blob: aad3ba74b0e9d121bd774e795460d5f2f73ac119 [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>
Arnd Bergmann46c9a942018-08-17 15:45:09 -070032#include <linux/random.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010033#include <linux/sched/signal.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040034#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080035#include <linux/swap.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080036#include <linux/uio.h>
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -070037#include <linux/khugepaged.h>
Mike Kravetz749df872017-09-06 16:24:16 -070038#include <linux/hugetlb.h>
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -080039#include <linux/frontswap.h>
Al Viro626c3922019-09-08 20:28:06 -040040#include <linux/fs_parser.h>
Matt Mackall853ac432009-01-06 14:40:20 -080041
Andrea Arcangeli95cc09d2017-02-22 15:43:31 -080042#include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
43
Matt Mackall853ac432009-01-06 14:40:20 -080044static struct vfsmount *shm_mnt;
45
46#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
51 */
52
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070053#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070054#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010055#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080056#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/string.h>
59#include <linux/slab.h>
60#include <linux/backing-dev.h>
61#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070064#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070065#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070066#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070067#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/security.h>
69#include <linux/swapops.h>
70#include <linux/mempolicy.h>
71#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000072#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070073#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070074#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080075#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040076#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070077#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070078#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070079#include <uapi/linux/memfd.h>
Mike Rapoportcfda0522017-02-22 15:43:37 -080080#include <linux/userfaultfd_k.h>
Mike Rapoport4c27fe42017-02-22 15:43:25 -080081#include <linux/rmap.h>
Amir Goldstein2b4db792017-05-18 15:29:33 +030082#include <linux/uuid.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070083
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080084#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#include <asm/pgtable.h>
86
Mel Gormandd56b042015-11-06 16:28:43 -080087#include "internal.h"
88
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030089#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
90#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Linus Torvalds1da177e2005-04-16 15:20:36 -070092/* Pretend that each entry is of this size in directory's i_size */
93#define BOGO_DIRENT_SIZE 20
94
Hugh Dickins69f07ec2011-08-03 16:21:26 -070095/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
96#define SHORT_SYMLINK_LEN 128
97
Hugh Dickins1aac1402012-05-29 15:06:42 -070098/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070099 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
100 * inode->i_private (with i_mutex making sure that it has only one user at
101 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700102 */
103struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -0700104 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -0700105 pgoff_t start; /* start of range currently being fallocated */
106 pgoff_t next; /* the next page offset to be fallocated */
107 pgoff_t nr_falloced; /* how many new pages have been fallocated */
108 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109};
110
Al Viro0b5071d2019-09-08 18:49:18 -0400111struct shmem_options {
112 unsigned long long blocks;
113 unsigned long long inodes;
114 struct mempolicy *mpol;
115 kuid_t uid;
116 kgid_t gid;
117 umode_t mode;
118 int huge;
119 int seen;
120#define SHMEM_SEEN_BLOCKS 1
121#define SHMEM_SEEN_INODES 2
122#define SHMEM_SEEN_HUGE 4
123};
124
Andrew Mortonb76db732008-02-08 04:21:49 -0800125#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800126static unsigned long shmem_default_max_blocks(void)
127{
Arun KSca79b0c2018-12-28 00:34:29 -0800128 return totalram_pages() / 2;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800129}
130
131static unsigned long shmem_default_max_inodes(void)
132{
Arun KSca79b0c2018-12-28 00:34:29 -0800133 unsigned long nr_pages = totalram_pages();
134
135 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800136}
Andrew Mortonb76db732008-02-08 04:21:49 -0800137#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800138
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700139static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
140static int shmem_replace_page(struct page **pagep, gfp_t gfp,
141 struct shmem_inode_info *info, pgoff_t index);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -0800142static int shmem_swapin_page(struct inode *inode, pgoff_t index,
143 struct page **pagep, enum sgp_type sgp,
144 gfp_t gfp, struct vm_area_struct *vma,
145 vm_fault_t *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700146static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700147 struct page **pagep, enum sgp_type sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800148 gfp_t gfp, struct vm_area_struct *vma,
Souptick Joarder2b740302018-08-23 17:01:36 -0700149 struct vm_fault *vmf, vm_fault_t *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700150
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700151int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700152 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700153{
154 return shmem_getpage_gfp(inode, index, pagep, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800155 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700156}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
159{
160 return sb->s_fs_info;
161}
162
163/*
164 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
165 * for shared memory and for shared anonymous (/dev/zero) mappings
166 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
167 * consistent with the pre-accounting of private mappings ...
