blob: 69106c600692fa223c17bd520ac8ac26f3683a43 [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>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010032#include <linux/sched/signal.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040033#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080034#include <linux/swap.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080035#include <linux/uio.h>
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -070036#include <linux/khugepaged.h>
Mike Kravetz749df872017-09-06 16:24:16 -070037#include <linux/hugetlb.h>
Matt Mackall853ac432009-01-06 14:40:20 -080038
Andrea Arcangeli95cc09d2017-02-22 15:43:31 -080039#include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
40
Matt Mackall853ac432009-01-06 14:40:20 -080041static struct vfsmount *shm_mnt;
42
43#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * This virtual memory filesystem is heavily based on the ramfs. It
46 * extends ramfs by the ability to use swap and honor resource limits
47 * which makes it a completely usable filesystem.
48 */
49
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070050#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070051#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010052#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080053#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/string.h>
56#include <linux/slab.h>
57#include <linux/backing-dev.h>
58#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070061#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070062#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070063#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070064#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <linux/security.h>
66#include <linux/swapops.h>
67#include <linux/mempolicy.h>
68#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000069#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070070#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070071#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080072#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040073#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070074#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070075#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070076#include <uapi/linux/memfd.h>
Mike Rapoportcfda0522017-02-22 15:43:37 -080077#include <linux/userfaultfd_k.h>
Mike Rapoport4c27fe42017-02-22 15:43:25 -080078#include <linux/rmap.h>
Amir Goldstein2b4db792017-05-18 15:29:33 +030079#include <linux/uuid.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070080
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080081#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#include <asm/pgtable.h>
83
Mel Gormandd56b042015-11-06 16:28:43 -080084#include "internal.h"
85
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030086#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/* Pretend that each entry is of this size in directory's i_size */
90#define BOGO_DIRENT_SIZE 20
91
Hugh Dickins69f07ec2011-08-03 16:21:26 -070092/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93#define SHORT_SYMLINK_LEN 128
94
Hugh Dickins1aac1402012-05-29 15:06:42 -070095/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070096 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_mutex making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070099 */
100struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -0700101 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -0700102 pgoff_t start; /* start of range currently being fallocated */
103 pgoff_t next; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
106};
107
Andrew Mortonb76db732008-02-08 04:21:49 -0800108#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800109static unsigned long shmem_default_max_blocks(void)
110{
111 return totalram_pages / 2;
112}
113
114static unsigned long shmem_default_max_inodes(void)
115{
116 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117}
Andrew Mortonb76db732008-02-08 04:21:49 -0800118#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800119
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700120static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700123static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700124 struct page **pagep, enum sgp_type sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800125 gfp_t gfp, struct vm_area_struct *vma,
126 struct vm_fault *vmf, int *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700127
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700128int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700129 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700130{
131 return shmem_getpage_gfp(inode, index, pagep, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800132 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700133}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
136{
137 return sb->s_fs_info;
138}
139
140/*
141 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
142 * for shared memory and for shared anonymous (/dev/zero) mappings
143 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
144 * consistent with the pre-accounting of private mappings ...
145 */
146static inline int shmem_acct_size(unsigned long flags, loff_t size)
147{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000148 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000149 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
151
152static inline void shmem_unacct_size(unsigned long flags, loff_t size)
153{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000154 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 vm_unacct_memory(VM_ACCT(size));
156}
157
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700158static inline int shmem_reacct_size(unsigned long flags,
159 loff_t oldsize, loff_t newsize)
160{
161 if (!(flags & VM_NORESERVE)) {
162 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
163 return security_vm_enough_memory_mm(current->mm,
164 VM_ACCT(newsize) - VM_ACCT(oldsize));
165 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
166 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
167 }
168 return 0;
169}
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171/*
172 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700173 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
175 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
176 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700177static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700179 if (!(flags & VM_NORESERVE))
180 return 0;
181
182 return security_vm_enough_memory_mm(current->mm,
183 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184}
185
186static inline void shmem_unacct_blocks(unsigned long flags, long pages)
187{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000188 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300189 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
Mike Rapoport0f079692017-09-06 16:22:59 -0700192static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
193{
194 struct shmem_inode_info *info = SHMEM_I(inode);
195 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
196
197 if (shmem_acct_block(info->flags, pages))
198 return false;
199
200 if (sbinfo->max_blocks) {
201 if (percpu_counter_compare(&sbinfo->used_blocks,
202 sbinfo->max_blocks - pages) > 0)
203 goto unacct;
204 percpu_counter_add(&sbinfo->used_blocks, pages);
205 }
206
207 return true;
208
209unacct:
210 shmem_unacct_blocks(info->flags, pages);
211 return false;
212}
213
214static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
215{
216 struct shmem_inode_info *info = SHMEM_I(inode);
217 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
218
219 if (sbinfo->max_blocks)
220 percpu_counter_sub(&sbinfo->used_blocks, pages);
221 shmem_unacct_blocks(info->flags, pages);
222}
223
Hugh Dickins759b9772007-03-05 00:30:28 -0800224static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700225static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800226static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800227static const struct inode_operations shmem_inode_operations;
228static const struct inode_operations shmem_dir_inode_operations;
229static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400230static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700231static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Mike Rapoportb0506e42017-02-22 15:43:28 -0800233bool vma_is_shmem(struct vm_area_struct *vma)
234{
235 return vma->vm_ops == &shmem_vm_ops;
236}
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800239static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800241static int shmem_reserve_inode(struct super_block *sb)
242{
243 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
244 if (sbinfo->max_inodes) {
245 spin_lock(&sbinfo->stat_lock);
246 if (!sbinfo->free_inodes) {
247 spin_unlock(&sbinfo->stat_lock);
248 return -ENOSPC;
249 }
250 sbinfo->free_inodes--;
251 spin_unlock(&sbinfo->stat_lock);
252 }
253 return 0;
254}
255
256static void shmem_free_inode(struct super_block *sb)
257{
258 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
259 if (sbinfo->max_inodes) {
260 spin_lock(&sbinfo->stat_lock);
261 sbinfo->free_inodes++;
262 spin_unlock(&sbinfo->stat_lock);
263 }
264}
265
Randy Dunlap46711812008-03-19 17:00:41 -0700266/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700267 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 * @inode: inode to recalc
269 *
270 * We have to calculate the free blocks since the mm can drop
271 * undirtied hole pages behind our back.
272 *
273 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
274 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
275 *
276 * It has to be called with the spinlock held.
277 */
278static void shmem_recalc_inode(struct inode *inode)
279{
280 struct shmem_inode_info *info = SHMEM_I(inode);
281 long freed;
282
283 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
284 if (freed > 0) {
285 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700286 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Mike Rapoport0f079692017-09-06 16:22:59 -0700287 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
289}
290
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700291bool shmem_charge(struct inode *inode, long pages)
292{
293 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700294 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700295
Mike Rapoport0f079692017-09-06 16:22:59 -0700296 if (!shmem_inode_acct_block(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700297 return false;
Mike Rapoportb1cc94a2017-09-06 16:22:56 -0700298
Hugh Dickins98f1ae12018-11-30 14:10:29 -0800299 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
300 inode->i_mapping->nrpages += pages;
301
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700302 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700303 info->alloced += pages;
304 inode->i_blocks += pages * BLOCKS_PER_PAGE;
305 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700306 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700307
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700308 return true;
309}
310
311void shmem_uncharge(struct inode *inode, long pages)
312{
313 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700314 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700315
Hugh Dickins98f1ae12018-11-30 14:10:29 -0800316 /* nrpages adjustment done by __delete_from_page_cache() or caller */
317
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700318 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700319 info->alloced -= pages;
320 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
321 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700322 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700323
Mike Rapoport0f079692017-09-06 16:22:59 -0700324 shmem_inode_unacct_blocks(inode, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700325}
326
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700327/*
328 * Replace item expected in radix tree by a new item, while holding tree lock.
329 */
330static int shmem_radix_tree_replace(struct address_space *mapping,
331 pgoff_t index, void *expected, void *replacement)
332{
Johannes Weinerf7942432016-12-12 16:43:41 -0800333 struct radix_tree_node *node;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700334 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700335 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700336
337 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700338 VM_BUG_ON(!replacement);
Johannes Weinerf7942432016-12-12 16:43:41 -0800339 item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
340 if (!item)
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700341 return -ENOENT;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700342 if (item != expected)
343 return -ENOENT;
Johannes Weiner4d693d02016-12-12 16:43:49 -0800344 __radix_tree_replace(&mapping->page_tree, node, pslot,
345 replacement, NULL, NULL);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700346 return 0;
347}
348
349/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700350 * Sometimes, before we decide whether to proceed or to fail, we must check
351 * that an entry was not already brought back from swap by a racing thread.
352 *
353 * Checking page is not enough: by the time a SwapCache page is locked, it
354 * might be reused, and again be SwapCache, using the same swap as before.
355 */
356static bool shmem_confirm_swap(struct address_space *mapping,
357 pgoff_t index, swp_entry_t swap)
358{
359 void *item;
360
361 rcu_read_lock();
362 item = radix_tree_lookup(&mapping->page_tree, index);
363 rcu_read_unlock();
364 return item == swp_to_radix_entry(swap);
365}
366
367/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700368 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
369 *
370 * SHMEM_HUGE_NEVER:
371 * disables huge pages for the mount;
372 * SHMEM_HUGE_ALWAYS:
373 * enables huge pages for the mount;
374 * SHMEM_HUGE_WITHIN_SIZE:
375 * only allocate huge pages if the page will be fully within i_size,
376 * also respect fadvise()/madvise() hints;
377 * SHMEM_HUGE_ADVISE:
378 * only allocate huge pages if requested with fadvise()/madvise();
379 */
380
381#define SHMEM_HUGE_NEVER 0
382#define SHMEM_HUGE_ALWAYS 1
383#define SHMEM_HUGE_WITHIN_SIZE 2
384#define SHMEM_HUGE_ADVISE 3
385
386/*
387 * Special values.
388 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
389 *
390 * SHMEM_HUGE_DENY:
391 * disables huge on shm_mnt and all mounts, for emergency use;
392 * SHMEM_HUGE_FORCE:
393 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
394 *
395 */
396#define SHMEM_HUGE_DENY (-1)
397#define SHMEM_HUGE_FORCE (-2)
398
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700399#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700400/* ifdef here to avoid bloating shmem.o when not necessary */
401
402int shmem_huge __read_mostly;
403
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800404#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700405static int shmem_parse_huge(const char *str)
406{
407 if (!strcmp(str, "never"))
408 return SHMEM_HUGE_NEVER;
409 if (!strcmp(str, "always"))
410 return SHMEM_HUGE_ALWAYS;
411 if (!strcmp(str, "within_size"))
412 return SHMEM_HUGE_WITHIN_SIZE;
413 if (!strcmp(str, "advise"))
414 return SHMEM_HUGE_ADVISE;
415 if (!strcmp(str, "deny"))
416 return SHMEM_HUGE_DENY;
417 if (!strcmp(str, "force"))
418 return SHMEM_HUGE_FORCE;
419 return -EINVAL;
420}
421
422static const char *shmem_format_huge(int huge)
423{
424 switch (huge) {
425 case SHMEM_HUGE_NEVER:
426 return "never";
427 case SHMEM_HUGE_ALWAYS:
428 return "always";
429 case SHMEM_HUGE_WITHIN_SIZE:
430 return "within_size";
431 case SHMEM_HUGE_ADVISE:
432 return "advise";
433 case SHMEM_HUGE_DENY:
434 return "deny";
435 case SHMEM_HUGE_FORCE:
436 return "force";
437 default:
438 VM_BUG_ON(1);
439 return "bad_val";
440 }
441}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800442#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700443
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700444static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
445 struct shrink_control *sc, unsigned long nr_to_split)
446{
447 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800448 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700449 struct inode *inode;
450 struct shmem_inode_info *info;
451 struct page *page;
452 unsigned long batch = sc ? sc->nr_to_scan : 128;
453 int removed = 0, split = 0;
454
455 if (list_empty(&sbinfo->shrinklist))
456 return SHRINK_STOP;
457
458 spin_lock(&sbinfo->shrinklist_lock);
459 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
460 info = list_entry(pos, struct shmem_inode_info, shrinklist);
461
462 /* pin the inode */
463 inode = igrab(&info->vfs_inode);
464
465 /* inode is about to be evicted */
466 if (!inode) {
467 list_del_init(&info->shrinklist);
468 removed++;
469 goto next;
470 }
471
472 /* Check if there's anything to gain */
473 if (round_up(inode->i_size, PAGE_SIZE) ==
474 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800475 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700476 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700477 goto next;
478 }
479
480 list_move(&info->shrinklist, &list);
481next:
482 if (!--batch)
483 break;
484 }
485 spin_unlock(&sbinfo->shrinklist_lock);
486
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800487 list_for_each_safe(pos, next, &to_remove) {
488 info = list_entry(pos, struct shmem_inode_info, shrinklist);
489 inode = &info->vfs_inode;
490 list_del_init(&info->shrinklist);
491 iput(inode);
492 }
493
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700494 list_for_each_safe(pos, next, &list) {
495 int ret;
496
497 info = list_entry(pos, struct shmem_inode_info, shrinklist);
498 inode = &info->vfs_inode;
499
Kirill A. Shutemovf5dad042018-03-22 16:17:35 -0700500 if (nr_to_split && split >= nr_to_split)
501 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700502
Kirill A. Shutemovf5dad042018-03-22 16:17:35 -0700503 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700504 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
505 if (!page)
506 goto drop;
507
Kirill A. Shutemovf5dad042018-03-22 16:17:35 -0700508 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700509 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700510 put_page(page);
511 goto drop;
512 }
513
Kirill A. Shutemovf5dad042018-03-22 16:17:35 -0700514 /*
515 * Leave the inode on the list if we failed to lock
516 * the page at this time.
517 *
518 * Waiting for the lock may lead to deadlock in the
519 * reclaim path.
