42006-10-17 00:09:45 -07003618 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003619 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620#else
3621 sb->s_flags |= MS_NOUSER;
3622#endif
3623
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003624 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003625 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003626 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003627 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003628 spin_lock_init(&sbinfo->shrinklist_lock);
3629 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003630
Hugh Dickins285b2c42011-08-03 16:21:20 -07003631 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003632 sb->s_blocksize = PAGE_SIZE;
3633 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634 sb->s_magic = TMPFS_MAGIC;
3635 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003636 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003637#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003638 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003639#endif
3640#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003641 sb->s_flags |= MS_POSIXACL;
3642#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003643
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003644 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645 if (!inode)
3646 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003647 inode->i_uid = sbinfo->uid;
3648 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003649 sb->s_root = d_make_root(inode);
3650 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003651 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 return 0;
3653
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654failed:
3655 shmem_put_super(sb);
3656 return err;
3657}
3658
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003659static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660
3661static struct inode *shmem_alloc_inode(struct super_block *sb)
3662{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003663 struct shmem_inode_info *info;
3664 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3665 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003667 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668}
3669
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003670static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003671{
3672 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003673 if (S_ISLNK(inode->i_mode))
3674 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003675 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3676}
3677
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678static void shmem_destroy_inode(struct inode *inode)
3679{
Al Viro09208d12011-07-26 03:15:03 -04003680 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003682 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683}
3684
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003685static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003687 struct shmem_inode_info *info = foo;
3688 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689}
3690
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003691static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692{
3693 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3694 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003695 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 return 0;
3697}
3698
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003699static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003701 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702}
3703
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003704static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003706 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003708 .write_begin = shmem_write_begin,
3709 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003711#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003712 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003713#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003714 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715};
3716
Helge Deller15ad7cd2006-12-06 20:40:36 -08003717static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003719 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003721 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003722 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003723 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003724 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003725 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003726 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003727 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728#endif
3729};
3730
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003731static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003732 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003733 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003734#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003735 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003736 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003737#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738};
3739
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003740static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741#ifdef CONFIG_TMPFS
3742 .create = shmem_create,
3743 .lookup = simple_lookup,
3744 .link = shmem_link,
3745 .unlink = shmem_unlink,
3746 .symlink = shmem_symlink,
3747 .mkdir = shmem_mkdir,
3748 .rmdir = shmem_rmdir,
3749 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003750 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003751 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003753#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003754 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003755#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003756#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003757 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003758 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003759#endif
3760};
3761
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003762static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003763#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003764 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003765#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003766#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003767 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003768 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003769#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770};