168 */
169static inline int shmem_acct_size(unsigned long flags, loff_t size)
170{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000171 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000172 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
175static inline void shmem_unacct_size(unsigned long flags, loff_t size)
176{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000177 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 vm_unacct_memory(VM_ACCT(size));
179}
180
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700181static inline int shmem_reacct_size(unsigned long flags,
182 loff_t oldsize, loff_t newsize)
183{
184 if (!(flags & VM_NORESERVE)) {
185 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
186 return security_vm_enough_memory_mm(current->mm,
187 VM_ACCT(newsize) - VM_ACCT(oldsize));
188 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
189 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
190 }
191 return 0;
192}
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/*
195 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700196 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
198 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
199 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700200static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700202 if (!(flags & VM_NORESERVE))
203 return 0;
204
205 return security_vm_enough_memory_mm(current->mm,
206 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static inline void shmem_unacct_blocks(unsigned long flags, long pages)
210{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000211 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300212 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Mike Rapoport0f079692017-09-06 16:22:59 -0700215static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
216{
217 struct shmem_inode_info *info = SHMEM_I(inode);
218 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
219
220 if (shmem_acct_block(info->flags, pages))
221 return false;
222
223 if (sbinfo->max_blocks) {
224 if (percpu_counter_compare(&sbinfo->used_blocks,
225 sbinfo->max_blocks - pages) > 0)
226 goto unacct;
227 percpu_counter_add(&sbinfo->used_blocks, pages);
228 }
229
230 return true;
231
232unacct:
233 shmem_unacct_blocks(info->flags, pages);
234 return false;
235}
236
237static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
238{
239 struct shmem_inode_info *info = SHMEM_I(inode);
240 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
241
242 if (sbinfo->max_blocks)
243 percpu_counter_sub(&sbinfo->used_blocks, pages);
244 shmem_unacct_blocks(info->flags, pages);
245}
246
Hugh Dickins759b9772007-03-05 00:30:28 -0800247static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700248static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800249static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800250static const struct inode_operations shmem_inode_operations;
251static const struct inode_operations shmem_dir_inode_operations;
252static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +0400253static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700254static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Mike Rapoportb0506e42017-02-22 15:43:28 -0800256bool vma_is_shmem(struct vm_area_struct *vma)
257{
258 return vma->vm_ops == &shmem_vm_ops;
259}
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800262static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800264static int shmem_reserve_inode(struct super_block *sb)
265{
266 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
267 if (sbinfo->max_inodes) {
268 spin_lock(&sbinfo->stat_lock);
269 if (!sbinfo->free_inodes) {
270 spin_unlock(&sbinfo->stat_lock);
271 return -ENOSPC;
272 }
273 sbinfo->free_inodes--;
274 spin_unlock(&sbinfo->stat_lock);
275 }
276 return 0;
277}
278
279static void shmem_free_inode(struct super_block *sb)
280{
281 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
282 if (sbinfo->max_inodes) {
283 spin_lock(&sbinfo->stat_lock);
284 sbinfo->free_inodes++;
285 spin_unlock(&sbinfo->stat_lock);
286 }
287}
288
Randy Dunlap46711812008-03-19 17:00:41 -0700289/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700290 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * @inode: inode to recalc
292 *
293 * We have to calculate the free blocks since the mm can drop
294 * undirtied hole pages behind our back.
295 *
296 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
297 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
298 *
299 * It has to be called with the spinlock held.
300 */
301static void shmem_recalc_inode(struct inode *inode)
302{
303 struct shmem_inode_info *info = SHMEM_I(inode);
304 long freed;
305
306 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
307 if (freed > 0) {
308 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700309 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Mike Rapoport0f079692017-09-06 16:22:59 -0700310 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312}
313
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700314bool shmem_charge(struct inode *inode, long pages)
315{
316 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700317 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700318
Mike Rapoport0f079692017-09-06 16:22:59 -0700319 if (!shmem_inode_acct_block(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700320 return false;
Mike Rapoportb1cc94a2017-09-06 16:22:56 -0700321
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800322 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
323 inode->i_mapping->nrpages += pages;
324
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700325 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700326 info->alloced += pages;
327 inode->i_blocks += pages * BLOCKS_PER_PAGE;
328 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700329 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700330
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700331 return true;
332}
333
334void shmem_uncharge(struct inode *inode, long pages)
335{
336 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700337 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700338
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800339 /* nrpages adjustment done by __delete_from_page_cache() or caller */
340
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700341 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700342 info->alloced -= pages;
343 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
344 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700345 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700346
Mike Rapoport0f079692017-09-06 16:22:59 -0700347 shmem_inode_unacct_blocks(inode, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700348}
349
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700350/*
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500351 * Replace item expected in xarray by a new item, while holding xa_lock.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700352 */
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500353static int shmem_replace_entry(struct address_space *mapping,
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700354 pgoff_t index, void *expected, void *replacement)
355{
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500356 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700357 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700358
359 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700360 VM_BUG_ON(!replacement);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500361 item = xas_load(&xas);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700362 if (item != expected)
363 return -ENOENT;
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500364 xas_store(&xas, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700365 return 0;
366}
367
368/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700369 * Sometimes, before we decide whether to proceed or to fail, we must check
370 * that an entry was not already brought back from swap by a racing thread.
371 *
372 * Checking page is not enough: by the time a SwapCache page is locked, it
373 * might be reused, and again be SwapCache, using the same swap as before.
374 */
375static bool shmem_confirm_swap(struct address_space *mapping,
376 pgoff_t index, swp_entry_t swap)
377{
Matthew Wilcoxa12831b2017-11-22 08:34:58 -0500378 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
Hugh Dickinsd1899222012-07-11 14:02:47 -0700379}
380
381/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700382 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
383 *
384 * SHMEM_HUGE_NEVER:
385 * disables huge pages for the mount;
386 * SHMEM_HUGE_ALWAYS:
387 * enables huge pages for the mount;
388 * SHMEM_HUGE_WITHIN_SIZE:
389 * only allocate huge pages if the page will be fully within i_size,
390 * also respect fadvise()/madvise() hints;
391 * SHMEM_HUGE_ADVISE:
392 * only allocate huge pages if requested with fadvise()/madvise();
393 */
394
395#define SHMEM_HUGE_NEVER 0
396#define SHMEM_HUGE_ALWAYS 1
397#define SHMEM_HUGE_WITHIN_SIZE 2
398#define SHMEM_HUGE_ADVISE 3
399
400/*
401 * Special values.