520 */
521 if (!trylock_page(page)) {
522 put_page(page);
523 goto leave;
524 }
525
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700526 ret = split_huge_page(page);
527 unlock_page(page);
528 put_page(page);
529
Kirill A. Shutemovf5dad042018-03-22 16:17:35 -0700530 /* If split failed leave the inode on the list */
531 if (ret)
532 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700533
534 split++;
535drop:
536 list_del_init(&info->shrinklist);
537 removed++;
Kirill A. Shutemovf5dad042018-03-22 16:17:35 -0700538leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700539 iput(inode);
540 }
541
542 spin_lock(&sbinfo->shrinklist_lock);
543 list_splice_tail(&list, &sbinfo->shrinklist);
544 sbinfo->shrinklist_len -= removed;
545 spin_unlock(&sbinfo->shrinklist_lock);
546
547 return split;
548}
549
550static long shmem_unused_huge_scan(struct super_block *sb,
551 struct shrink_control *sc)
552{
553 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
554
555 if (!READ_ONCE(sbinfo->shrinklist_len))
556 return SHRINK_STOP;
557
558 return shmem_unused_huge_shrink(sbinfo, sc, 0);
559}
560
561static long shmem_unused_huge_count(struct super_block *sb,
562 struct shrink_control *sc)
563{
564 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
565 return READ_ONCE(sbinfo->shrinklist_len);
566}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700567#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700568
569#define shmem_huge SHMEM_HUGE_DENY
570
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700571static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
572 struct shrink_control *sc, unsigned long nr_to_split)
573{
574 return 0;
575}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700576#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700577
578/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700579 * Like add_to_page_cache_locked, but error if expected item has gone.
580 */
581static int shmem_add_to_page_cache(struct page *page,
582 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700583 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700584{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700585 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700586
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700587 VM_BUG_ON_PAGE(PageTail(page), page);
588 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800589 VM_BUG_ON_PAGE(!PageLocked(page), page);
590 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700591 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700592
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700593 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700594 page->mapping = mapping;
595 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700596
Hugh Dickinsb065b432012-07-11 14:02:48 -0700597 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700598 if (PageTransHuge(page)) {
599 void __rcu **results;
600 pgoff_t idx;
601 int i;
602
603 error = 0;
604 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
605 &results, &idx, index, 1) &&
606 idx < index + HPAGE_PMD_NR) {
607 error = -EEXIST;
608 }
609
610 if (!error) {
611 for (i = 0; i < HPAGE_PMD_NR; i++) {
612 error = radix_tree_insert(&mapping->page_tree,
613 index + i, page + i);
614 VM_BUG_ON(error);
615 }
616 count_vm_event(THP_FILE_ALLOC);
617 }
618 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700619 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700620 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700621 error = shmem_radix_tree_replace(mapping, index, expected,
622 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700623 }
624
Hugh Dickinsb065b432012-07-11 14:02:48 -0700625 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700626 mapping->nrpages += nr;
627 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700628 __inc_node_page_state(page, NR_SHMEM_THPS);
629 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
630 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700631 spin_unlock_irq(&mapping->tree_lock);
632 } else {
633 page->mapping = NULL;
634 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700635 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700636 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700637 return error;
638}
639
640/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700641 * Like delete_from_page_cache, but substitutes swap for page.
642 */
643static void shmem_delete_from_page_cache(struct page *page, void *radswap)
644{
645 struct address_space *mapping = page->mapping;
646 int error;
647
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700648 VM_BUG_ON_PAGE(PageCompound(page), page);
649
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700650 spin_lock_irq(&mapping->tree_lock);
651 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
652 page->mapping = NULL;
653 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700654 __dec_node_page_state(page, NR_FILE_PAGES);
655 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700656 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300657 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700658 BUG_ON(error);
659}
660
661/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700662 * Remove swap entry from radix tree, free the swap and its page cache.
663 */
664static int shmem_free_swap(struct address_space *mapping,
665 pgoff_t index, void *radswap)
666{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700667 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700668
669 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700670 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700671 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700672 if (old != radswap)
673 return -ENOENT;
674 free_swap_and_cache(radix_to_swp_entry(radswap));
675 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700676}
677
678/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800679 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800680 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800681 *
682 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
683 * as long as the inode doesn't go away and racy results are not a problem.
684 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800685unsigned long shmem_partial_swap_usage(struct address_space *mapping,
686 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800687{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800688 struct radix_tree_iter iter;
689 void **slot;
690 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800691 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800692
693 rcu_read_lock();
694
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800695 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
696 if (iter.index >= end)
697 break;
698
699 page = radix_tree_deref_slot(slot);
700
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700701 if (radix_tree_deref_retry(page)) {
702 slot = radix_tree_iter_retry(&iter);
703 continue;
704 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800705
706 if (radix_tree_exceptional_entry(page))
707 swapped++;
708
709 if (need_resched()) {
Matthew Wilcox148deab2016-12-14 15:08:49 -0800710 slot = radix_tree_iter_resume(slot, &iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800711 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800712 }
713 }
714
715 rcu_read_unlock();
716
717 return swapped << PAGE_SHIFT;
718}
719
720/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800721 * Determine (in bytes) how many of the shmem object's pages mapped by the
722 * given vma is swapped out.
723 *
724 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
725 * as long as the inode doesn't go away and racy results are not a problem.
726 */
727unsigned long shmem_swap_usage(struct vm_area_struct *vma)
728{
729 struct inode *inode = file_inode(vma->vm_file);
730 struct shmem_inode_info *info = SHMEM_I(inode);
731 struct address_space *mapping = inode->i_mapping;
732 unsigned long swapped;
733
734 /* Be careful as we don't hold info->lock */
735 swapped = READ_ONCE(info->swapped);
736
737 /*
738 * The easier cases are when the shmem object has nothing in swap, or
739 * the vma maps it whole. Then we can simply use the stats that we
740 * already track.
741 */
742 if (!swapped)
743 return 0;
744
745 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
746 return swapped << PAGE_SHIFT;
747
748 /* Here comes the more involved part */
749 return shmem_partial_swap_usage(mapping,
750 linear_page_index(vma, vma->vm_start),
751 linear_page_index(vma, vma->vm_end));
752}
753
754/*
Hugh Dickins24513262012-01-20 14:34:21 -0800755 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
756 */
757void shmem_unlock_mapping(struct address_space *mapping)
758{
759 struct pagevec pvec;
760 pgoff_t indices[PAGEVEC_SIZE];
761 pgoff_t index = 0;
762
763 pagevec_init(&pvec, 0);
764 /*
765 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
766 */
767 while (!mapping_unevictable(mapping)) {
768 /*
769 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
770 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
771 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700772 pvec.nr = find_get_entries(mapping, index,
773 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800774 if (!pvec.nr)
775 break;
776 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700777 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800778 check_move_unevictable_pages(pvec.pages, pvec.nr);
779 pagevec_release(&pvec);
780 cond_resched();
781 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700782}
783
784/*
785 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700786 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700787 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700788static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
789 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700791 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300793 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
794 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
795 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
796 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700797 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700798 pgoff_t indices[PAGEVEC_SIZE];
799 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700800 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700801 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700803 if (lend == -1)
804 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700805
806 pagevec_init(&pvec, 0);
807 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700808 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700809 pvec.nr = find_get_entries(mapping, index,
810 min(end - index, (pgoff_t)PAGEVEC_SIZE),
811 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700812 if (!pvec.nr)
813 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700814 for (i = 0; i < pagevec_count(&pvec); i++) {
815 struct page *page = pvec.pages[i];
816
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700817 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700818 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700819 break;
820
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700821 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700822 if (unfalloc)
823 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700824 nr_swaps_freed += !shmem_free_swap(mapping,
825 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700826 continue;
827 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700828
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700829 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
830
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700831 if (!trylock_page(page))
832 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700833
834 if (PageTransTail(page)) {
835 /* Middle of THP: zero out the page */
836 clear_highpage(page);
837 unlock_page(page);
838 continue;
839 } else if (PageTransHuge(page)) {
840 if (index == round_down(end, HPAGE_PMD_NR)) {
841 /*
842 * Range ends in the middle of THP:
843 * zero out the page
844 */
845 clear_highpage(page);
846 unlock_page(page);
847 continue;
848 }
849 index += HPAGE_PMD_NR - 1;
850 i += HPAGE_PMD_NR - 1;
851 }
852
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700853 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700854 VM_BUG_ON_PAGE(PageTail(page), page);
855 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800856 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700857 truncate_inode_page(mapping, page);
858 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700859 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700860 unlock_page(page);
861 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700862 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800863 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700864 cond_resched();
865 index++;
866 }
867
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700868 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700869 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700870 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700871 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300872 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700873 if (start > end) {
874 top = partial_end;
875 partial_end = 0;
876 }
877 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700878 set_page_dirty(page);
879 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300880 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700881 }
882 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700883 if (partial_end) {
884 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700885 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700886 if (page) {
887 zero_user_segment(page, 0, partial_end);
888 set_page_dirty(page);
889 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300890 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700891 }
892 }
893 if (start >= end)
894 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700895
896 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700897 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700898 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700899
900 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700901 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700902 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700903 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700904 /* If all gone or hole-punch or unfalloc, we're done */
905 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700906 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700907 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700908 index = start;
909 continue;
910 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700911 for (i = 0; i < pagevec_count(&pvec); i++) {
912 struct page *page = pvec.pages[i];
913
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700914 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700915 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700916 break;
917
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700918 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700919 if (unfalloc)
920 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700921 if (shmem_free_swap(mapping, index, page)) {
922 /* Swap was replaced by page: retry */
923 index--;
924 break;
925 }
926 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700927 continue;
928 }
929
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700930 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700931
932 if (PageTransTail(page)) {
933 /* Middle of THP: zero out the page */
934 clear_highpage(page);
935 unlock_page(page);
936 /*
937 * Partial thp truncate due 'start' in middle
938 * of THP: don't need to look on these pages
939 * again on !pvec.nr restart.
940 */
941 if (index != round_down(end, HPAGE_PMD_NR))
942 start++;
943 continue;
944 } else if (PageTransHuge(page)) {
945 if (index == round_down(end, HPAGE_PMD_NR)) {
946 /*
947 * Range ends in the middle of THP:
948 * zero out the page
949 */
950 clear_highpage(page);
951 unlock_page(page);
952 continue;
953 }
954 index += HPAGE_PMD_NR - 1;
955 i += HPAGE_PMD_NR - 1;
956 }
957
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700958 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700959 VM_BUG_ON_PAGE(PageTail(page), page);
960 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800961 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700962 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700963 } else {
964 /* Page was replaced by swap: retry */
965 unlock_page(page);
966 index--;
967 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700968 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700969 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700970 unlock_page(page);
971 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700972 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800973 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700974 index++;
975 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700976
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700977 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700978 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700980 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700981}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700983void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
984{
985 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700986 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700988EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
David Howellsa528d352017-01-31 16:46:22 +0000990static int shmem_getattr(const struct path *path, struct kstat *stat,
991 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -0700992{
David Howellsa528d352017-01-31 16:46:22 +0000993 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -0700994 struct shmem_inode_info *info = SHMEM_I(inode);
995
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800996 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700997 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800998 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700999 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001000 }
Yu Zhao44a30222015-09-08 15:03:33 -07001001 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -07001002 return 0;
1003}
1004
Hugh Dickins94c1e622011-06-27 16:18:03 -07001005static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006{
David Howells75c3cfa2015-03-17 22:26:12 +00001007 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -07001008 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001009 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 int error;
1011
Jan Kara31051c82016-05-26 16:55:18 +02001012 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001013 if (error)
1014 return error;
1015
Hugh Dickins94c1e622011-06-27 16:18:03 -07001016 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1017 loff_t oldsize = inode->i_size;
1018 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001019
David Herrmann40e041a2014-08-08 14:25:27 -07001020 /* protected by i_mutex */
1021 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1022 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1023 return -EPERM;
1024
Hugh Dickins94c1e622011-06-27 16:18:03 -07001025 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001026 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1027 oldsize, newsize);
1028 if (error)
1029 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001030 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001031 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001032 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001033 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001034 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001035 if (oldsize > holebegin)
1036 unmap_mapping_range(inode->i_mapping,
1037 holebegin, 0, 1);
1038 if (info->alloced)
1039 shmem_truncate_range(inode,
1040 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001041 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001042 if (oldsize > holebegin)
1043 unmap_mapping_range(inode->i_mapping,
1044 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001045
1046 /*
1047 * Part of the huge page can be beyond i_size: subject
1048 * to shrink under memory pressure.
1049 */
1050 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1051 spin_lock(&sbinfo->shrinklist_lock);
Cong Wangd0413532017-08-10 15:24:24 -07001052 /*
1053 * _careful to defend against unlocked access to
1054 * ->shrink_list in shmem_unused_huge_shrink()
1055 */
1056 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001057 list_add_tail(&info->shrinklist,
1058 &sbinfo->shrinklist);
1059 sbinfo->shrinklist_len++;
1060 }
1061 spin_unlock(&sbinfo->shrinklist_lock);
1062 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
1065
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001066 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001067 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001068 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 return error;
1070}
1071
Al Viro1f895f72010-06-05 19:10:41 -04001072static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001075 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001077 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 shmem_unacct_size(info->flags, inode->i_size);
1079 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001080 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001081 if (!list_empty(&info->shrinklist)) {
1082 spin_lock(&sbinfo->shrinklist_lock);
1083 if (!list_empty(&info->shrinklist)) {
1084 list_del_init(&info->shrinklist);
1085 sbinfo->shrinklist_len--;
1086 }
1087 spin_unlock(&sbinfo->shrinklist_lock);
1088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001090 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001092 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 }
Al Viro3ed47db2016-01-22 18:08:52 -05001094 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001095
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001096 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001097 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001098 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001099 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100}
1101
Matthew Wilcox478922e2016-12-14 15:08:52 -08001102static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1103{
1104 struct radix_tree_iter iter;
1105 void **slot;
1106 unsigned long found = -1;
1107 unsigned int checked = 0;
1108
1109 rcu_read_lock();
1110 radix_tree_for_each_slot(slot, root, &iter, 0) {
1111 if (*slot == item) {
1112 found = iter.index;
1113 break;
1114 }
1115 checked++;
1116 if ((checked % 4096) != 0)
1117 continue;
1118 slot = radix_tree_iter_resume(slot, &iter);
1119 cond_resched_rcu();
1120 }
1121
1122 rcu_read_unlock();
1123 return found;
1124}
1125
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001126/*
1127 * If swap found in inode, free it and move page from swapcache to filecache.