3771
Hugh Dickins759b9772007-03-05 00:30:28 -08003772static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 .alloc_inode = shmem_alloc_inode,
3774 .destroy_inode = shmem_destroy_inode,
3775#ifdef CONFIG_TMPFS
3776 .statfs = shmem_statfs,
3777 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003778 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779#endif
Al Viro1f895f72010-06-05 19:10:41 -04003780 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 .drop_inode = generic_delete_inode,
3782 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003783#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3784 .nr_cached_objects = shmem_unused_huge_count,
3785 .free_cached_objects = shmem_unused_huge_scan,
3786#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787};
3788
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003789static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003790 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003791 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792#ifdef CONFIG_NUMA
3793 .set_policy = shmem_set_policy,
3794 .get_policy = shmem_get_policy,
3795#endif
3796};
3797
Al Viro3c26ff62010-07-25 11:46:36 +04003798static struct dentry *shmem_mount(struct file_system_type *fs_type,
3799 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800{
Al Viro3c26ff62010-07-25 11:46:36 +04003801 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802}
3803
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003804static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805 .owner = THIS_MODULE,
3806 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003807 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003809 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003812int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813{
3814 int error;
3815
Rob Landley16203a72013-09-11 14:26:12 -07003816 /* If rootfs called this, don't re-init */
3817 if (shmem_inode_cachep)
3818 return 0;
3819
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003820 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 if (error)
3822 goto out3;
3823
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003824 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003826 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 goto out2;
3828 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003829
Al Viroca4e0512013-08-31 12:57:10 -04003830 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 if (IS_ERR(shm_mnt)) {
3832 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003833 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834 goto out1;
3835 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003836
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003837#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5d8b3cc2017-08-25 15:55:33 -07003838 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003839 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3840 else
3841 shmem_huge = 0; /* just in case it was patched */
3842#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843 return 0;
3844
3845out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003846 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003848 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849out3:
3850 shm_mnt = ERR_PTR(error);
3851 return error;
3852}
Matt Mackall853ac432009-01-06 14:40:20 -08003853
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003854#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003855static ssize_t shmem_enabled_show(struct kobject *kobj,
3856 struct kobj_attribute *attr, char *buf)
3857{
3858 int values[] = {
3859 SHMEM_HUGE_ALWAYS,
3860 SHMEM_HUGE_WITHIN_SIZE,
3861 SHMEM_HUGE_ADVISE,
3862 SHMEM_HUGE_NEVER,
3863 SHMEM_HUGE_DENY,
3864 SHMEM_HUGE_FORCE,
3865 };
3866 int i, count;
3867
3868 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3869 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3870
3871 count += sprintf(buf + count, fmt,
3872 shmem_format_huge(values[i]));
3873 }
3874 buf[count - 1] = '\n';
3875 return count;
3876}
3877
3878static ssize_t shmem_enabled_store(struct kobject *kobj,
3879 struct kobj_attribute *attr, const char *buf, size_t count)
3880{
3881 char tmp[16];
3882 int huge;
3883
3884 if (count + 1 > sizeof(tmp))
3885 return -EINVAL;
3886 memcpy(tmp, buf, count);
3887 tmp[count] = '\0';
3888 if (count && tmp[count - 1] == '\n')
3889 tmp[count - 1] = '\0';
3890
3891 huge = shmem_parse_huge(tmp);
3892 if (huge == -EINVAL)
3893 return -EINVAL;
3894 if (!has_transparent_hugepage() &&
3895 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3896 return -EINVAL;
3897
3898 shmem_huge = huge;
Kirill A. Shutemov5d8b3cc2017-08-25 15:55:33 -07003899 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003900 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3901 return count;
3902}
3903
3904struct kobj_attribute shmem_enabled_attr =
3905 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003906#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003907
Arnd Bergmann3b337192016-08-10 16:27:44 -07003908#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003909bool shmem_huge_enabled(struct vm_area_struct *vma)
3910{
3911 struct inode *inode = file_inode(vma->vm_file);
3912 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3913 loff_t i_size;
3914 pgoff_t off;
3915
3916 if (shmem_huge == SHMEM_HUGE_FORCE)
3917 return true;
3918 if (shmem_huge == SHMEM_HUGE_DENY)
3919 return false;
3920 switch (sbinfo->huge) {
3921 case SHMEM_HUGE_NEVER:
3922 return false;
3923 case SHMEM_HUGE_ALWAYS:
3924 return true;
3925 case SHMEM_HUGE_WITHIN_SIZE:
3926 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3927 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3928 if (i_size >= HPAGE_PMD_SIZE &&
3929 i_size >> PAGE_SHIFT >= off)
3930 return true;
3931 case SHMEM_HUGE_ADVISE:
3932 /* TODO: implement fadvise() hints */
3933 return (vma->vm_flags & VM_HUGEPAGE);
3934 default:
3935 VM_BUG_ON(1);
3936 return false;
3937 }
3938}
Arnd Bergmann3b337192016-08-10 16:27:44 -07003939#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003940
Matt Mackall853ac432009-01-06 14:40:20 -08003941#else /* !CONFIG_SHMEM */
3942
3943/*
3944 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3945 *
3946 * This is intended for small system where the benefits of the full
3947 * shmem code (swap-backed and resource-limited) are outweighed by
3948 * their complexity. On systems without swap this code should be
3949 * effectively equivalent, but much lighter weight.