402 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
403 *
404 * SHMEM_HUGE_DENY:
405 * disables huge on shm_mnt and all mounts, for emergency use;
406 * SHMEM_HUGE_FORCE:
407 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
408 *
409 */
410#define SHMEM_HUGE_DENY (-1)
411#define SHMEM_HUGE_FORCE (-2)
412
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700413#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700414/* ifdef here to avoid bloating shmem.o when not necessary */
415
Mike Kravetz5b9c98f2018-06-07 17:05:53 -0700416static int shmem_huge __read_mostly;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700417
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700418#if defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700419static int shmem_parse_huge(const char *str)
420{
421 if (!strcmp(str, "never"))
422 return SHMEM_HUGE_NEVER;
423 if (!strcmp(str, "always"))
424 return SHMEM_HUGE_ALWAYS;
425 if (!strcmp(str, "within_size"))
426 return SHMEM_HUGE_WITHIN_SIZE;
427 if (!strcmp(str, "advise"))
428 return SHMEM_HUGE_ADVISE;
429 if (!strcmp(str, "deny"))
430 return SHMEM_HUGE_DENY;
431 if (!strcmp(str, "force"))
432 return SHMEM_HUGE_FORCE;
433 return -EINVAL;
434}
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700435#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700436
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700437#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700438static const char *shmem_format_huge(int huge)
439{
440 switch (huge) {
441 case SHMEM_HUGE_NEVER:
442 return "never";
443 case SHMEM_HUGE_ALWAYS:
444 return "always";
445 case SHMEM_HUGE_WITHIN_SIZE:
446 return "within_size";
447 case SHMEM_HUGE_ADVISE:
448 return "advise";
449 case SHMEM_HUGE_DENY:
450 return "deny";
451 case SHMEM_HUGE_FORCE:
452 return "force";
453 default:
454 VM_BUG_ON(1);
455 return "bad_val";
456 }
457}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800458#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700459
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700460static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
461 struct shrink_control *sc, unsigned long nr_to_split)
462{
463 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800464 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700465 struct inode *inode;
466 struct shmem_inode_info *info;
467 struct page *page;
468 unsigned long batch = sc ? sc->nr_to_scan : 128;
469 int removed = 0, split = 0;
470
471 if (list_empty(&sbinfo->shrinklist))
472 return SHRINK_STOP;
473
474 spin_lock(&sbinfo->shrinklist_lock);
475 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
476 info = list_entry(pos, struct shmem_inode_info, shrinklist);
477
478 /* pin the inode */
479 inode = igrab(&info->vfs_inode);
480
481 /* inode is about to be evicted */
482 if (!inode) {
483 list_del_init(&info->shrinklist);
484 removed++;
485 goto next;
486 }
487
488 /* Check if there's anything to gain */
489 if (round_up(inode->i_size, PAGE_SIZE) ==
490 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800491 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700492 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700493 goto next;
494 }
495
496 list_move(&info->shrinklist, &list);
497next:
498 if (!--batch)
499 break;
500 }
501 spin_unlock(&sbinfo->shrinklist_lock);
502
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800503 list_for_each_safe(pos, next, &to_remove) {
504 info = list_entry(pos, struct shmem_inode_info, shrinklist);
505 inode = &info->vfs_inode;
506 list_del_init(&info->shrinklist);
507 iput(inode);
508 }
509
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700510 list_for_each_safe(pos, next, &list) {
511 int ret;
512
513 info = list_entry(pos, struct shmem_inode_info, shrinklist);
514 inode = &info->vfs_inode;
515
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700516 if (nr_to_split && split >= nr_to_split)
517 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700518
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700519 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700520 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
521 if (!page)
522 goto drop;
523
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700524 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700525 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700526 put_page(page);
527 goto drop;
528 }
529
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700530 /*
531 * Leave the inode on the list if we failed to lock
532 * the page at this time.
533 *
534 * Waiting for the lock may lead to deadlock in the
535 * reclaim path.
536 */
537 if (!trylock_page(page)) {
538 put_page(page);
539 goto leave;
540 }
541
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700542 ret = split_huge_page(page);
543 unlock_page(page);
544 put_page(page);
545
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700546 /* If split failed leave the inode on the list */
547 if (ret)
548 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700549
550 split++;
551drop:
552 list_del_init(&info->shrinklist);
553 removed++;
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700554leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700555 iput(inode);
556 }
557
558 spin_lock(&sbinfo->shrinklist_lock);
559 list_splice_tail(&list, &sbinfo->shrinklist);
560 sbinfo->shrinklist_len -= removed;
561 spin_unlock(&sbinfo->shrinklist_lock);
562
563 return split;
564}
565
566static long shmem_unused_huge_scan(struct super_block *sb,
567 struct shrink_control *sc)
568{
569 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
570
571 if (!READ_ONCE(sbinfo->shrinklist_len))
572 return SHRINK_STOP;
573
574 return shmem_unused_huge_shrink(sbinfo, sc, 0);
575}
576
577static long shmem_unused_huge_count(struct super_block *sb,
578 struct shrink_control *sc)
579{
580 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
581 return READ_ONCE(sbinfo->shrinklist_len);
582}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700583#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700584
585#define shmem_huge SHMEM_HUGE_DENY
586
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700587static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
588 struct shrink_control *sc, unsigned long nr_to_split)
589{
590 return 0;
591}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700592#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700593
Yang Shi89fdcd22018-06-07 17:06:59 -0700594static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
595{
596 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
597 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
598 shmem_huge != SHMEM_HUGE_DENY)
599 return true;
600 return false;
601}
602
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700603/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700604 * Like add_to_page_cache_locked, but error if expected item has gone.
605 */
606static int shmem_add_to_page_cache(struct page *page,
607 struct address_space *mapping,
Matthew Wilcox552446a2017-12-01 13:25:14 -0500608 pgoff_t index, void *expected, gfp_t gfp)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700609{
Matthew Wilcox552446a2017-12-01 13:25:14 -0500610 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
611 unsigned long i = 0;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -0700612 unsigned long nr = compound_nr(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700613
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700614 VM_BUG_ON_PAGE(PageTail(page), page);
615 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800616 VM_BUG_ON_PAGE(!PageLocked(page), page);
617 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700618 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700619
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700620 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700621 page->mapping = mapping;
622 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700623
Matthew Wilcox552446a2017-12-01 13:25:14 -0500624 do {
625 void *entry;
626 xas_lock_irq(&xas);
627 entry = xas_find_conflict(&xas);
628 if (entry != expected)
629 xas_set_err(&xas, -EEXIST);
630 xas_create_range(&xas);
631 if (xas_error(&xas))
632 goto unlock;
633next:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700634 xas_store(&xas, page);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500635 if (++i < nr) {
636 xas_next(&xas);
637 goto next;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700638 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500639 if (PageTransHuge(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700640 count_vm_event(THP_FILE_ALLOC);
Mel Gorman11fb9982016-07-28 15:46:20 -0700641 __inc_node_page_state(page, NR_SHMEM_THPS);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500642 }
643 mapping->nrpages += nr;
Mel Gorman11fb9982016-07-28 15:46:20 -0700644 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
645 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500646unlock:
647 xas_unlock_irq(&xas);
648 } while (xas_nomem(&xas, gfp));
649
650 if (xas_error(&xas)) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700651 page->mapping = NULL;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700652 page_ref_sub(page, nr);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500653 return xas_error(&xas);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700654 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500655
656 return 0;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700657}
658
659/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700660 * Like delete_from_page_cache, but substitutes swap for page.