1128 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001129static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001130 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001132 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001133 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001134 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001135 gfp_t gfp;
1136 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001138 radswap = swp_to_radix_entry(swap);
Matthew Wilcox478922e2016-12-14 15:08:52 -08001139 index = find_swap_entry(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001140 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001141 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001142
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001143 /*
1144 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001145 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001146 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001147 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001148 */
1149 if (shmem_swaplist.next != &info->swaplist)
1150 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001151
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001152 gfp = mapping_gfp_mask(mapping);
1153 if (shmem_should_replace_page(*pagep, gfp)) {
1154 mutex_unlock(&shmem_swaplist_mutex);
1155 error = shmem_replace_page(pagep, gfp, info, index);
1156 mutex_lock(&shmem_swaplist_mutex);
1157 /*
1158 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001159 * allocation, but the inode might have been freed while we
1160 * dropped it: although a racing shmem_evict_inode() cannot
1161 * complete without emptying the radix_tree, our page lock
1162 * on this swapcache page is not enough to prevent that -
1163 * free_swap_and_cache() of our swap entry will only
1164 * trylock_page(), removing swap from radix_tree whatever.
1165 *
1166 * We must not proceed to shmem_add_to_page_cache() if the
1167 * inode has been freed, but of course we cannot rely on
1168 * inode or mapping or info to check that. However, we can
1169 * safely check if our swap entry is still in use (and here
1170 * it can't have got reused for another page): if it's still
1171 * in use, then the inode cannot have been freed yet, and we
1172 * can safely proceed (if it's no longer in use, that tells
1173 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001174 */
1175 if (!page_swapcount(*pagep))
1176 error = -ENOENT;
1177 }
1178
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001179 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001180 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1181 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1182 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001183 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001184 if (!error)
1185 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001186 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001187 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001188 /*
1189 * Truncation and eviction use free_swap_and_cache(), which
1190 * only does trylock page: if we raced, best clean up here.
1191 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001192 delete_from_swap_cache(*pagep);
1193 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001194 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001195 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001196 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001197 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001198 swap_free(swap);
1199 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001201 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
1204/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001205 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001207int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001209 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001211 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001212 int error = 0;
1213
1214 /*
1215 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001216 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001217 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001218 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001219 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001220
1221 /*
1222 * Charge page using GFP_KERNEL while we can wait, before taking
1223 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1224 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001225 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001226 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1227 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001228 if (error)
1229 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001230 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001231 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001233 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001234 list_for_each_safe(this, next, &shmem_swaplist) {
1235 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001236 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001237 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001238 else
1239 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001240 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001241 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001242 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001243 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001245 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001246
Johannes Weiner00501b52014-08-08 14:19:20 -07001247 if (error) {
1248 if (error != -ENOMEM)
1249 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001250 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001251 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001252 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001253out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001254 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001255 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001256 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
1259/*
1260 * Move the page from the page cache to the swap cache.
1261 */
1262static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1263{
1264 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001267 swp_entry_t swap;
1268 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001270 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 mapping = page->mapping;
1273 index = page->index;
1274 inode = mapping->host;
1275 info = SHMEM_I(inode);
1276 if (info->flags & VM_LOCKED)
1277 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001278 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 goto redirty;
1280
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001281 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001282 * Our capabilities prevent regular writeback or sync from ever calling
1283 * shmem_writepage; but a stacking filesystem might use ->writepage of
1284 * its underlying filesystem, in which case tmpfs should write out to
1285 * swap only in response to memory pressure, and not for the writeback
1286 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001287 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001288 if (!wbc->for_reclaim) {
1289 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1290 goto redirty;
1291 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001292
1293 /*
1294 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1295 * value into swapfile.c, the only way we can correctly account for a
1296 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001297 *
1298 * That's okay for a page already fallocated earlier, but if we have
1299 * not yet completed the fallocation, then (a) we want to keep track
1300 * of this page in case we have to undo it, and (b) it may not be a
1301 * good idea to continue anyway, once we're pushing into swap. So
1302 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001303 */
1304 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001305 if (inode->i_private) {
1306 struct shmem_falloc *shmem_falloc;
1307 spin_lock(&inode->i_lock);
1308 shmem_falloc = inode->i_private;
1309 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001310 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001311 index >= shmem_falloc->start &&
1312 index < shmem_falloc->next)
1313 shmem_falloc->nr_unswapped++;
1314 else
1315 shmem_falloc = NULL;
1316 spin_unlock(&inode->i_lock);
1317 if (shmem_falloc)
1318 goto redirty;
1319 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001320 clear_highpage(page);
1321 flush_dcache_page(page);
1322 SetPageUptodate(page);
1323 }
1324
Huang Ying38d8b4e2017-07-06 15:37:18 -07001325 swap = get_swap_page(page);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001326 if (!swap.val)
1327 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001328
Vladimir Davydov37e84352016-01-20 15:02:56 -08001329 if (mem_cgroup_try_charge_swap(page, swap))
1330 goto free_swap;
1331
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001332 /*
1333 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001334 * if it's not already there. Do it now before the page is
1335 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001336 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001337 * we've incremented swapped, because shmem_unuse_inode() will
1338 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001339 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001340 mutex_lock(&shmem_swaplist_mutex);
1341 if (list_empty(&info->swaplist))
1342 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001343
Hugh Dickins48f170f2011-07-25 17:12:37 -07001344 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001345 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001346 shmem_recalc_inode(inode);
1347 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001348 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001349
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001350 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001351 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1352
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001353 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001354 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001355 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return 0;
1357 }
1358
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001359 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001360free_swap:
Minchan Kim75f6d6d2017-07-06 15:37:21 -07001361 put_swap_page(page, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362redirty:
1363 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001364 if (wbc->for_reclaim)
1365 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1366 unlock_page(page);
1367 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368}
1369
Hugh Dickins75edd342016-05-19 17:12:44 -07001370#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001371static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001372{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001373 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001374
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001375 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001376 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001377
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001378 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001379
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001380 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001381}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001382
1383static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1384{
1385 struct mempolicy *mpol = NULL;
1386 if (sbinfo->mpol) {
1387 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1388 mpol = sbinfo->mpol;
1389 mpol_get(mpol);
1390 spin_unlock(&sbinfo->stat_lock);
1391 }
1392 return mpol;
1393}
Hugh Dickins75edd342016-05-19 17:12:44 -07001394#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1395static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1396{
1397}
1398static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1399{
1400 return NULL;
1401}
1402#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1403#ifndef CONFIG_NUMA
1404#define vm_policy vm_private_data
1405#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001406
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001407static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1408 struct shmem_inode_info *info, pgoff_t index)
1409{
1410 /* Create a pseudo vma that just contains the policy */
1411 vma->vm_start = 0;
1412 /* Bias interleave by inode number to distribute better across nodes */
1413 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1414 vma->vm_ops = NULL;
1415 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1416}
1417
1418static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1419{
1420 /* Drop reference taken by mpol_shared_policy_lookup() */
1421 mpol_cond_put(vma->vm_policy);
1422}
1423
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001424static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1425 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001428 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001430 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001431 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001432 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001433
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001434 return page;
1435}
Mel Gorman18a2f372012-12-05 14:01:41 -08001436
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001437static struct page *shmem_alloc_hugepage(gfp_t gfp,
1438 struct shmem_inode_info *info, pgoff_t index)
1439{
1440 struct vm_area_struct pvma;
1441 struct inode *inode = &info->vfs_inode;
1442 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001443 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001444 void __rcu **results;
1445 struct page *page;
1446
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001447 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001448 return NULL;
1449
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001450 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001451 rcu_read_lock();
1452 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1453 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1454 rcu_read_unlock();
1455 return NULL;
1456 }
1457 rcu_read_unlock();
1458
1459 shmem_pseudo_vma_init(&pvma, info, hindex);
1460 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1461 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1462 shmem_pseudo_vma_destroy(&pvma);
1463 if (page)
1464 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001465 return page;
1466}
1467
1468static struct page *shmem_alloc_page(gfp_t gfp,
1469 struct shmem_inode_info *info, pgoff_t index)
1470{
1471 struct vm_area_struct pvma;
1472 struct page *page;
1473
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001474 shmem_pseudo_vma_init(&pvma, info, index);
1475 page = alloc_page_vma(gfp, &pvma, 0);
1476 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001477
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001478 return page;
1479}
1480
1481static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
Mike Rapoport0f079692017-09-06 16:22:59 -07001482 struct inode *inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001483 pgoff_t index, bool huge)
1484{
Mike Rapoport0f079692017-09-06 16:22:59 -07001485 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001486 struct page *page;
1487 int nr;
1488 int err = -ENOSPC;
1489
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001490 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001491 huge = false;
1492 nr = huge ? HPAGE_PMD_NR : 1;
1493
Mike Rapoport0f079692017-09-06 16:22:59 -07001494 if (!shmem_inode_acct_block(inode, nr))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001495 goto failed;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001496
1497 if (huge)
1498 page = shmem_alloc_hugepage(gfp, info, index);
1499 else
1500 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001501 if (page) {
1502 __SetPageLocked(page);
1503 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001504 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001505 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001506
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001507 err = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07001508 shmem_inode_unacct_blocks(inode, nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001509failed:
1510 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001514 * When a page is moved from swapcache to shmem filecache (either by the
1515 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1516 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1517 * ignorance of the mapping it belongs to. If that mapping has special
1518 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1519 * we may need to copy to a suitable page before moving to filecache.
1520 *
1521 * In a future release, this may well be extended to respect cpuset and
1522 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1523 * but for now it is a simple matter of zone.
1524 */
1525static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1526{
1527 return page_zonenum(page) > gfp_zone(gfp);
1528}
1529
1530static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1531 struct shmem_inode_info *info, pgoff_t index)
1532{
1533 struct page *oldpage, *newpage;
1534 struct address_space *swap_mapping;
Yu Zhao7c1ba1a2018-11-30 14:09:03 -08001535 swp_entry_t entry;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001536 pgoff_t swap_index;
1537 int error;
1538
1539 oldpage = *pagep;
Yu Zhao7c1ba1a2018-11-30 14:09:03 -08001540 entry.val = page_private(oldpage);
1541 swap_index = swp_offset(entry);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001542 swap_mapping = page_mapping(oldpage);
1543
1544 /*
1545 * We have arrived here because our zones are constrained, so don't
1546 * limit chance of success by further cpuset and node constraints.
1547 */
1548 gfp &= ~GFP_CONSTRAINT_MASK;
1549 newpage = shmem_alloc_page(gfp, info, index);
1550 if (!newpage)
1551 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001552
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001553 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001554 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001555 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001556
Hugh Dickins9956edf2016-11-10 10:46:11 -08001557 __SetPageLocked(newpage);
1558 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001559 SetPageUptodate(newpage);
Yu Zhao7c1ba1a2018-11-30 14:09:03 -08001560 set_page_private(newpage, entry.val);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001561 SetPageSwapCache(newpage);
1562
1563 /*
1564 * Our caller will very soon move newpage out of swapcache, but it's
1565 * a nice clean interface for us to replace oldpage by newpage there.
1566 */
1567 spin_lock_irq(&swap_mapping->tree_lock);
1568 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1569 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001570 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001571 __inc_node_page_state(newpage, NR_FILE_PAGES);
1572 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001573 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001574 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001575
Hugh Dickins0142ef62012-06-07 14:21:09 -07001576 if (unlikely(error)) {
1577 /*
1578 * Is this possible? I think not, now that our callers check
1579 * both PageSwapCache and page_private after getting page lock;
1580 * but be defensive. Reverse old to newpage for clear and free.
1581 */
1582 oldpage = newpage;
1583 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001584 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001585 lru_cache_add_anon(newpage);
1586 *pagep = newpage;
1587 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001588
1589 ClearPageSwapCache(oldpage);
1590 set_page_private(oldpage, 0);
1591
1592 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001593 put_page(oldpage);
1594 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001595 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001596}
1597
1598/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001599 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 *
1601 * If we allocate a new one we do not mark it dirty. That's up to the
1602 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001603 * entry since a page cannot live in both the swap and page cache.
1604 *
1605 * fault_mm and fault_type are only supplied by shmem_fault:
1606 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001608static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001609 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08001610 struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
1612 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann23f919d2016-12-12 16:42:28 -08001613 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001615 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001616 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001617 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001619 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001620 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001622 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001623 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001625 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001627 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1628 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001630 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001631 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001632 if (radix_tree_exceptional_entry(page)) {
1633 swap = radix_to_swp_entry(page);
1634 page = NULL;
1635 }
1636
Hugh Dickins75edd342016-05-19 17:12:44 -07001637 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001638 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001639 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001640 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001641 }
1642
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001643 if (page && sgp == SGP_WRITE)
1644 mark_page_accessed(page);
1645
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001646 /* fallocated page? */
1647 if (page && !PageUptodate(page)) {
1648 if (sgp != SGP_READ)
1649 goto clear;
1650 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001651 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001652 page = NULL;
1653 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001654 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001655 *pagep = page;
1656 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001660 * Fast cache lookup did not find it:
1661 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001663 sbinfo = SHMEM_SB(inode->i_sb);
Mike Rapoportcfda0522017-02-22 15:43:37 -08001664 charge_mm = vma ? vma->vm_mm : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 if (swap.val) {
1667 /* Look it up and read it in.. */
Huang Yingec560172017-09-06 16:24:36 -07001668 page = lookup_swap_cache(swap, NULL, 0);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001669 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001670 /* Or update major stats only when swapin succeeds?? */
1671 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001672 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001673 count_vm_event(PGMAJFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07001674 count_memcg_event_mm(charge_mm, PGMAJFAULT);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001675 }
1676 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001677 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001678 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001679 error = -ENOMEM;
1680 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683
1684 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001685 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001686 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001687 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001688 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001689 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001690 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001691 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001693 goto failed;
1694 }
1695 wait_on_page_writeback(page);
1696
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001697 if (shmem_should_replace_page(page, gfp)) {
1698 error = shmem_replace_page(&page, gfp, info, index);
1699 if (error)
1700 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 }
1702
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001703 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001704 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001705 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001706 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001707 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001708 /*
1709 * We already confirmed swap under page lock, and make
1710 * no memory allocation here, so usually no possibility
1711 * of error; but free_swap_and_cache() only trylocks a
1712 * page, so it is just possible that the entry has been
1713 * truncated or holepunched since swap was confirmed.
1714 * shmem_undo_range() will have done some of the
1715 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001716 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001717 * Reset swap.val? No, leave it so "failed" goes back to
1718 * "repeat": reading a hole and writing should succeed.