3950 */
3951
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003952static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08003953 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003954 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08003955 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003956 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08003957};
3958
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003959int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08003960{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003961 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08003962
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003963 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08003964 BUG_ON(IS_ERR(shm_mnt));
3965
3966 return 0;
3967}
3968
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003969int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08003970{
3971 return 0;
3972}
3973
Hugh Dickins3f96b792009-09-21 17:03:37 -07003974int shmem_lock(struct file *file, int lock, struct user_struct *user)
3975{
3976 return 0;
3977}
3978
Hugh Dickins24513262012-01-20 14:34:21 -08003979void shmem_unlock_mapping(struct address_space *mapping)
3980{
3981}
3982
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003983#ifdef CONFIG_MMU
3984unsigned long shmem_get_unmapped_area(struct file *file,
3985 unsigned long addr, unsigned long len,
3986 unsigned long pgoff, unsigned long flags)
3987{
3988 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3989}
3990#endif
3991
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003992void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07003993{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003994 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07003995}
3996EXPORT_SYMBOL_GPL(shmem_truncate_range);
3997
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003998#define shmem_vm_ops generic_file_vm_ops
3999#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004000#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004001#define shmem_acct_size(flags, size) 0
4002#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004003
4004#endif /* CONFIG_SHMEM */
4005
4006/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007
Rasmus Villemoes19938e32016-10-07 17:01:01 -07004008static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04004009 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05004010};
4011
Eric Parisc7277092013-12-02 11:24:19 +00004012static struct file *__shmem_file_setup(const char *name, loff_t size,
4013 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014{
Al Viro6b4d0b22013-02-14 21:37:26 -05004015 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04004017 struct path path;
Al Viro34515382013-02-14 22:38:02 -05004018 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 struct qstr this;
4020
4021 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05004022 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023
Hugh Dickins285b2c42011-08-03 16:21:20 -07004024 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 return ERR_PTR(-EINVAL);
4026
4027 if (shmem_acct_size(flags, size))
4028 return ERR_PTR(-ENOMEM);
4029
Al Viro6b4d0b22013-02-14 21:37:26 -05004030 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 this.name = name;
4032 this.len = strlen(name);
4033 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004034 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004035 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004036 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004037 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004039 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040
Al Viro6b4d0b22013-02-14 21:37:26 -05004041 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004042 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004044 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045
Eric Parisc7277092013-12-02 11:24:19 +00004046 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004047 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004049 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004050 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4051 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004052 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004053
Al Viro6b4d0b22013-02-14 21:37:26 -05004054 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004055 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004056 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004057 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004058
Al Viro6b4d0b22013-02-14 21:37:26 -05004059 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061put_memory:
4062 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004063put_path:
4064 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004065 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066}
Eric Parisc7277092013-12-02 11:24:19 +00004067
4068/**
4069 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4070 * kernel internal. There will be NO LSM permission checks against the
4071 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004072 * higher layer. The users are the big_key and shm implementations. LSM
4073 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004074 * @name: name for dentry (to be seen in /proc/<pid>/maps
4075 * @size: size to be set for the file
4076 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4077 */
4078struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4079{
4080 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4081}
4082
4083/**
4084 * shmem_file_setup - get an unlinked file living in tmpfs
4085 * @name: name for dentry (to be seen in /proc/<pid>/maps
4086 * @size: size to be set for the file
4087 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4088 */
4089struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4090{
4091 return __shmem_file_setup(name, size, flags, 0);
4092}
Keith Packard395e0dd2008-06-20 00:08:06 -07004093EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094
John Stultz6f193672011-12-15 18:51:04 -08004095void shmem_set_file(struct vm_area_struct *vma, struct file *file)
4096{
4097 if (vma->vm_file)
4098 fput(vma->vm_file);
4099 vma->vm_file = file;
4100 vma->vm_ops = &shmem_vm_ops;
4101}
4102
Randy Dunlap46711812008-03-19 17:00:41 -07004103/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4106 */
4107int shmem_zero_setup(struct vm_area_struct *vma)
4108{
4109 struct file *file;
4110 loff_t size = vma->vm_end - vma->vm_start;
4111
Hugh Dickins66fc1302015-06-14 09:48:09 -07004112 /*
4113 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4114 * between XFS directory reading and selinux: since this file is only
4115 * accessible to the user through its mapping, use S_PRIVATE flag to
4116 * bypass file security, in the same way as shmem_kernel_file_setup().
4117 */
4118 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119 if (IS_ERR(file))
4120 return PTR_ERR(file);
4121
John Stultz6f193672011-12-15 18:51:04 -08004122 shmem_set_file(vma, file);
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004123
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004124 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004125 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4126 (vma->vm_end & HPAGE_PMD_MASK)) {
4127 khugepaged_enter(vma, vma->vm_flags);
4128 }
4129
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130 return 0;
4131}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004132
4133/**
4134 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4135 * @mapping: the page's address_space
4136 * @index: the page index
4137 * @gfp: the page allocator flags to use if allocating
4138 *
4139 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4140 * with any new page allocations done using the specified allocation flags.
4141 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4142 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4143 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4144 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004145 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4146 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004147 */
4148struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4149 pgoff_t index, gfp_t gfp)
4150{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004151#ifdef CONFIG_SHMEM
4152 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004153 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004154 int error;
4155
4156 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004157 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4158 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004159 if (error)
4160 page = ERR_PTR(error);
4161 else
4162 unlock_page(page);
4163 return page;
4164#else
4165 /*
4166 * The tiny !SHMEM case uses ramfs without swap
4167 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004168 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004169#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004170}
4171EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);