661 */
662static void shmem_delete_from_page_cache(struct page *page, void *radswap)
663{
664 struct address_space *mapping = page->mapping;
665 int error;
666
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700667 VM_BUG_ON_PAGE(PageCompound(page), page);
668
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700669 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500670 error = shmem_replace_entry(mapping, page->index, page, radswap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700671 page->mapping = NULL;
672 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700673 __dec_node_page_state(page, NR_FILE_PAGES);
674 __dec_node_page_state(page, NR_SHMEM);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700675 xa_unlock_irq(&mapping->i_pages);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300676 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700677 BUG_ON(error);
678}
679
680/*
Matthew Wilcoxc121d3b2017-12-04 03:13:54 -0500681 * Remove swap entry from page cache, free the swap and its page cache.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700682 */
683static int shmem_free_swap(struct address_space *mapping,
684 pgoff_t index, void *radswap)
685{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700686 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700687
Matthew Wilcox55f3f7e2018-11-26 16:08:43 -0500688 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700689 if (old != radswap)
690 return -ENOENT;
691 free_swap_and_cache(radix_to_swp_entry(radswap));
692 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700693}
694
695/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800696 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800697 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800698 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700699 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800700 * as long as the inode doesn't go away and racy results are not a problem.
701 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800702unsigned long shmem_partial_swap_usage(struct address_space *mapping,
703 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800704{
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500705 XA_STATE(xas, &mapping->i_pages, start);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800706 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800707 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800708
709 rcu_read_lock();
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500710 xas_for_each(&xas, page, end - 1) {
711 if (xas_retry(&xas, page))
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700712 continue;
Matthew Wilcox3159f942017-11-03 13:30:42 -0400713 if (xa_is_value(page))
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800714 swapped++;
715
716 if (need_resched()) {
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500717 xas_pause(&xas);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800718 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800719 }
720 }
721
722 rcu_read_unlock();
723
724 return swapped << PAGE_SHIFT;
725}
726
727/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800728 * Determine (in bytes) how many of the shmem object's pages mapped by the
729 * given vma is swapped out.
730 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700731 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka48131e02016-01-14 15:19:23 -0800732 * as long as the inode doesn't go away and racy results are not a problem.
733 */
734unsigned long shmem_swap_usage(struct vm_area_struct *vma)
735{
736 struct inode *inode = file_inode(vma->vm_file);
737 struct shmem_inode_info *info = SHMEM_I(inode);
738 struct address_space *mapping = inode->i_mapping;
739 unsigned long swapped;
740
741 /* Be careful as we don't hold info->lock */
742 swapped = READ_ONCE(info->swapped);
743
744 /*
745 * The easier cases are when the shmem object has nothing in swap, or
746 * the vma maps it whole. Then we can simply use the stats that we
747 * already track.
748 */
749 if (!swapped)
750 return 0;
751
752 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
753 return swapped << PAGE_SHIFT;
754
755 /* Here comes the more involved part */
756 return shmem_partial_swap_usage(mapping,
757 linear_page_index(vma, vma->vm_start),
758 linear_page_index(vma, vma->vm_end));
759}
760
761/*
Hugh Dickins24513262012-01-20 14:34:21 -0800762 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
763 */
764void shmem_unlock_mapping(struct address_space *mapping)
765{
766 struct pagevec pvec;
767 pgoff_t indices[PAGEVEC_SIZE];
768 pgoff_t index = 0;
769
Mel Gorman86679822017-11-15 17:37:52 -0800770 pagevec_init(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800771 /*
772 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
773 */
774 while (!mapping_unevictable(mapping)) {
775 /*
776 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
777 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
778 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700779 pvec.nr = find_get_entries(mapping, index,
780 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800781 if (!pvec.nr)
782 break;
783 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700784 pagevec_remove_exceptionals(&pvec);
Kuo-Hsin Yang64e3d122018-11-06 13:23:24 +0000785 check_move_unevictable_pages(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800786 pagevec_release(&pvec);
787 cond_resched();
788 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700789}
790
791/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -0500792 * Remove range of pages and swap entries from page cache, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700793 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700794 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700795static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
796 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700798 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300800 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
801 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
802 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
803 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700804 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700805 pgoff_t indices[PAGEVEC_SIZE];
806 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700807 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700808 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700810 if (lend == -1)
811 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700812
Mel Gorman86679822017-11-15 17:37:52 -0800813 pagevec_init(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700814 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700815 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700816 pvec.nr = find_get_entries(mapping, index,
817 min(end - index, (pgoff_t)PAGEVEC_SIZE),
818 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700819 if (!pvec.nr)
820 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700821 for (i = 0; i < pagevec_count(&pvec); i++) {
822 struct page *page = pvec.pages[i];
823
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700824 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700825 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700826 break;
827
Matthew Wilcox3159f942017-11-03 13:30:42 -0400828 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700829 if (unfalloc)
830 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700831 nr_swaps_freed += !shmem_free_swap(mapping,
832 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700833 continue;
834 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700835
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700836 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
837
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700838 if (!trylock_page(page))
839 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700840
841 if (PageTransTail(page)) {
842 /* Middle of THP: zero out the page */
843 clear_highpage(page);
844 unlock_page(page);
845 continue;
846 } else if (PageTransHuge(page)) {
847 if (index == round_down(end, HPAGE_PMD_NR)) {
848 /*
849 * Range ends in the middle of THP:
850 * zero out the page
851 */
852 clear_highpage(page);
853 unlock_page(page);
854 continue;
855 }
856 index += HPAGE_PMD_NR - 1;
857 i += HPAGE_PMD_NR - 1;
858 }
859