1719 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001720 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001721 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001722 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001723 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001724 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001725 if (error)
1726 goto failed;
1727
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001728 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001729
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001730 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001731 info->swapped--;
1732 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001733 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001734
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001735 if (sgp == SGP_WRITE)
1736 mark_page_accessed(page);
1737
Hugh Dickins27ab7002011-07-25 17:12:36 -07001738 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001739 set_page_dirty(page);
1740 swap_free(swap);
1741
Hugh Dickins54af60422011-08-03 16:21:24 -07001742 } else {
Mike Rapoportcfda0522017-02-22 15:43:37 -08001743 if (vma && userfaultfd_missing(vma)) {
1744 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1745 return 0;
1746 }
1747
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001748 /* shmem_symlink() */
1749 if (mapping->a_ops != &shmem_aops)
1750 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001751 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001752 goto alloc_nohuge;
1753 if (shmem_huge == SHMEM_HUGE_FORCE)
1754 goto alloc_huge;
1755 switch (sbinfo->huge) {
1756 loff_t i_size;
1757 pgoff_t off;
1758 case SHMEM_HUGE_NEVER:
1759 goto alloc_nohuge;
1760 case SHMEM_HUGE_WITHIN_SIZE:
1761 off = round_up(index, HPAGE_PMD_NR);
1762 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1763 if (i_size >= HPAGE_PMD_SIZE &&
1764 i_size >> PAGE_SHIFT >= off)
1765 goto alloc_huge;
1766 /* fallthrough */
1767 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001768 if (sgp_huge == SGP_HUGE)
1769 goto alloc_huge;
1770 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001771 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001774alloc_huge:
Mike Rapoport0f079692017-09-06 16:22:59 -07001775 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001776 if (IS_ERR(page)) {
Mike Rapoport0f079692017-09-06 16:22:59 -07001777alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001778 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001779 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001780 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001781 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001782 error = PTR_ERR(page);
1783 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001784 if (error != -ENOSPC)
1785 goto failed;
1786 /*
1787 * Try to reclaim some spece by splitting a huge page
1788 * beyond i_size on the filesystem.
1789 */
1790 while (retry--) {
1791 int ret;
1792 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1793 if (ret == SHRINK_STOP)
1794 break;
1795 if (ret)
1796 goto alloc_nohuge;
1797 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001798 goto failed;
1799 }
1800
1801 if (PageTransHuge(page))
1802 hindex = round_down(index, HPAGE_PMD_NR);
1803 else
1804 hindex = index;
1805
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001806 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001807 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001808
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001809 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001810 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001811 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001812 goto unacct;
1813 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1814 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001815 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001816 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001817 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001818 radix_tree_preload_end();
1819 }
1820 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001821 mem_cgroup_cancel_charge(page, memcg,
1822 PageTransHuge(page));
1823 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001824 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001825 mem_cgroup_commit_charge(page, memcg, false,
1826 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001827 lru_cache_add_anon(page);
1828
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001829 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001830 info->alloced += 1 << compound_order(page);
1831 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001832 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001833 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001834 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001835
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001836 if (PageTransHuge(page) &&
1837 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1838 hindex + HPAGE_PMD_NR - 1) {
1839 /*
1840 * Part of the huge page is beyond i_size: subject
1841 * to shrink under memory pressure.
1842 */
1843 spin_lock(&sbinfo->shrinklist_lock);
Cong Wangd0413532017-08-10 15:24:24 -07001844 /*
1845 * _careful to defend against unlocked access to
1846 * ->shrink_list in shmem_unused_huge_shrink()
1847 */
1848 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001849 list_add_tail(&info->shrinklist,
1850 &sbinfo->shrinklist);
1851 sbinfo->shrinklist_len++;
1852 }
1853 spin_unlock(&sbinfo->shrinklist_lock);
1854 }
1855
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001856 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001857 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1858 */
1859 if (sgp == SGP_FALLOC)
1860 sgp = SGP_WRITE;
1861clear:
1862 /*
1863 * Let SGP_WRITE caller clear ends if write does not fill page;
1864 * but SGP_FALLOC on a page fallocated earlier must initialize
1865 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001866 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001867 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1868 struct page *head = compound_head(page);
1869 int i;
1870
1871 for (i = 0; i < (1 << compound_order(head)); i++) {
1872 clear_highpage(head + i);
1873 flush_dcache_page(head + i);
1874 }
1875 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001876 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001877 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001878
Hugh Dickins54af60422011-08-03 16:21:24 -07001879 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001880 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001881 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001882 if (alloced) {
1883 ClearPageDirty(page);
1884 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001885 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001886 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001887 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001888 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001889 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001890 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001891 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001892 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001893 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001894
Nick Piggind0217ac2007-07-19 01:47:03 -07001895 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001896 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001898unacct:
Mike Rapoport0f079692017-09-06 16:22:59 -07001899 shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001900
1901 if (PageTransHuge(page)) {
1902 unlock_page(page);
1903 put_page(page);
1904 goto alloc_nohuge;
1905 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001906failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001907 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001908 error = -EEXIST;
1909unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001910 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001911 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001912 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001913 }
1914 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001915 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001916 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001917 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001918 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001919 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001920 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001921 goto repeat;
1922 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923}
1924
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001925/*
1926 * This is like autoremove_wake_function, but it removes the wait queue
1927 * entry unconditionally - even if something else had already woken the
1928 * target.
1929 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02001930static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001931{
1932 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02001933 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001934 return ret;
1935}
1936
Dave Jiang11bac802017-02-24 14:56:41 -08001937static int shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938{
Dave Jiang11bac802017-02-24 14:56:41 -08001939 struct vm_area_struct *vma = vmf->vma;
Al Viro496ad9a2013-01-23 17:07:38 -05001940 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001941 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001942 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001944 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001946 /*
1947 * Trinity finds that probing a hole which tmpfs is punching can
1948 * prevent the hole-punch from ever completing: which in turn
1949 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001950 * faulting pages into the hole while it's being punched. Although
1951 * shmem_undo_range() does remove the additions, it may be unable to
1952 * keep up, as each new page needs its own unmap_mapping_range() call,
1953 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1954 *
1955 * It does not matter if we sometimes reach this check just before the
1956 * hole-punch begins, so that one fault then races with the punch:
1957 * we just need to make racing faults a rare case.
1958 *
1959 * The implementation below would be much simpler if we just used a
1960 * standard mutex or completion: but we cannot take i_mutex in fault,
1961 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001962 */
1963 if (unlikely(inode->i_private)) {
1964 struct shmem_falloc *shmem_falloc;
1965
1966 spin_lock(&inode->i_lock);
1967 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001968 if (shmem_falloc &&
1969 shmem_falloc->waitq &&
1970 vmf->pgoff >= shmem_falloc->start &&
1971 vmf->pgoff < shmem_falloc->next) {
1972 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001973 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001974
1975 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001976 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1977 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001978 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001979 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001980 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001981 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001982
1983 shmem_falloc_waitq = shmem_falloc->waitq;
1984 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1985 TASK_UNINTERRUPTIBLE);
1986 spin_unlock(&inode->i_lock);
1987 schedule();
1988
1989 /*
1990 * shmem_falloc_waitq points into the shmem_fallocate()
1991 * stack of the hole-punching task: shmem_falloc_waitq
1992 * is usually invalid by the time we reach here, but
1993 * finish_wait() does not dereference it in that case;
1994 * though i_lock needed lest racing with wake_up_all().
1995 */
1996 spin_lock(&inode->i_lock);
1997 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1998 spin_unlock(&inode->i_lock);
1999 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002000 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07002001 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002002 }
2003
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002004 sgp = SGP_CACHE;
Michal Hocko18600332017-07-10 15:48:02 -07002005
2006 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2007 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002008 sgp = SGP_NOHUGE;
Michal Hocko18600332017-07-10 15:48:02 -07002009 else if (vma->vm_flags & VM_HUGEPAGE)
2010 sgp = SGP_HUGE;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002011
2012 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08002013 gfp, vma, vmf, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 if (error)
2015 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002016 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017}
2018
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002019unsigned long shmem_get_unmapped_area(struct file *file,
2020 unsigned long uaddr, unsigned long len,
2021 unsigned long pgoff, unsigned long flags)
2022{
2023 unsigned long (*get_area)(struct file *,
2024 unsigned long, unsigned long, unsigned long, unsigned long);
2025 unsigned long addr;
2026 unsigned long offset;
2027 unsigned long inflated_len;
2028 unsigned long inflated_addr;
2029 unsigned long inflated_offset;
2030
2031 if (len > TASK_SIZE)
2032 return -ENOMEM;
2033
2034 get_area = current->mm->get_unmapped_area;
2035 addr = get_area(file, uaddr, len, pgoff, flags);
2036
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002037 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002038 return addr;
2039 if (IS_ERR_VALUE(addr))
2040 return addr;
2041 if (addr & ~PAGE_MASK)
2042 return addr;
2043 if (addr > TASK_SIZE - len)
2044 return addr;
2045
2046 if (shmem_huge == SHMEM_HUGE_DENY)
2047 return addr;
2048 if (len < HPAGE_PMD_SIZE)
2049 return addr;
2050 if (flags & MAP_FIXED)
2051 return addr;
2052 /*
2053 * Our priority is to support MAP_SHARED mapped hugely;
2054 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2055 * But if caller specified an address hint, respect that as before.
2056 */
2057 if (uaddr)
2058 return addr;
2059
2060 if (shmem_huge != SHMEM_HUGE_FORCE) {
2061 struct super_block *sb;
2062
2063 if (file) {
2064 VM_BUG_ON(file->f_op != &shmem_file_operations);
2065 sb = file_inode(file)->i_sb;
2066 } else {
2067 /*
2068 * Called directly from mm/mmap.c, or drivers/char/mem.c
2069 * for "/dev/zero", to create a shared anonymous object.
2070 */
2071 if (IS_ERR(shm_mnt))
2072 return addr;
2073 sb = shm_mnt->mnt_sb;
2074 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002075 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002076 return addr;
2077 }
2078
2079 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2080 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2081 return addr;
2082 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2083 return addr;
2084
2085 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2086 if (inflated_len > TASK_SIZE)
2087 return addr;
2088 if (inflated_len < len)
2089 return addr;
2090
2091 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2092 if (IS_ERR_VALUE(inflated_addr))
2093 return addr;
2094 if (inflated_addr & ~PAGE_MASK)
2095 return addr;
2096
2097 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2098 inflated_addr += offset - inflated_offset;
2099 if (inflated_offset > offset)
2100 inflated_addr += HPAGE_PMD_SIZE;
2101
2102 if (inflated_addr > TASK_SIZE - len)
2103 return addr;
2104 return inflated_addr;
2105}
2106
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002108static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109{
Al Viro496ad9a2013-01-23 17:07:38 -05002110 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002111 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112}
2113
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002114static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2115 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116{
Al Viro496ad9a2013-01-23 17:07:38 -05002117 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002118 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002120 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2121 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122}
2123#endif
2124
2125int shmem_lock(struct file *file, int lock, struct user_struct *user)
2126{
Al Viro496ad9a2013-01-23 17:07:38 -05002127 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 struct shmem_inode_info *info = SHMEM_I(inode);
2129 int retval = -ENOMEM;
2130
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002131 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (lock && !(info->flags & VM_LOCKED)) {
2133 if (!user_shm_lock(inode->i_size, user))
2134 goto out_nomem;
2135 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002136 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 }
2138 if (!lock && (info->flags & VM_LOCKED) && user) {
2139 user_shm_unlock(inode->i_size, user);
2140 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002141 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 }
2143 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002144
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002146 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 return retval;
2148}
2149
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002150static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151{
2152 file_accessed(file);
2153 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002154 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002155 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2156 (vma->vm_end & HPAGE_PMD_MASK)) {
2157 khugepaged_enter(vma, vma->vm_flags);
2158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 return 0;
2160}
2161
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002162static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002163 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164{
2165 struct inode *inode;
2166 struct shmem_inode_info *info;
2167 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2168
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002169 if (shmem_reserve_inode(sb))
2170 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
2172 inode = new_inode(sb);
2173 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002174 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002175 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002177 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002178 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 info = SHMEM_I(inode);
2180 memset(info, 0, (char *)inode - (char *)info);
2181 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002182 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002183 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002184 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002186 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002187 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188
2189 switch (mode & S_IFMT) {
2190 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002191 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 init_special_inode(inode, mode, dev);
2193 break;
2194 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002195 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 inode->i_op = &shmem_inode_operations;
2197 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002198 mpol_shared_policy_init(&info->policy,
2199 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 break;
2201 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002202 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 /* Some things misbehave if size == 0 on a directory */
2204 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2205 inode->i_op = &shmem_dir_inode_operations;
2206 inode->i_fop = &simple_dir_operations;
2207 break;
2208 case S_IFLNK:
2209 /*
2210 * Must not load anything in the rbtree,
2211 * mpol_free_shared_policy will not be called.