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700860 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700861 VM_BUG_ON_PAGE(PageTail(page), page);
862 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800863 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700864 truncate_inode_page(mapping, page);
865 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700866 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700867 unlock_page(page);
868 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700869 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800870 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700871 cond_resched();
872 index++;
873 }
874
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700875 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700876 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700877 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700878 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300879 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700880 if (start > end) {
881 top = partial_end;
882 partial_end = 0;
883 }
884 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700885 set_page_dirty(page);
886 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300887 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700888 }
889 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700890 if (partial_end) {
891 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700892 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700893 if (page) {
894 zero_user_segment(page, 0, partial_end);
895 set_page_dirty(page);
896 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300897 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700898 }
899 }
900 if (start >= end)
901 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700902
903 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700904 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700905 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700906
907 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700908 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700909 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700910 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700911 /* If all gone or hole-punch or unfalloc, we're done */
912 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700913 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700914 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700915 index = start;
916 continue;
917 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700918 for (i = 0; i < pagevec_count(&pvec); i++) {
919 struct page *page = pvec.pages[i];
920
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700921 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700922 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700923 break;
924
Matthew Wilcox3159f942017-11-03 13:30:42 -0400925 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700926 if (unfalloc)
927 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700928 if (shmem_free_swap(mapping, index, page)) {
929 /* Swap was replaced by page: retry */
930 index--;
931 break;
932 }
933 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700934 continue;
935 }
936
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700937 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700938
939 if (PageTransTail(page)) {
940 /* Middle of THP: zero out the page */
941 clear_highpage(page);
942 unlock_page(page);
943 /*
944 * Partial thp truncate due 'start' in middle
945 * of THP: don't need to look on these pages
946 * again on !pvec.nr restart.
947 */
948 if (index != round_down(end, HPAGE_PMD_NR))
949 start++;
950 continue;
951 } else if (PageTransHuge(page)) {
952 if (index == round_down(end, HPAGE_PMD_NR)) {
953 /*
954 * Range ends in the middle of THP:
955 * zero out the page
956 */
957 clear_highpage(page);
958 unlock_page(page);
959 continue;
960 }
961 index += HPAGE_PMD_NR - 1;
962 i += HPAGE_PMD_NR - 1;
963 }
964
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700965 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700966 VM_BUG_ON_PAGE(PageTail(page), page);
967 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800968 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700969 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700970 } else {
971 /* Page was replaced by swap: retry */
972 unlock_page(page);
973 index--;
974 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700975 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700976 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700977 unlock_page(page);
978 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700979 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800980 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700981 index++;
982 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700983
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700984 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700985 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700987 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700988}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700990void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
991{
992 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700993 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700995EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
David Howellsa528d352017-01-31 16:46:22 +0000997static int shmem_getattr(const struct path *path, struct kstat *stat,
998 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -0700999{
David Howellsa528d352017-01-31 16:46:22 +00001000 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -07001001 struct shmem_inode_info *info = SHMEM_I(inode);
Yang Shi89fdcd22018-06-07 17:06:59 -07001002 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
Yu Zhao44a30222015-09-08 15:03:33 -07001003
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001004 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001005 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001006 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001007 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001008 }
Yu Zhao44a30222015-09-08 15:03:33 -07001009 generic_fillattr(inode, stat);
Yang Shi89fdcd22018-06-07 17:06:59 -07001010
1011 if (is_huge_enabled(sb_info))
1012 stat->blksize = HPAGE_PMD_SIZE;
1013
Yu Zhao44a30222015-09-08 15:03:33 -07001014 return 0;
1015}
1016
Hugh Dickins94c1e622011-06-27 16:18:03 -07001017static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
David Howells75c3cfa2015-03-17 22:26:12 +00001019 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -07001020 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001021 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 int error;
1023
Jan Kara31051c82016-05-26 16:55:18 +02001024 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001025 if (error)
1026 return error;
1027
Hugh Dickins94c1e622011-06-27 16:18:03 -07001028 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1029 loff_t oldsize = inode->i_size;
1030 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001031
David Herrmann40e041a2014-08-08 14:25:27 -07001032 /* protected by i_mutex */
1033 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1034 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1035 return -EPERM;
1036
Hugh Dickins94c1e622011-06-27 16:18:03 -07001037 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001038 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1039 oldsize, newsize);
1040 if (error)
1041 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001042 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001043 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001044 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001045 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001046 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001047 if (oldsize > holebegin)
1048 unmap_mapping_range(inode->i_mapping,
1049 holebegin, 0, 1);
1050 if (info->alloced)
1051 shmem_truncate_range(inode,
1052 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001053 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001054 if (oldsize > holebegin)
1055 unmap_mapping_range(inode->i_mapping,
1056 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001057
1058 /*
1059 * Part of the huge page can be beyond i_size: subject
1060 * to shrink under memory pressure.