2212 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002213 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 break;
2215 }
Joel Fernandes (Google)6447b342018-09-20 12:22:39 -07002216
2217 lockdep_annotate_inode_mutex_key(inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002218 } else
2219 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 return inode;
2221}
2222
Johannes Weiner0cd61442014-04-03 14:47:46 -07002223bool shmem_mapping(struct address_space *mapping)
2224{
Hugh Dickinsf8005452017-02-22 15:41:42 -08002225 return mapping->a_ops == &shmem_aops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002226}
2227
Mike Rapoport8d103962017-09-06 16:23:02 -07002228static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2229 pmd_t *dst_pmd,
2230 struct vm_area_struct *dst_vma,
2231 unsigned long dst_addr,
2232 unsigned long src_addr,
2233 bool zeropage,
2234 struct page **pagep)
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002235{
2236 struct inode *inode = file_inode(dst_vma->vm_file);
2237 struct shmem_inode_info *info = SHMEM_I(inode);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002238 struct address_space *mapping = inode->i_mapping;
2239 gfp_t gfp = mapping_gfp_mask(mapping);
2240 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2241 struct mem_cgroup *memcg;
2242 spinlock_t *ptl;
2243 void *page_kaddr;
2244 struct page *page;
2245 pte_t _dst_pte, *dst_pte;
2246 int ret;
Andrea Arcangeliaf3edb32018-11-30 14:09:37 -08002247 pgoff_t offset, max_off;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002248
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002249 ret = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07002250 if (!shmem_inode_acct_block(inode, 1))
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002251 goto out;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002252
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002253 if (!*pagep) {
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002254 page = shmem_alloc_page(gfp, info, pgoff);
2255 if (!page)
Mike Rapoport0f079692017-09-06 16:22:59 -07002256 goto out_unacct_blocks;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002257
Mike Rapoport8d103962017-09-06 16:23:02 -07002258 if (!zeropage) { /* mcopy_atomic */
2259 page_kaddr = kmap_atomic(page);
2260 ret = copy_from_user(page_kaddr,
2261 (const void __user *)src_addr,
2262 PAGE_SIZE);
2263 kunmap_atomic(page_kaddr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002264
Mike Rapoport8d103962017-09-06 16:23:02 -07002265 /* fallback to copy_from_user outside mmap_sem */
2266 if (unlikely(ret)) {
2267 *pagep = page;
2268 shmem_inode_unacct_blocks(inode, 1);
2269 /* don't free the page */
Andrea Arcangeli82c5a8c2018-11-30 14:09:25 -08002270 return -ENOENT;
Mike Rapoport8d103962017-09-06 16:23:02 -07002271 }
2272 } else { /* mfill_zeropage_atomic */
2273 clear_highpage(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002274 }
2275 } else {
2276 page = *pagep;
2277 *pagep = NULL;
2278 }
2279
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002280 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2281 __SetPageLocked(page);
2282 __SetPageSwapBacked(page);
Andrea Arcangelia425d352017-02-22 15:43:52 -08002283 __SetPageUptodate(page);
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002284
Andrea Arcangeliaf3edb32018-11-30 14:09:37 -08002285 ret = -EFAULT;
2286 offset = linear_page_index(dst_vma, dst_addr);
2287 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2288 if (unlikely(offset >= max_off))
2289 goto out_release;
2290
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002291 ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false);
2292 if (ret)
2293 goto out_release;
2294
2295 ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2296 if (!ret) {
2297 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2298 radix_tree_preload_end();
2299 }
2300 if (ret)
2301 goto out_release_uncharge;
2302
2303 mem_cgroup_commit_charge(page, memcg, false, false);
2304
2305 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2306 if (dst_vma->vm_flags & VM_WRITE)
2307 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
Andrea Arcangeli46466e22018-11-30 14:09:43 -08002308 else {
2309 /*
2310 * We don't set the pte dirty if the vma has no
2311 * VM_WRITE permission, so mark the page dirty or it
2312 * could be freed from under us. We could do it
2313 * unconditionally before unlock_page(), but doing it
2314 * only if VM_WRITE is not set is faster.
2315 */
2316 set_page_dirty(page);
2317 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002318
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002319 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
Andrea Arcangeliaf3edb32018-11-30 14:09:37 -08002320
2321 ret = -EFAULT;
2322 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2323 if (unlikely(offset >= max_off))
2324 goto out_release_uncharge_unlock;
2325
2326 ret = -EEXIST;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002327 if (!pte_none(*dst_pte))
2328 goto out_release_uncharge_unlock;
2329
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002330 lru_cache_add_anon(page);
2331
2332 spin_lock(&info->lock);
2333 info->alloced++;
2334 inode->i_blocks += BLOCKS_PER_PAGE;
2335 shmem_recalc_inode(inode);
2336 spin_unlock(&info->lock);
2337
2338 inc_mm_counter(dst_mm, mm_counter_file(page));
2339 page_add_file_rmap(page, false);
2340 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2341
2342 /* No need to invalidate - it was non-present before */
2343 update_mmu_cache(dst_vma, dst_addr, dst_pte);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002344 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangeliaf3edb32018-11-30 14:09:37 -08002345 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002346 ret = 0;
2347out:
2348 return ret;
2349out_release_uncharge_unlock:
2350 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangeli46466e22018-11-30 14:09:43 -08002351 ClearPageDirty(page);
Andrea Arcangeliaf3edb32018-11-30 14:09:37 -08002352 delete_from_page_cache(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002353out_release_uncharge:
2354 mem_cgroup_cancel_charge(page, memcg, false);
2355out_release:
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002356 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002357 put_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002358out_unacct_blocks:
Mike Rapoport0f079692017-09-06 16:22:59 -07002359 shmem_inode_unacct_blocks(inode, 1);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002360 goto out;
2361}
2362
Mike Rapoport8d103962017-09-06 16:23:02 -07002363int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2364 pmd_t *dst_pmd,
2365 struct vm_area_struct *dst_vma,
2366 unsigned long dst_addr,
2367 unsigned long src_addr,
2368 struct page **pagep)
2369{
2370 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2371 dst_addr, src_addr, false, pagep);
2372}
2373
2374int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2375 pmd_t *dst_pmd,
2376 struct vm_area_struct *dst_vma,
2377 unsigned long dst_addr)
2378{
2379 struct page *page = NULL;
2380
2381 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2382 dst_addr, 0, true, &page);
2383}
2384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002386static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002387static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002389#ifdef CONFIG_TMPFS_XATTR
2390static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2391#else
2392#define shmem_initxattrs NULL
2393#endif
2394
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002396shmem_write_begin(struct file *file, struct address_space *mapping,
2397 loff_t pos, unsigned len, unsigned flags,
2398 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399{
Nick Piggin800d15a2007-10-16 01:25:03 -07002400 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002401 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002402 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002403
2404 /* i_mutex is held by caller */
Steven Rostedt (VMware)3f472cc2017-02-24 14:59:10 -08002405 if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) {
David Herrmann40e041a2014-08-08 14:25:27 -07002406 if (info->seals & F_SEAL_WRITE)
2407 return -EPERM;
2408 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2409 return -EPERM;
2410 }
2411
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002412 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002413}
2414
2415static int
2416shmem_write_end(struct file *file, struct address_space *mapping,
2417 loff_t pos, unsigned len, unsigned copied,
2418 struct page *page, void *fsdata)
2419{
2420 struct inode *inode = mapping->host;
2421
Hugh Dickinsd3602442008-02-04 22:28:44 -08002422 if (pos + copied > inode->i_size)
2423 i_size_write(inode, pos + copied);
2424
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002425 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002426 struct page *head = compound_head(page);
2427 if (PageTransCompound(page)) {
2428 int i;
2429
2430 for (i = 0; i < HPAGE_PMD_NR; i++) {
2431 if (head + i == page)
2432 continue;
2433 clear_highpage(head + i);
2434 flush_dcache_page(head + i);
2435 }
2436 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002437 if (copied < PAGE_SIZE) {
2438 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002439 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002440 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002441 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002442 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002443 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002444 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002445 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002446 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002447
Nick Piggin800d15a2007-10-16 01:25:03 -07002448 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449}
2450
Al Viro2ba5bbe2014-04-02 20:00:02 -04002451static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452{
Al Viro6e58e792014-02-03 17:07:03 -05002453 struct file *file = iocb->ki_filp;
2454 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002456 pgoff_t index;
2457 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002458 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002459 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002460 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002461 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002462
2463 /*
2464 * Might this read be for a stacking filesystem? Then when reading
2465 * holes of a sparse file, we actually need to allocate those pages,
2466 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2467 */
Al Viro777eda22014-12-17 04:46:46 -05002468 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002469 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002471 index = *ppos >> PAGE_SHIFT;
2472 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473
2474 for (;;) {
2475 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002476 pgoff_t end_index;
2477 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 loff_t i_size = i_size_read(inode);
2479
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002480 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 if (index > end_index)
2482 break;
2483 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002484 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (nr <= offset)
2486 break;
2487 }
2488
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002489 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002490 if (error) {
2491 if (error == -EINVAL)
2492 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 break;
2494 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002495 if (page) {
2496 if (sgp == SGP_CACHE)
2497 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002498 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500
2501 /*
2502 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002503 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002505 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002507 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002509 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 if (nr <= offset) {
2511 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002512 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 break;
2514 }
2515 }
2516 nr -= offset;
2517
2518 if (page) {
2519 /*
2520 * If users can be writing to this page using arbitrary
2521 * virtual addresses, take care about potential aliasing
2522 * before reading the page on the kernel side.
2523 */
2524 if (mapping_writably_mapped(mapping))
2525 flush_dcache_page(page);
2526 /*
2527 * Mark the page accessed if we read the beginning.
2528 */
2529 if (!offset)
2530 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002531 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002533 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
2536 /*
2537 * Ok, we have the page, and it's up-to-date, so
2538 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002540 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002541 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002543 index += offset >> PAGE_SHIFT;
2544 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002546 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002547 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 break;
Al Viro6e58e792014-02-03 17:07:03 -05002549 if (ret < nr) {
2550 error = -EFAULT;
2551 break;
2552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 cond_resched();
2554 }
2555
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002556 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002557 file_accessed(file);
2558 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559}
2560
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002561/*
2562 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2563 */
2564static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002565 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002566{
2567 struct page *page;
2568 struct pagevec pvec;
2569 pgoff_t indices[PAGEVEC_SIZE];
2570 bool done = false;
2571 int i;
2572
2573 pagevec_init(&pvec, 0);
2574 pvec.nr = 1; /* start small: we may be there already */
2575 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002576 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002577 pvec.nr, pvec.pages, indices);
2578 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002579 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002580 index = end;
2581 break;
2582 }
2583 for (i = 0; i < pvec.nr; i++, index++) {
2584 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002585 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002586 done = true;
2587 break;
2588 }
2589 index = indices[i];
2590 }
2591 page = pvec.pages[i];
2592 if (page && !radix_tree_exceptional_entry(page)) {
2593 if (!PageUptodate(page))
2594 page = NULL;
2595 }
2596 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002597 (page && whence == SEEK_DATA) ||
2598 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002599 done = true;
2600 break;
2601 }
2602 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002603 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002604 pagevec_release(&pvec);
2605 pvec.nr = PAGEVEC_SIZE;
2606 cond_resched();
2607 }
2608 return index;
2609}
2610
Andrew Morton965c8e52012-12-17 15:59:39 -08002611static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002612{
2613 struct address_space *mapping = file->f_mapping;
2614 struct inode *inode = mapping->host;
2615 pgoff_t start, end;
2616 loff_t new_offset;
2617
Andrew Morton965c8e52012-12-17 15:59:39 -08002618 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2619 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002620 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002621 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002622 /* We're holding i_mutex so we can access i_size directly */
2623
Yufen Yua84c8722018-11-16 15:08:39 -08002624 if (offset < 0 || offset >= inode->i_size)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002625 offset = -ENXIO;
2626 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002627 start = offset >> PAGE_SHIFT;
2628 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002629 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002630 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002631 if (new_offset > offset) {
2632 if (new_offset < inode->i_size)
2633 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002634 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002635 offset = -ENXIO;
2636 else
2637 offset = inode->i_size;
2638 }
2639 }
2640
Hugh Dickins387aae62013-08-04 11:30:25 -07002641 if (offset >= 0)
2642 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002643 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002644 return offset;
2645}
2646
David Herrmann05f65b52014-08-08 14:25:36 -07002647/*
2648 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2649 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2650 */
2651#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2652#define LAST_SCAN 4 /* about 150ms max */
2653
2654static void shmem_tag_pins(struct address_space *mapping)
2655{
2656 struct radix_tree_iter iter;
2657 void **slot;
2658 pgoff_t start;
2659 struct page *page;
Matthew Wilcox (Oracle)391d4ee2019-10-25 09:58:35 -07002660 unsigned int tagged = 0;
David Herrmann05f65b52014-08-08 14:25:36 -07002661
2662 lru_add_drain();
2663 start = 0;
David Herrmann05f65b52014-08-08 14:25:36 -07002664
Matthew Wilcox (Oracle)391d4ee2019-10-25 09:58:35 -07002665 spin_lock_irq(&mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002666 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
zhong jiang988f7012019-11-18 11:26:07 +08002667 page = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002668 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002669 if (radix_tree_deref_retry(page)) {
2670 slot = radix_tree_iter_retry(&iter);
2671 continue;
2672 }
David Herrmann05f65b52014-08-08 14:25:36 -07002673 } else if (page_count(page) - page_mapcount(page) > 1) {
David Herrmann05f65b52014-08-08 14:25:36 -07002674 radix_tree_tag_set(&mapping->page_tree, iter.index,
2675 SHMEM_TAG_PINNED);
David Herrmann05f65b52014-08-08 14:25:36 -07002676 }
2677
Matthew Wilcox (Oracle)391d4ee2019-10-25 09:58:35 -07002678 if (++tagged % 1024)
2679 continue;
2680
2681 slot = radix_tree_iter_resume(slot, &iter);
2682 spin_unlock_irq(&mapping->tree_lock);
2683 cond_resched();
2684 spin_lock_irq(&mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002685 }
Matthew Wilcox (Oracle)391d4ee2019-10-25 09:58:35 -07002686 spin_unlock_irq(&mapping->tree_lock);
David Herrmann05f65b52014-08-08 14:25:36 -07002687}
2688
2689/*
2690 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2691 * via get_user_pages(), drivers might have some pending I/O without any active
2692 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2693 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2694 * them to be dropped.
2695 * The caller must guarantee that no new user will acquire writable references
2696 * to those pages to avoid races.
2697 */
David Herrmann40e041a2014-08-08 14:25:27 -07002698static int shmem_wait_for_pins(struct address_space *mapping)
2699{
David Herrmann05f65b52014-08-08 14:25:36 -07002700 struct radix_tree_iter iter;
2701 void **slot;
2702 pgoff_t start;
2703 struct page *page;
2704 int error, scan;
2705
2706 shmem_tag_pins(mapping);
2707
2708 error = 0;
2709 for (scan = 0; scan <= LAST_SCAN; scan++) {
2710 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2711 break;
2712
2713 if (!scan)
2714 lru_add_drain_all();
2715 else if (schedule_timeout_killable((HZ << scan) / 200))
2716 scan = LAST_SCAN;
2717
2718 start = 0;
2719 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002720 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2721 start, SHMEM_TAG_PINNED) {
2722
2723 page = radix_tree_deref_slot(slot);
2724 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002725 if (radix_tree_deref_retry(page)) {
2726 slot = radix_tree_iter_retry(&iter);
2727 continue;
2728 }
David Herrmann05f65b52014-08-08 14:25:36 -07002729
2730 page = NULL;
2731 }
2732
2733 if (page &&
2734 page_count(page) - page_mapcount(page) != 1) {
2735 if (scan < LAST_SCAN)
2736 goto continue_resched;
2737
2738 /*
2739 * On the last scan, we clean up all those tags
2740 * we inserted; but make a note that we still
2741 * found pages pinned.