1061 */
1062 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1063 spin_lock(&sbinfo->shrinklist_lock);
Cong Wangd0413532017-08-10 15:24:24 -07001064 /*
1065 * _careful to defend against unlocked access to
1066 * ->shrink_list in shmem_unused_huge_shrink()
1067 */
1068 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001069 list_add_tail(&info->shrinklist,
1070 &sbinfo->shrinklist);
1071 sbinfo->shrinklist_len++;
1072 }
1073 spin_unlock(&sbinfo->shrinklist_lock);
1074 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001075 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
1077
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001078 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001079 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001080 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return error;
1082}
1083
Al Viro1f895f72010-06-05 19:10:41 -04001084static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001087 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001089 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 shmem_unacct_size(info->flags, inode->i_size);
1091 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001092 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001093 if (!list_empty(&info->shrinklist)) {
1094 spin_lock(&sbinfo->shrinklist_lock);
1095 if (!list_empty(&info->shrinklist)) {
1096 list_del_init(&info->shrinklist);
1097 sbinfo->shrinklist_len--;
1098 }
1099 spin_unlock(&sbinfo->shrinklist_lock);
1100 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001101 while (!list_empty(&info->swaplist)) {
1102 /* Wait while shmem_unuse() is scanning this inode... */
1103 wait_var_event(&info->stop_eviction,
1104 !atomic_read(&info->stop_eviction));
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001105 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001106 /* ...but beware of the race if we peeked too early */
1107 if (!atomic_read(&info->stop_eviction))
1108 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001109 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
Al Viro3ed47db2016-01-22 18:08:52 -05001111 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001112
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001113 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001114 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001115 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001116 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117}
1118
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001119extern struct swap_info_struct *swap_info[];
1120
1121static int shmem_find_swap_entries(struct address_space *mapping,
1122 pgoff_t start, unsigned int nr_entries,
1123 struct page **entries, pgoff_t *indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001124 unsigned int type, bool frontswap)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001125{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001126 XA_STATE(xas, &mapping->i_pages, start);
1127 struct page *page;
Hugh Dickins87039542019-04-18 17:49:58 -07001128 swp_entry_t entry;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001129 unsigned int ret = 0;
1130
1131 if (!nr_entries)
1132 return 0;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001133
1134 rcu_read_lock();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001135 xas_for_each(&xas, page, ULONG_MAX) {
1136 if (xas_retry(&xas, page))
Mike Kravetz5b9c98f2018-06-07 17:05:53 -07001137 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001138
1139 if (!xa_is_value(page))
1140 continue;
1141
Hugh Dickins87039542019-04-18 17:49:58 -07001142 entry = radix_to_swp_entry(page);
1143 if (swp_type(entry) != type)
1144 continue;
1145 if (frontswap &&
1146 !frontswap_test(swap_info[type], swp_offset(entry)))
1147 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001148
1149 indices[ret] = xas.xa_index;
1150 entries[ret] = page;
1151
1152 if (need_resched()) {
1153 xas_pause(&xas);
1154 cond_resched_rcu();
1155 }
1156 if (++ret == nr_entries)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001157 break;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001158 }
Matthew Wilcox478922e2016-12-14 15:08:52 -08001159 rcu_read_unlock();
Matthew Wilcoxe21a2952017-11-22 08:36:00 -05001160
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001161 return ret;
1162}
1163
1164/*
1165 * Move the swapped pages for an inode to page cache. Returns the count
1166 * of pages swapped in, or the error in case of failure.
1167 */
1168static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1169 pgoff_t *indices)
1170{
1171 int i = 0;
1172 int ret = 0;
1173 int error = 0;
1174 struct address_space *mapping = inode->i_mapping;
1175
1176 for (i = 0; i < pvec.nr; i++) {
1177 struct page *page = pvec.pages[i];
1178
1179 if (!xa_is_value(page))
1180 continue;
1181 error = shmem_swapin_page(inode, indices[i],
1182 &page, SGP_CACHE,
1183 mapping_gfp_mask(mapping),
1184 NULL, NULL);
1185 if (error == 0) {
1186 unlock_page(page);
1187 put_page(page);
1188 ret++;
1189 }
1190 if (error == -ENOMEM)
1191 break;
1192 error = 0;
1193 }
1194 return error ? error : ret;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001195}
1196
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001197/*
1198 * If swap found in inode, free it and move page from swapcache to filecache.
1199 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001200static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1201 bool frontswap, unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001203 struct address_space *mapping = inode->i_mapping;
1204 pgoff_t start = 0;
1205 struct pagevec pvec;
1206 pgoff_t indices[PAGEVEC_SIZE];
1207 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1208 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001210 pagevec_init(&pvec);
1211 do {
1212 unsigned int nr_entries = PAGEVEC_SIZE;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001213
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001214 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1215 nr_entries = *fs_pages_to_unuse;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001216
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001217 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1218 pvec.pages, indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001219 type, frontswap);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001220 if (pvec.nr == 0) {
1221 ret = 0;
1222 break;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001223 }
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001224
1225 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1226 if (ret < 0)
1227 break;
1228
1229 if (frontswap_partial) {
1230 *fs_pages_to_unuse -= ret;
1231 if (*fs_pages_to_unuse == 0) {
1232 ret = FRONTSWAP_PAGES_UNUSED;
1233 break;
1234 }
1235 }
1236
1237 start = indices[pvec.nr - 1];
1238 } while (true);
1239
1240 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241}
1242
1243/*
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001244 * Read all the shared memory data that resides in the swap
1245 * device 'type' back into memory, so the swap device can be
1246 * unused.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001248int shmem_unuse(unsigned int type, bool frontswap,
1249 unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001251 struct shmem_inode_info *info, *next;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001252 int error = 0;
1253
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001254 if (list_empty(&shmem_swaplist))
1255 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001257 mutex_lock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001258 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1259 if (!info->swapped) {
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001260 list_del_init(&info->swaplist);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001261 continue;
1262 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001263 /*
1264 * Drop the swaplist mutex while searching the inode for swap;
1265 * but before doing so, make sure shmem_evict_inode() will not
1266 * remove placeholder inode from swaplist, nor let it be freed
1267 * (igrab() would protect from unlink, but not from unmount).
1268 */
1269 atomic_inc(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001270 mutex_unlock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001271
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001272 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001273 fs_pages_to_unuse);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001274 cond_resched();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001275
1276 mutex_lock(&shmem_swaplist_mutex);
1277 next = list_next_entry(info, swaplist);
1278 if (!info->swapped)
1279 list_del_init(&info->swaplist);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001280 if (atomic_dec_and_test(&info->stop_eviction))
1281 wake_up_var(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001282 if (error)
Hugh Dickins778dd892011-05-11 15:13:37 -07001283 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001285 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001286
Hugh Dickins778dd892011-05-11 15:13:37 -07001287 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288}
1289
1290/*
1291 * Move the page from the page cache to the swap cache.