2742 */
2743 error = -EBUSY;
2744 }
2745
2746 spin_lock_irq(&mapping->tree_lock);
2747 radix_tree_tag_clear(&mapping->page_tree,
2748 iter.index, SHMEM_TAG_PINNED);
2749 spin_unlock_irq(&mapping->tree_lock);
2750continue_resched:
2751 if (need_resched()) {
Matthew Wilcox148deab2016-12-14 15:08:49 -08002752 slot = radix_tree_iter_resume(slot, &iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002753 cond_resched_rcu();
David Herrmann05f65b52014-08-08 14:25:36 -07002754 }
2755 }
2756 rcu_read_unlock();
2757 }
2758
2759 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002760}
2761
2762#define F_ALL_SEALS (F_SEAL_SEAL | \
2763 F_SEAL_SHRINK | \
2764 F_SEAL_GROW | \
2765 F_SEAL_WRITE)
2766
2767int shmem_add_seals(struct file *file, unsigned int seals)
2768{
2769 struct inode *inode = file_inode(file);
2770 struct shmem_inode_info *info = SHMEM_I(inode);
2771 int error;
2772
2773 /*
2774 * SEALING
2775 * Sealing allows multiple parties to share a shmem-file but restrict
2776 * access to a specific subset of file operations. Seals can only be
2777 * added, but never removed. This way, mutually untrusted parties can
2778 * share common memory regions with a well-defined policy. A malicious
2779 * peer can thus never perform unwanted operations on a shared object.
2780 *
2781 * Seals are only supported on special shmem-files and always affect
2782 * the whole underlying inode. Once a seal is set, it may prevent some
2783 * kinds of access to the file. Currently, the following seals are
2784 * defined:
2785 * SEAL_SEAL: Prevent further seals from being set on this file
2786 * SEAL_SHRINK: Prevent the file from shrinking
2787 * SEAL_GROW: Prevent the file from growing
2788 * SEAL_WRITE: Prevent write access to the file
2789 *
2790 * As we don't require any trust relationship between two parties, we
2791 * must prevent seals from being removed. Therefore, sealing a file
2792 * only adds a given set of seals to the file, it never touches
2793 * existing seals. Furthermore, the "setting seals"-operation can be
2794 * sealed itself, which basically prevents any further seal from being
2795 * added.
2796 *
2797 * Semantics of sealing are only defined on volatile files. Only
2798 * anonymous shmem files support sealing. More importantly, seals are
2799 * never written to disk. Therefore, there's no plan to support it on
2800 * other file types.
2801 */
2802
2803 if (file->f_op != &shmem_file_operations)
2804 return -EINVAL;
2805 if (!(file->f_mode & FMODE_WRITE))
2806 return -EPERM;
2807 if (seals & ~(unsigned int)F_ALL_SEALS)
2808 return -EINVAL;
2809
Al Viro59551022016-01-22 15:40:57 -05002810 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002811
2812 if (info->seals & F_SEAL_SEAL) {
2813 error = -EPERM;
2814 goto unlock;
2815 }
2816
2817 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2818 error = mapping_deny_writable(file->f_mapping);
2819 if (error)
2820 goto unlock;
2821
2822 error = shmem_wait_for_pins(file->f_mapping);
2823 if (error) {
2824 mapping_allow_writable(file->f_mapping);
2825 goto unlock;
2826 }
2827 }
2828
2829 info->seals |= seals;
2830 error = 0;
2831
2832unlock:
Al Viro59551022016-01-22 15:40:57 -05002833 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002834 return error;
2835}
2836EXPORT_SYMBOL_GPL(shmem_add_seals);
2837
2838int shmem_get_seals(struct file *file)
2839{
2840 if (file->f_op != &shmem_file_operations)
2841 return -EINVAL;
2842
2843 return SHMEM_I(file_inode(file))->seals;
2844}
2845EXPORT_SYMBOL_GPL(shmem_get_seals);
2846
2847long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2848{
2849 long error;
2850
2851 switch (cmd) {
2852 case F_ADD_SEALS:
2853 /* disallow upper 32bit */
2854 if (arg > UINT_MAX)
2855 return -EINVAL;
2856
2857 error = shmem_add_seals(file, arg);
2858 break;
2859 case F_GET_SEALS:
2860 error = shmem_get_seals(file);
2861 break;
2862 default:
2863 error = -EINVAL;
2864 break;
2865 }
2866
2867 return error;
2868}
2869
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002870static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2871 loff_t len)
2872{
Al Viro496ad9a2013-01-23 17:07:38 -05002873 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002874 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002875 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002876 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002877 pgoff_t start, index, end;
2878 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002879
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002880 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2881 return -EOPNOTSUPP;
2882
Al Viro59551022016-01-22 15:40:57 -05002883 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002884
2885 if (mode & FALLOC_FL_PUNCH_HOLE) {
2886 struct address_space *mapping = file->f_mapping;
2887 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2888 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002889 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002890
David Herrmann40e041a2014-08-08 14:25:27 -07002891 /* protected by i_mutex */
2892 if (info->seals & F_SEAL_WRITE) {
2893 error = -EPERM;
2894 goto out;
2895 }
2896
Hugh Dickins8e205f72014-07-23 14:00:10 -07002897 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junf49aa242019-11-30 17:58:11 -08002898 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002899 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2900 spin_lock(&inode->i_lock);
2901 inode->i_private = &shmem_falloc;
2902 spin_unlock(&inode->i_lock);
2903
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002904 if ((u64)unmap_end > (u64)unmap_start)
2905 unmap_mapping_range(mapping, unmap_start,
2906 1 + unmap_end - unmap_start, 0);
2907 shmem_truncate_range(inode, offset, offset + len - 1);
2908 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002909
2910 spin_lock(&inode->i_lock);
2911 inode->i_private = NULL;
2912 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02002913 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002914 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002915 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002916 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002917 }
2918
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002919 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2920 error = inode_newsize_ok(inode, offset + len);
2921 if (error)
2922 goto out;
2923
David Herrmann40e041a2014-08-08 14:25:27 -07002924 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2925 error = -EPERM;
2926 goto out;
2927 }
2928
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002929 start = offset >> PAGE_SHIFT;
2930 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002931 /* Try to avoid a swapstorm if len is impossible to satisfy */
2932 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2933 error = -ENOSPC;
2934 goto out;
2935 }
2936
Hugh Dickins8e205f72014-07-23 14:00:10 -07002937 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002938 shmem_falloc.start = start;
2939 shmem_falloc.next = start;
2940 shmem_falloc.nr_falloced = 0;
2941 shmem_falloc.nr_unswapped = 0;
2942 spin_lock(&inode->i_lock);
2943 inode->i_private = &shmem_falloc;
2944 spin_unlock(&inode->i_lock);
2945
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002946 for (index = start; index < end; index++) {
2947 struct page *page;
2948
2949 /*
2950 * Good, the fallocate(2) manpage permits EINTR: we may have
2951 * been interrupted because we are using up too much memory.
2952 */
2953 if (signal_pending(current))
2954 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002955 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2956 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002957 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002958 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002959 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002960 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002961 if (index > start) {
2962 shmem_undo_range(inode,
2963 (loff_t)start << PAGE_SHIFT,
2964 ((loff_t)index << PAGE_SHIFT) - 1, true);
2965 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002966 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002967 }
2968
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002969 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002970 * Inform shmem_writepage() how far we have reached.
2971 * No need for lock or barrier: we have the page lock.
2972 */
2973 shmem_falloc.next++;
2974 if (!PageUptodate(page))
2975 shmem_falloc.nr_falloced++;
2976
2977 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002978 * If !PageUptodate, leave it that way so that freeable pages
2979 * can be recognized if we need to rollback on error later.
2980 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002981 * than free the pages we are allocating (and SGP_CACHE pages
2982 * might still be clean: we now need to mark those dirty too).
2983 */
2984 set_page_dirty(page);
2985 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002986 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002987 cond_resched();
2988 }
2989
2990 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2991 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002992 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002993undone:
2994 spin_lock(&inode->i_lock);
2995 inode->i_private = NULL;
2996 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002997out:
Al Viro59551022016-01-22 15:40:57 -05002998 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002999 return error;
3000}
3001
David Howells726c3342006-06-23 02:02:58 -07003002static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003{
David Howells726c3342006-06-23 02:02:58 -07003004 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005
3006 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003007 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003009 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003011 buf->f_bavail =
3012 buf->f_bfree = sbinfo->max_blocks -
3013 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003014 }
3015 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 buf->f_files = sbinfo->max_inodes;
3017 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 }
3019 /* else leave those fields 0 like simple_statfs */
3020 return 0;
3021}
3022
3023/*
3024 * File creation. Allocate an inode, and we're done..
3025 */
3026static int
Al Viro1a67aaf2011-07-26 01:52:52 -04003027shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003029 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 int error = -ENOSPC;
3031
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003032 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003034 error = simple_acl_create(dir, inode);
3035 if (error)
3036 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05003037 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003038 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003039 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003040 if (error && error != -EOPNOTSUPP)
3041 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04003042
Al Viro718deb62009-12-16 19:35:36 -05003043 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003045 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 d_instantiate(dentry, inode);
3047 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 }
3049 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003050out_iput:
3051 iput(inode);
3052 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053}
3054
Al Viro60545d02013-06-07 01:20:27 -04003055static int
3056shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
3057{
3058 struct inode *inode;
3059 int error = -ENOSPC;
3060
3061 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
3062 if (inode) {
3063 error = security_inode_init_security(inode, dir,
3064 NULL,
3065 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003066 if (error && error != -EOPNOTSUPP)
3067 goto out_iput;
3068 error = simple_acl_create(dir, inode);
3069 if (error)
3070 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04003071 d_tmpfile(dentry, inode);
3072 }
3073 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003074out_iput:
3075 iput(inode);
3076 return error;
Al Viro60545d02013-06-07 01:20:27 -04003077}
3078
Al Viro18bb1db2011-07-26 01:41:39 -04003079static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080{
3081 int error;
3082
3083 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3084 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07003085 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 return 0;
3087}
3088
Al Viro4acdaf22011-07-26 01:42:34 -04003089static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04003090 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091{
3092 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3093}
3094
3095/*
3096 * Link a file..
3097 */
3098static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3099{
David Howells75c3cfa2015-03-17 22:26:12 +00003100 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong4cf6ad02019-02-22 22:35:32 -08003101 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102
3103 /*
3104 * No ordinary (disk based) filesystem counts links as inodes;
3105 * but each new link needs a new dentry, pinning lowmem, and
3106 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wong00df9622019-02-21 08:48:09 -08003107 * But if an O_TMPFILE file is linked into the tmpfs, the
3108 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 */
Darrick J. Wong00df9622019-02-21 08:48:09 -08003110 if (inode->i_nlink) {
3111 ret = shmem_reserve_inode(inode->i_sb);
3112 if (ret)
3113 goto out;
3114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115
3116 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003117 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07003118 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04003119 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 dget(dentry); /* Extra pinning count for the created dentry */
3121 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003122out:
3123 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124}
3125
3126static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3127{
David Howells75c3cfa2015-03-17 22:26:12 +00003128 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003130 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3131 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132
3133 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003134 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003135 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 dput(dentry); /* Undo the count from "create" - this does all the work */
3137 return 0;
3138}
3139
3140static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3141{
3142 if (!simple_empty(dentry))
3143 return -ENOTEMPTY;
3144
David Howells75c3cfa2015-03-17 22:26:12 +00003145 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003146 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 return shmem_unlink(dir, dentry);
3148}
3149
Miklos Szeredi37456772014-07-23 15:15:34 +02003150static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3151{
David Howellse36cb0b2015-01-29 12:02:35 +00003152 bool old_is_dir = d_is_dir(old_dentry);
3153 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003154
3155 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3156 if (old_is_dir) {
3157 drop_nlink(old_dir);
3158 inc_nlink(new_dir);
3159 } else {
3160 drop_nlink(new_dir);
3161 inc_nlink(old_dir);
3162 }
3163 }
3164 old_dir->i_ctime = old_dir->i_mtime =
3165 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003166 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003167 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02003168
3169 return 0;
3170}
3171
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003172static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3173{
3174 struct dentry *whiteout;
3175 int error;
3176
3177 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3178 if (!whiteout)
3179 return -ENOMEM;
3180
3181 error = shmem_mknod(old_dir, whiteout,
3182 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3183 dput(whiteout);
3184 if (error)
3185 return error;
3186
3187 /*
3188 * Cheat and hash the whiteout while the old dentry is still in
3189 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3190 *
3191 * d_lookup() will consistently find one of them at this point,
3192 * not sure which one, but that isn't even important.
3193 */
3194 d_rehash(whiteout);
3195 return 0;
3196}
3197
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198/*
3199 * The VFS layer already does all the dentry stuff for rename,
3200 * we just have to decrement the usage count for the target if
3201 * it exists so that the VFS layer correctly free's it when it
3202 * gets overwritten.