1292 */
1293static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1294{
1295 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001298 swp_entry_t swap;
1299 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001301 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 mapping = page->mapping;
1304 index = page->index;
1305 inode = mapping->host;
1306 info = SHMEM_I(inode);
1307 if (info->flags & VM_LOCKED)
1308 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001309 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 goto redirty;
1311
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001312 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001313 * Our capabilities prevent regular writeback or sync from ever calling
1314 * shmem_writepage; but a stacking filesystem might use ->writepage of
1315 * its underlying filesystem, in which case tmpfs should write out to
1316 * swap only in response to memory pressure, and not for the writeback
1317 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001318 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001319 if (!wbc->for_reclaim) {
1320 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1321 goto redirty;
1322 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001323
1324 /*
1325 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1326 * value into swapfile.c, the only way we can correctly account for a
1327 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001328 *
1329 * That's okay for a page already fallocated earlier, but if we have
1330 * not yet completed the fallocation, then (a) we want to keep track
1331 * of this page in case we have to undo it, and (b) it may not be a
1332 * good idea to continue anyway, once we're pushing into swap. So
1333 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001334 */
1335 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001336 if (inode->i_private) {
1337 struct shmem_falloc *shmem_falloc;
1338 spin_lock(&inode->i_lock);
1339 shmem_falloc = inode->i_private;
1340 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001341 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001342 index >= shmem_falloc->start &&
1343 index < shmem_falloc->next)
1344 shmem_falloc->nr_unswapped++;
1345 else
1346 shmem_falloc = NULL;
1347 spin_unlock(&inode->i_lock);
1348 if (shmem_falloc)
1349 goto redirty;
1350 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001351 clear_highpage(page);
1352 flush_dcache_page(page);
1353 SetPageUptodate(page);
1354 }
1355
Huang Ying38d8b4e2017-07-06 15:37:18 -07001356 swap = get_swap_page(page);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001357 if (!swap.val)
1358 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001359
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001360 /*
1361 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001362 * if it's not already there. Do it now before the page is
1363 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001364 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001365 * we've incremented swapped, because shmem_unuse_inode() will
1366 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001367 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001368 mutex_lock(&shmem_swaplist_mutex);
1369 if (list_empty(&info->swaplist))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001370 list_add(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001371
Yang Shi4afab1c2019-11-30 17:58:07 -08001372 if (add_to_swap_cache(page, swap,
1373 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001374 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001375 shmem_recalc_inode(inode);
1376 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001377 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001378
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001379 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001380 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1381
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001382 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001383 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001384 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return 0;
1386 }
1387
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001388 mutex_unlock(&shmem_swaplist_mutex);
Minchan Kim75f6d6d2017-07-06 15:37:21 -07001389 put_swap_page(page, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390redirty:
1391 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001392 if (wbc->for_reclaim)
1393 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1394 unlock_page(page);
1395 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396}
1397
Hugh Dickins75edd342016-05-19 17:12:44 -07001398#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001399static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001400{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001401 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001402
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001403 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001404 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001405
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001406 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001407
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001408 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001409}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001410
1411static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1412{
1413 struct mempolicy *mpol = NULL;
1414 if (sbinfo->mpol) {
1415 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1416 mpol = sbinfo->mpol;
1417 mpol_get(mpol);
1418 spin_unlock(&sbinfo->stat_lock);
1419 }
1420 return mpol;
1421}
Hugh Dickins75edd342016-05-19 17:12:44 -07001422#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1423static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1424{
1425}
1426static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1427{
1428 return NULL;
1429}
1430#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1431#ifndef CONFIG_NUMA
1432#define vm_policy vm_private_data
1433#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001434
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001435static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1436 struct shmem_inode_info *info, pgoff_t index)
1437{
1438 /* Create a pseudo vma that just contains the policy */
Kirill A. Shutemov2c4541e2018-07-26 16:37:30 -07001439 vma_init(vma, NULL);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001440 /* Bias interleave by inode number to distribute better across nodes */
1441 vma->vm_pgoff = index + info->vfs_inode.i_ino;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001442 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1443}
1444
1445static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1446{
1447 /* Drop reference taken by mpol_shared_policy_lookup() */
1448 mpol_cond_put(vma->vm_policy);
1449}
1450
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001451static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1452 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001455 struct page *page;
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001456 struct vm_fault vmf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001458 shmem_pseudo_vma_init(&pvma, info, index);
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001459 vmf.vma = &pvma;
1460 vmf.address = 0;
1461 page = swap_cluster_readahead(swap, gfp, &vmf);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001462 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001463
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001464 return page;
1465}
Mel Gorman18a2f372012-12-05 14:01:41 -08001466
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001467static struct page *shmem_alloc_hugepage(gfp_t gfp,
1468 struct shmem_inode_info *info, pgoff_t index)
1469{
1470 struct vm_area_struct pvma;
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001471 struct address_space *mapping = info->vfs_inode.i_mapping;
1472 pgoff_t hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001473 struct page *page;
1474
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001475 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001476 return NULL;
1477
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001478 hindex = round_down(index, HPAGE_PMD_NR);
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001479 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1480 XA_PRESENT))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001481 return NULL;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001482
1483 shmem_pseudo_vma_init(&pvma, info, hindex);
1484 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
David Rientjes19deb762019-09-04 12:54:20 -07001485 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001486 shmem_pseudo_vma_destroy(&pvma);
1487 if (page)
1488 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001489 return page;
1490}
1491
1492static struct page *shmem_alloc_page(gfp_t gfp,
1493 struct shmem_inode_info *info, pgoff_t index)
1494{
1495 struct vm_area_struct pvma;
1496 struct page *page;
1497
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001498 shmem_pseudo_vma_init(&pvma, info, index);
1499 page = alloc_page_vma(gfp, &pvma, 0);
1500 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001501
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001502 return page;
1503}
1504
1505static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
Mike Rapoport0f079692017-09-06 16:22:59 -07001506 struct inode *inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001507 pgoff_t index, bool huge)
1508{
Mike Rapoport0f079692017-09-06 16:22:59 -07001509 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001510 struct page *page;
1511 int nr;
1512 int err = -ENOSPC;
1513
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001514 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001515 huge = false;
1516 nr = huge ? HPAGE_PMD_NR : 1;
1517
Mike Rapoport0f079692017-09-06 16:22:59 -07001518 if (!shmem_inode_acct_block(inode, nr))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001519 goto failed;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001520
1521 if (huge)
1522 page = shmem_alloc_hugepage(gfp, info, index);
1523 else
1524 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001525 if (page) {
1526 __SetPageLocked(page);
1527 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001528 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001529 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001530
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001531 err = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07001532 shmem_inode_unacct_blocks(inode, nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001533failed:
1534 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001536
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001538 * When a page is moved from swapcache to shmem filecache (either by the
1539 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1540 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1541 * ignorance of the mapping it belongs to. If that mapping has special
1542 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1543 * we may need to copy to a suitable page before moving to filecache.