3203 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003204static 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 -07003205{
David Howells75c3cfa2015-03-17 22:26:12 +00003206 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 int they_are_dirs = S_ISDIR(inode->i_mode);
3208
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003209 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003210 return -EINVAL;
3211
Miklos Szeredi37456772014-07-23 15:15:34 +02003212 if (flags & RENAME_EXCHANGE)
3213 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3214
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 if (!simple_empty(new_dentry))
3216 return -ENOTEMPTY;
3217
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003218 if (flags & RENAME_WHITEOUT) {
3219 int error;
3220
3221 error = shmem_whiteout(old_dir, old_dentry);
3222 if (error)
3223 return error;
3224 }
3225
David Howells75c3cfa2015-03-17 22:26:12 +00003226 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003228 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003229 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003230 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003233 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003234 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235 }
3236
3237 old_dir->i_size -= BOGO_DIRENT_SIZE;
3238 new_dir->i_size += BOGO_DIRENT_SIZE;
3239 old_dir->i_ctime = old_dir->i_mtime =
3240 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003241 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 return 0;
3243}
3244
3245static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3246{
3247 int error;
3248 int len;
3249 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003250 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 struct shmem_inode_info *info;
3252
3253 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003254 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 return -ENAMETOOLONG;
3256
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003257 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 if (!inode)
3259 return -ENOSPC;
3260
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003261 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003262 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003263 if (error) {
3264 if (error != -EOPNOTSUPP) {
3265 iput(inode);
3266 return error;
3267 }
3268 error = 0;
3269 }
3270
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 info = SHMEM_I(inode);
3272 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003273 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003274 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3275 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003276 iput(inode);
3277 return -ENOMEM;
3278 }
3279 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003281 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003282 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 if (error) {
3284 iput(inode);
3285 return error;
3286 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003287 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003289 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003290 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003292 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003293 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003296 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 d_instantiate(dentry, inode);
3298 dget(dentry);
3299 return 0;
3300}
3301
Al Virofceef392015-12-29 15:58:39 -05003302static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303{
Al Virofceef392015-12-29 15:58:39 -05003304 mark_page_accessed(arg);
3305 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306}
3307
Al Viro6b255392015-11-17 10:20:54 -05003308static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003309 struct inode *inode,
3310 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003313 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003314 if (!dentry) {
3315 page = find_get_page(inode->i_mapping, 0);
3316 if (!page)
3317 return ERR_PTR(-ECHILD);
3318 if (!PageUptodate(page)) {
3319 put_page(page);
3320 return ERR_PTR(-ECHILD);
3321 }
3322 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003323 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003324 if (error)
3325 return ERR_PTR(error);
3326 unlock_page(page);
3327 }
Al Virofceef392015-12-29 15:58:39 -05003328 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003329 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330}
3331
Eric Parisb09e0fa2011-05-24 17:12:39 -07003332#ifdef CONFIG_TMPFS_XATTR
3333/*
3334 * Superblocks without xattr inode operations may get some security.* xattr
3335 * support from the LSM "for free". As soon as we have any other xattrs
3336 * like ACLs, we also need to implement the security.* handlers at
3337 * filesystem level, though.
3338 */
3339
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003340/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003341 * Callback for security_inode_init_security() for acquiring xattrs.
3342 */
3343static int shmem_initxattrs(struct inode *inode,
3344 const struct xattr *xattr_array,
3345 void *fs_info)
3346{
3347 struct shmem_inode_info *info = SHMEM_I(inode);
3348 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003349 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003350 size_t len;
3351
3352 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003353 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003354 if (!new_xattr)
3355 return -ENOMEM;
3356
3357 len = strlen(xattr->name) + 1;
3358 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3359 GFP_KERNEL);
3360 if (!new_xattr->name) {
3361 kfree(new_xattr);
3362 return -ENOMEM;
3363 }
3364
3365 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3366 XATTR_SECURITY_PREFIX_LEN);
3367 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3368 xattr->name, len);
3369
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003370 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003371 }
3372
3373 return 0;
3374}
3375
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003376static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003377 struct dentry *unused, struct inode *inode,
3378 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003379{
Al Virob2968212016-04-10 20:48:24 -04003380 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003381
3382 name = xattr_full_name(handler, name);
3383 return simple_xattr_get(&info->xattrs, name, buffer, size);
3384}
3385
3386static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003387 struct dentry *unused, struct inode *inode,
3388 const char *name, const void *value,
3389 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003390{
Al Viro59301222016-05-27 10:19:30 -04003391 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003392
3393 name = xattr_full_name(handler, name);
3394 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3395}
3396
3397static const struct xattr_handler shmem_security_xattr_handler = {
3398 .prefix = XATTR_SECURITY_PREFIX,
3399 .get = shmem_xattr_handler_get,
3400 .set = shmem_xattr_handler_set,
3401};
3402
3403static const struct xattr_handler shmem_trusted_xattr_handler = {
3404 .prefix = XATTR_TRUSTED_PREFIX,
3405 .get = shmem_xattr_handler_get,
3406 .set = shmem_xattr_handler_set,
3407};
3408
Eric Parisb09e0fa2011-05-24 17:12:39 -07003409static const struct xattr_handler *shmem_xattr_handlers[] = {
3410#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003411 &posix_acl_access_xattr_handler,
3412 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003413#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003414 &shmem_security_xattr_handler,
3415 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003416 NULL
3417};
3418
Eric Parisb09e0fa2011-05-24 17:12:39 -07003419static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3420{
David Howells75c3cfa2015-03-17 22:26:12 +00003421 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003422 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003423}
3424#endif /* CONFIG_TMPFS_XATTR */
3425
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003426static const struct inode_operations shmem_short_symlink_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003427 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003428#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003429 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003430#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431};
3432
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003433static const struct inode_operations shmem_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003434 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003435#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003436 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003437#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003438};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003439
David M. Grimes91828a42006-10-17 00:09:45 -07003440static struct dentry *shmem_get_parent(struct dentry *child)
3441{
3442 return ERR_PTR(-ESTALE);
3443}
3444
3445static int shmem_match(struct inode *ino, void *vfh)
3446{
3447 __u32 *fh = vfh;
3448 __u64 inum = fh[2];
3449 inum = (inum << 32) | fh[1];
3450 return ino->i_ino == inum && fh[0] == ino->i_generation;
3451}
3452
Christoph Hellwig480b1162007-10-21 16:42:13 -07003453static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3454 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003455{
David M. Grimes91828a42006-10-17 00:09:45 -07003456 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003457 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003458 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003459
Christoph Hellwig480b1162007-10-21 16:42:13 -07003460 if (fh_len < 3)
3461 return NULL;
3462
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003463 inum = fid->raw[2];
3464 inum = (inum << 32) | fid->raw[1];
3465
Christoph Hellwig480b1162007-10-21 16:42:13 -07003466 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3467 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003468 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003469 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003470 iput(inode);
3471 }
3472
Christoph Hellwig480b1162007-10-21 16:42:13 -07003473 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003474}
3475
Al Virob0b03822012-04-02 14:34:06 -04003476static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3477 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003478{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303479 if (*len < 3) {
3480 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003481 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303482 }
David M. Grimes91828a42006-10-17 00:09:45 -07003483
Al Viro1d3382cb2010-10-23 15:19:20 -04003484 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003485 /* Unfortunately insert_inode_hash is not idempotent,
3486 * so as we hash inodes here rather than at creation
3487 * time, we need a lock to ensure we only try
3488 * to do it once
3489 */
3490 static DEFINE_SPINLOCK(lock);
3491 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003492 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003493 __insert_inode_hash(inode,
3494 inode->i_ino + inode->i_generation);
3495 spin_unlock(&lock);
3496 }
3497
3498 fh[0] = inode->i_generation;
3499 fh[1] = inode->i_ino;
3500 fh[2] = ((__u64)inode->i_ino) >> 32;
3501
3502 *len = 3;
3503 return 1;
3504}
3505
Christoph Hellwig39655162007-10-21 16:42:17 -07003506static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003507 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003508 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003509 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003510};
3511
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003512static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3513 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514{
3515 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003516 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003517 uid_t uid;
3518 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003520 while (options != NULL) {
3521 this_char = options;
3522 for (;;) {
3523 /*
3524 * NUL-terminate this option: unfortunately,
3525 * mount options form a comma-separated list,
3526 * but mpol's nodelist may also contain commas.
3527 */
3528 options = strchr(options, ',');
3529 if (options == NULL)
3530 break;
3531 options++;
3532 if (!isdigit(*options)) {
3533 options[-1] = '\0';
3534 break;
3535 }
3536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 if (!*this_char)
3538 continue;
3539 if ((value = strchr(this_char,'=')) != NULL) {
3540 *value++ = 0;
3541 } else {
Joe Perches11705322016-03-17 14:19:50 -07003542 pr_err("tmpfs: No value for mount option '%s'\n",
3543 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003544 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 }
3546
3547 if (!strcmp(this_char,"size")) {
3548 unsigned long long size;
3549 size = memparse(value,&rest);
3550 if (*rest == '%') {
3551 size <<= PAGE_SHIFT;
3552 size *= totalram_pages;
3553 do_div(size, 100);
3554 rest++;
3555 }
3556 if (*rest)
3557 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003558 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003559 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003561 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 if (*rest)
3563 goto bad_val;
3564 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003565 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566 if (*rest)
3567 goto bad_val;
3568 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003569 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003571 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572 if (*rest)
3573 goto bad_val;
3574 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003575 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003577 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 if (*rest)
3579 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003580 sbinfo->uid = make_kuid(current_user_ns(), uid);
3581 if (!uid_valid(sbinfo->uid))
3582 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003584 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003586 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 if (*rest)
3588 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003589 sbinfo->gid = make_kgid(current_user_ns(), gid);
3590 if (!gid_valid(sbinfo->gid))
3591 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003592#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003593 } else if (!strcmp(this_char, "huge")) {
3594 int huge;
3595 huge = shmem_parse_huge(value);
3596 if (huge < 0)
3597 goto bad_val;
3598 if (!has_transparent_hugepage() &&
3599 huge != SHMEM_HUGE_NEVER)
3600 goto bad_val;
3601 sbinfo->huge = huge;
3602#endif
3603#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003604 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003605 mpol_put(mpol);
3606 mpol = NULL;
3607 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003608 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003609#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 } else {
Joe Perches11705322016-03-17 14:19:50 -07003611 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003612 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613 }
3614 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003615 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 return 0;
3617
3618bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003619 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003621error:
3622 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 return 1;
3624
3625}
3626
3627static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3628{
3629 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003630 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003631 unsigned long inodes;
3632 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633
Greg Thelen5f001102013-02-22 16:36:01 -08003634 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003635 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003636 return error;
3637
3638 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003639 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003640 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003641 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003642 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003643 goto out;
3644 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003645 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003646 * but we must separately disallow unlimited->limited, because
3647 * in that case we have no record of how much is already in use.
3648 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003649 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003650 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003651 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003652 goto out;
3653
3654 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003655 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003656 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003657 sbinfo->max_inodes = config.max_inodes;
3658 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003659
Greg Thelen5f001102013-02-22 16:36:01 -08003660 /*
3661 * Preserve previous mempolicy unless mpol remount option was specified.
3662 */
3663 if (config.mpol) {
3664 mpol_put(sbinfo->mpol);
3665 sbinfo->mpol = config.mpol; /* transfers initial ref */
3666 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003667out:
3668 spin_unlock(&sbinfo->stat_lock);
3669 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003671
Al Viro34c80b12011-12-08 21:32:45 -05003672static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003673{
Al Viro34c80b12011-12-08 21:32:45 -05003674 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003675
3676 if (sbinfo->max_blocks != shmem_default_max_blocks())
3677 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003678 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003679 if (sbinfo->max_inodes != shmem_default_max_inodes())
3680 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3681 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003682 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003683 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3684 seq_printf(seq, ",uid=%u",
3685 from_kuid_munged(&init_user_ns, sbinfo->uid));
3686 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3687 seq_printf(seq, ",gid=%u",
3688 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003689#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003690 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3691 if (sbinfo->huge)
3692 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3693#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003694 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003695 return 0;
3696}
David Herrmann9183df22014-08-08 14:25:29 -07003697
3698#define MFD_NAME_PREFIX "memfd:"
3699#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3700#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3701
Mike Kravetz749df872017-09-06 16:24:16 -07003702#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
David Herrmann9183df22014-08-08 14:25:29 -07003703
3704SYSCALL_DEFINE2(memfd_create,
3705 const char __user *, uname,
3706 unsigned int, flags)
3707{
3708 struct shmem_inode_info *info;
3709 struct file *file;
3710 int fd, error;
3711 char *name;
3712 long len;
3713
Mike Kravetz749df872017-09-06 16:24:16 -07003714 if (!(flags & MFD_HUGETLB)) {
3715 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3716 return -EINVAL;
3717 } else {
3718 /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */
3719 if (flags & MFD_ALLOW_SEALING)
3720 return -EINVAL;
3721 /* Allow huge page size encoding in flags. */
3722 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
3723 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
3724 return -EINVAL;
3725 }
David Herrmann9183df22014-08-08 14:25:29 -07003726
3727 /* length includes terminating zero */
3728 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3729 if (len <= 0)
3730 return -EFAULT;
3731 if (len > MFD_NAME_MAX_LEN + 1)
3732 return -EINVAL;
3733
Michal Hocko0ee931c2017-09-13 16:28:29 -07003734 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL);
David Herrmann9183df22014-08-08 14:25:29 -07003735 if (!name)
3736 return -ENOMEM;
3737
3738 strcpy(name, MFD_NAME_PREFIX);
3739 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3740 error = -EFAULT;
3741 goto err_name;
3742 }
3743
3744 /* terminating-zero may have changed after strnlen_user() returned */
3745 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3746 error = -EFAULT;
3747 goto err_name;
3748 }
3749
3750 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3751 if (fd < 0) {
3752 error = fd;
3753 goto err_name;
3754 }
3755
Mike Kravetz749df872017-09-06 16:24:16 -07003756 if (flags & MFD_HUGETLB) {
3757 struct user_struct *user = NULL;
3758
3759 file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
3760 HUGETLB_ANONHUGE_INODE,
3761 (flags >> MFD_HUGE_SHIFT) &
3762 MFD_HUGE_MASK);
3763 } else
3764 file = shmem_file_setup(name, 0, VM_NORESERVE);
David Herrmann9183df22014-08-08 14:25:29 -07003765 if (IS_ERR(file)) {
3766 error = PTR_ERR(file);
3767 goto err_fd;
3768 }
David Herrmann9183df22014-08-08 14:25:29 -07003769 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3770 file->f_flags |= O_RDWR | O_LARGEFILE;
Mike Kravetz749df872017-09-06 16:24:16 -07003771
3772 if (flags & MFD_ALLOW_SEALING) {
3773 /*
3774 * flags check at beginning of function ensures
3775 * this is not a hugetlbfs (MFD_HUGETLB) file.
3776 */
3777 info = SHMEM_I(file_inode(file));
David Herrmann9183df22014-08-08 14:25:29 -07003778 info->seals &= ~F_SEAL_SEAL;
Mike Kravetz749df872017-09-06 16:24:16 -07003779 }
David Herrmann9183df22014-08-08 14:25:29 -07003780
3781 fd_install(fd, file);
3782 kfree(name);
3783 return fd;
3784
3785err_fd:
3786 put_unused_fd(fd);
3787err_name:
3788 kfree(name);
3789 return error;
3790}
3791
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003792#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793
3794static void shmem_put_super(struct super_block *sb)
3795{
Hugh Dickins602586a2010-08-17 15:23:56 -07003796 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3797
3798 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003799 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003800 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 sb->s_fs_info = NULL;
3802}
3803
Kay Sievers2b2af542009-04-30 15:23:42 +02003804int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805{
3806 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003807 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003808 int err = -ENOMEM;
3809
3810 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003811 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003812 L1_CACHE_BYTES), GFP_KERNEL);
3813 if (!sbinfo)
3814 return -ENOMEM;
3815
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003816 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003817 sbinfo->uid = current_fsuid();
3818 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003819 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003821#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 /*
3823 * Per default we only allow half of the physical ram per
3824 * tmpfs instance, limiting inodes to one per page of lowmem;
3825 * but the internal instance is left unlimited.
3826 */
Al Viroca4e0512013-08-31 12:57:10 -04003827 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003828 sbinfo->max_blocks = shmem_default_max_blocks();
3829 sbinfo->max_inodes = shmem_default_max_inodes();
3830 if (shmem_parse_options(data, sbinfo, false)) {
3831 err = -EINVAL;
3832 goto failed;
3833 }
Al Viroca4e0512013-08-31 12:57:10 -04003834 } else {
3835 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836 }
David M. Grimes91828a42006-10-17 00:09:45 -07003837 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003838 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839#else
3840 sb->s_flags |= MS_NOUSER;
3841#endif
3842
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003843 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003844 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003845 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003846 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003847 spin_lock_init(&sbinfo->shrinklist_lock);
3848 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003849
Hugh Dickins285b2c42011-08-03 16:21:20 -07003850 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003851 sb->s_blocksize = PAGE_SIZE;
3852 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 sb->s_magic = TMPFS_MAGIC;
3854 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003855 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003856#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003857 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003858#endif
3859#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003860 sb->s_flags |= MS_POSIXACL;
3861#endif
Amir Goldstein2b4db792017-05-18 15:29:33 +03003862 uuid_gen(&sb->s_uuid);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003863
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003864 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 if (!inode)
3866 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003867 inode->i_uid = sbinfo->uid;
3868 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003869 sb->s_root = d_make_root(inode);
3870 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003871 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 return 0;
3873
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874failed:
3875 shmem_put_super(sb);
3876 return err;
3877}
3878
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003879static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880
3881static struct inode *shmem_alloc_inode(struct super_block *sb)
3882{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003883 struct shmem_inode_info *info;
3884 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3885 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003887 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888}
3889
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003890static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003891{
3892 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003893 if (S_ISLNK(inode->i_mode))
3894 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003895 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3896}
3897
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898static void shmem_destroy_inode(struct inode *inode)
3899{
Al Viro09208d12011-07-26 03:15:03 -04003900 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003902 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903}
3904
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003905static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003907 struct shmem_inode_info *info = foo;
3908 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909}
3910
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003911static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912{
3913 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3914 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003915 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916 return 0;
3917}
3918
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003919static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003921 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922}
3923
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003924static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003926 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003928 .write_begin = shmem_write_begin,
3929 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003931#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003932 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003933#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003934 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935};
3936
Helge Deller15ad7cd2006-12-06 20:40:36 -08003937static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003939 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003941 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003942 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003943 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003944 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003945 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003946 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003947 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948#endif
3949};
3950
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003951static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003952 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003953 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003954#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003955 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003956 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003957#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958};
3959
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003960static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961#ifdef CONFIG_TMPFS
3962 .create = shmem_create,
3963 .lookup = simple_lookup,
3964 .link = shmem_link,
3965 .unlink = shmem_unlink,
3966 .symlink = shmem_symlink,
3967 .mkdir = shmem_mkdir,
3968 .rmdir = shmem_rmdir,
3969 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003970 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003971 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003973#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003974 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003975#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003976#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003977 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003978 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003979#endif
3980};
3981
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003982static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003983#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003984 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003985#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003986#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003987 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003988 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003989#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990};
3991
Hugh Dickins759b9772007-03-05 00:30:28 -08003992static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 .alloc_inode = shmem_alloc_inode,
3994 .destroy_inode = shmem_destroy_inode,
3995#ifdef CONFIG_TMPFS
3996 .statfs = shmem_statfs,
3997 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003998 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999#endif
Al Viro1f895f72010-06-05 19:10:41 -04004000 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 .drop_inode = generic_delete_inode,
4002 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07004003#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4004 .nr_cached_objects = shmem_unused_huge_count,
4005 .free_cached_objects = shmem_unused_huge_scan,
4006#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007};
4008
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04004009static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07004010 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07004011 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012#ifdef CONFIG_NUMA
4013 .set_policy = shmem_set_policy,
4014 .get_policy = shmem_get_policy,
4015#endif
4016};
4017
Al Viro3c26ff62010-07-25 11:46:36 +04004018static struct dentry *shmem_mount(struct file_system_type *fs_type,
4019 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020{
Al Viro3c26ff62010-07-25 11:46:36 +04004021 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022}
4023
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004024static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 .owner = THIS_MODULE,
4026 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04004027 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004029 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004032int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033{
4034 int error;
4035
Rob Landley16203a72013-09-11 14:26:12 -07004036 /* If rootfs called this, don't re-init */
4037 if (shmem_inode_cachep)
4038 return 0;
4039
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004040 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041 if (error)
4042 goto out3;
4043
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004044 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07004046 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047 goto out2;
4048 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07004049
Al Viroca4e0512013-08-31 12:57:10 -04004050 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051 if (IS_ERR(shm_mnt)) {
4052 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07004053 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054 goto out1;
4055 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004056
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004057#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004058 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004059 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4060 else
4061 shmem_huge = 0; /* just in case it was patched */
4062#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 return 0;
4064
4065out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004066 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004068 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069out3:
4070 shm_mnt = ERR_PTR(error);
4071 return error;
4072}
Matt Mackall853ac432009-01-06 14:40:20 -08004073
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004074#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004075static ssize_t shmem_enabled_show(struct kobject *kobj,
4076 struct kobj_attribute *attr, char *buf)
4077{
4078 int values[] = {
4079 SHMEM_HUGE_ALWAYS,
4080 SHMEM_HUGE_WITHIN_SIZE,
4081 SHMEM_HUGE_ADVISE,
4082 SHMEM_HUGE_NEVER,
4083 SHMEM_HUGE_DENY,
4084 SHMEM_HUGE_FORCE,
4085 };
4086 int i, count;
4087
4088 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4089 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4090
4091 count += sprintf(buf + count, fmt,
4092 shmem_format_huge(values[i]));
4093 }
4094 buf[count - 1] = '\n';
4095 return count;
4096}
4097
4098static ssize_t shmem_enabled_store(struct kobject *kobj,
4099 struct kobj_attribute *attr, const char *buf, size_t count)
4100{
4101 char tmp[16];
4102 int huge;
4103
4104 if (count + 1 > sizeof(tmp))
4105 return -EINVAL;
4106 memcpy(tmp, buf, count);
4107 tmp[count] = '\0';
4108 if (count && tmp[count - 1] == '\n')
4109 tmp[count - 1] = '\0';
4110
4111 huge = shmem_parse_huge(tmp);
4112 if (huge == -EINVAL)
4113 return -EINVAL;
4114 if (!has_transparent_hugepage() &&
4115 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4116 return -EINVAL;
4117
4118 shmem_huge = huge;
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004119 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004120 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4121 return count;
4122}
4123
4124struct kobj_attribute shmem_enabled_attr =
4125 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07004126#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004127
Arnd Bergmann3b337192016-08-10 16:27:44 -07004128#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004129bool shmem_huge_enabled(struct vm_area_struct *vma)
4130{
4131 struct inode *inode = file_inode(vma->vm_file);
4132 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4133 loff_t i_size;
4134 pgoff_t off;
4135
4136 if (shmem_huge == SHMEM_HUGE_FORCE)
4137 return true;
4138 if (shmem_huge == SHMEM_HUGE_DENY)
4139 return false;
4140 switch (sbinfo->huge) {
4141 case SHMEM_HUGE_NEVER:
4142 return false;
4143 case SHMEM_HUGE_ALWAYS:
4144 return true;
4145 case SHMEM_HUGE_WITHIN_SIZE:
4146 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4147 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4148 if (i_size >= HPAGE_PMD_SIZE &&
4149 i_size >> PAGE_SHIFT >= off)
4150 return true;
4151 case SHMEM_HUGE_ADVISE:
4152 /* TODO: implement fadvise() hints */
4153 return (vma->vm_flags & VM_HUGEPAGE);
4154 default:
4155 VM_BUG_ON(1);
4156 return false;
4157 }
4158}
Arnd Bergmann3b337192016-08-10 16:27:44 -07004159#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004160
Matt Mackall853ac432009-01-06 14:40:20 -08004161#else /* !CONFIG_SHMEM */
4162
4163/*
4164 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4165 *
4166 * This is intended for small system where the benefits of the full
4167 * shmem code (swap-backed and resource-limited) are outweighed by
4168 * their complexity. On systems without swap this code should be
4169 * effectively equivalent, but much lighter weight.
4170 */
4171
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004172static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004173 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04004174 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08004175 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004176 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004177};
4178
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004179int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004180{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004181 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004182
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004183 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004184 BUG_ON(IS_ERR(shm_mnt));
4185
4186 return 0;
4187}
4188
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004189int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08004190{
4191 return 0;
4192}
4193
Hugh Dickins3f96b792009-09-21 17:03:37 -07004194int shmem_lock(struct file *file, int lock, struct user_struct *user)
4195{
4196 return 0;
4197}
4198
Hugh Dickins24513262012-01-20 14:34:21 -08004199void shmem_unlock_mapping(struct address_space *mapping)
4200{
4201}
4202
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004203#ifdef CONFIG_MMU
4204unsigned long shmem_get_unmapped_area(struct file *file,
4205 unsigned long addr, unsigned long len,
4206 unsigned long pgoff, unsigned long flags)
4207{
4208 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4209}
4210#endif
4211
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004212void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004213{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004214 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004215}
4216EXPORT_SYMBOL_GPL(shmem_truncate_range);
4217
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004218#define shmem_vm_ops generic_file_vm_ops
4219#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004220#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004221#define shmem_acct_size(flags, size) 0
4222#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004223
4224#endif /* CONFIG_SHMEM */
4225
4226/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227
Rasmus Villemoes19938e32016-10-07 17:01:01 -07004228static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04004229 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05004230};
4231
Eric Parisc7277092013-12-02 11:24:19 +00004232static struct file *__shmem_file_setup(const char *name, loff_t size,
4233 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234{
Al Viro6b4d0b22013-02-14 21:37:26 -05004235 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04004237 struct path path;
Al Viro34515382013-02-14 22:38:02 -05004238 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239 struct qstr this;
4240
4241 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05004242 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243
Hugh Dickins285b2c42011-08-03 16:21:20 -07004244 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245 return ERR_PTR(-EINVAL);
4246
4247 if (shmem_acct_size(flags, size))
4248 return ERR_PTR(-ENOMEM);
4249
Al Viro6b4d0b22013-02-14 21:37:26 -05004250 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251 this.name = name;
4252 this.len = strlen(name);
4253 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004254 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004255 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004256 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004257 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004258 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004259 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260
Al Viro6b4d0b22013-02-14 21:37:26 -05004261 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004262 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004264 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265
Eric Parisc7277092013-12-02 11:24:19 +00004266 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004267 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004269 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004270 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4271 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004272 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004273
Al Viro6b4d0b22013-02-14 21:37:26 -05004274 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004275 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004276 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004277 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004278
Al Viro6b4d0b22013-02-14 21:37:26 -05004279 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004280
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281put_memory:
4282 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004283put_path:
4284 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004285 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286}
Eric Parisc7277092013-12-02 11:24:19 +00004287
4288/**
4289 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4290 * kernel internal. There will be NO LSM permission checks against the
4291 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004292 * higher layer. The users are the big_key and shm implementations. LSM
4293 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004294 * @name: name for dentry (to be seen in /proc/<pid>/maps
4295 * @size: size to be set for the file
4296 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4297 */
4298struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4299{
4300 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4301}
4302
4303/**
4304 * shmem_file_setup - get an unlinked file living in tmpfs
4305 * @name: name for dentry (to be seen in /proc/<pid>/maps
4306 * @size: size to be set for the file
4307 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4308 */
4309struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4310{
4311 return __shmem_file_setup(name, size, flags, 0);
4312}
Keith Packard395e0dd2008-06-20 00:08:06 -07004313EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314
Randy Dunlap46711812008-03-19 17:00:41 -07004315/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004317 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4318 */
4319int shmem_zero_setup(struct vm_area_struct *vma)
4320{
4321 struct file *file;
4322 loff_t size = vma->vm_end - vma->vm_start;
4323
Hugh Dickins66fc1302015-06-14 09:48:09 -07004324 /*
4325 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4326 * between XFS directory reading and selinux: since this file is only
4327 * accessible to the user through its mapping, use S_PRIVATE flag to
4328 * bypass file security, in the same way as shmem_kernel_file_setup().
4329 */
4330 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331 if (IS_ERR(file))
4332 return PTR_ERR(file);
4333
4334 if (vma->vm_file)
4335 fput(vma->vm_file);
4336 vma->vm_file = file;
4337 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004338
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004339 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004340 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4341 (vma->vm_end & HPAGE_PMD_MASK)) {
4342 khugepaged_enter(vma, vma->vm_flags);
4343 }
4344
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345 return 0;
4346}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004347
4348/**
4349 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4350 * @mapping: the page's address_space
4351 * @index: the page index
4352 * @gfp: the page allocator flags to use if allocating
4353 *
4354 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4355 * with any new page allocations done using the specified allocation flags.
4356 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4357 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4358 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4359 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004360 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4361 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004362 */
4363struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4364 pgoff_t index, gfp_t gfp)
4365{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004366#ifdef CONFIG_SHMEM
4367 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004368 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004369 int error;
4370
4371 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004372 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
Mike Rapoportcfda0522017-02-22 15:43:37 -08004373 gfp, NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004374 if (error)
4375 page = ERR_PTR(error);
4376 else
4377 unlock_page(page);
4378 return page;
4379#else
4380 /*
4381 * The tiny !SHMEM case uses ramfs without swap
4382 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004383 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004384#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004385}
4386EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);