1544 *
1545 * In a future release, this may well be extended to respect cpuset and
1546 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1547 * but for now it is a simple matter of zone.
1548 */
1549static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1550{
1551 return page_zonenum(page) > gfp_zone(gfp);
1552}
1553
1554static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1555 struct shmem_inode_info *info, pgoff_t index)
1556{
1557 struct page *oldpage, *newpage;
1558 struct address_space *swap_mapping;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001559 swp_entry_t entry;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001560 pgoff_t swap_index;
1561 int error;
1562
1563 oldpage = *pagep;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001564 entry.val = page_private(oldpage);
1565 swap_index = swp_offset(entry);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001566 swap_mapping = page_mapping(oldpage);
1567
1568 /*
1569 * We have arrived here because our zones are constrained, so don't
1570 * limit chance of success by further cpuset and node constraints.
1571 */
1572 gfp &= ~GFP_CONSTRAINT_MASK;
1573 newpage = shmem_alloc_page(gfp, info, index);
1574 if (!newpage)
1575 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001576
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001577 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001578 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001579 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001580
Hugh Dickins9956edf2016-11-10 10:46:11 -08001581 __SetPageLocked(newpage);
1582 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001583 SetPageUptodate(newpage);
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001584 set_page_private(newpage, entry.val);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001585 SetPageSwapCache(newpage);
1586
1587 /*
1588 * Our caller will very soon move newpage out of swapcache, but it's
1589 * a nice clean interface for us to replace oldpage by newpage there.
1590 */
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001591 xa_lock_irq(&swap_mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -05001592 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001593 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001594 __inc_node_page_state(newpage, NR_FILE_PAGES);
1595 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001596 }
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001597 xa_unlock_irq(&swap_mapping->i_pages);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001598
Hugh Dickins0142ef62012-06-07 14:21:09 -07001599 if (unlikely(error)) {
1600 /*
1601 * Is this possible? I think not, now that our callers check
1602 * both PageSwapCache and page_private after getting page lock;
1603 * but be defensive. Reverse old to newpage for clear and free.
1604 */
1605 oldpage = newpage;
1606 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001607 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001608 lru_cache_add_anon(newpage);
1609 *pagep = newpage;
1610 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001611
1612 ClearPageSwapCache(oldpage);
1613 set_page_private(oldpage, 0);
1614
1615 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001616 put_page(oldpage);
1617 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001618 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001619}
1620
1621/*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001622 * Swap in the page pointed to by *pagep.
1623 * Caller has to make sure that *pagep contains a valid swapped page.
1624 * Returns 0 and the page in pagep if success. On failure, returns the
1625 * the error code and NULL in *pagep.
1626 */
1627static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1628 struct page **pagep, enum sgp_type sgp,
1629 gfp_t gfp, struct vm_area_struct *vma,
1630 vm_fault_t *fault_type)
1631{
1632 struct address_space *mapping = inode->i_mapping;
1633 struct shmem_inode_info *info = SHMEM_I(inode);
1634 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1635 struct mem_cgroup *memcg;
1636 struct page *page;
1637 swp_entry_t swap;
1638 int error;
1639
1640 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1641 swap = radix_to_swp_entry(*pagep);
1642 *pagep = NULL;
1643
1644 /* Look it up and read it in.. */
1645 page = lookup_swap_cache(swap, NULL, 0);
1646 if (!page) {
1647 /* Or update major stats only when swapin succeeds?? */
1648 if (fault_type) {
1649 *fault_type |= VM_FAULT_MAJOR;
1650 count_vm_event(PGMAJFAULT);
1651 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1652 }
1653 /* Here we actually start the io */
1654 page = shmem_swapin(swap, gfp, info, index);
1655 if (!page) {
1656 error = -ENOMEM;
1657 goto failed;
1658 }
1659 }
1660
1661 /* We have to do this with page locked to prevent races */
1662 lock_page(page);
1663 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1664 !shmem_confirm_swap(mapping, index, swap)) {
1665 error = -EEXIST;
1666 goto unlock;
1667 }
1668 if (!PageUptodate(page)) {
1669 error = -EIO;
1670 goto failed;
1671 }
1672 wait_on_page_writeback(page);
1673
1674 if (shmem_should_replace_page(page, gfp)) {
1675 error = shmem_replace_page(&page, gfp, info, index);
1676 if (error)
1677 goto failed;
1678 }
1679
1680 error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1681 false);
1682 if (!error) {
1683 error = shmem_add_to_page_cache(page, mapping, index,
1684 swp_to_radix_entry(swap), gfp);
1685 /*
1686 * We already confirmed swap under page lock, and make
1687 * no memory allocation here, so usually no possibility
1688 * of error; but free_swap_and_cache() only trylocks a
1689 * page, so it is just possible that the entry has been
1690 * truncated or holepunched since swap was confirmed.
1691 * shmem_undo_range() will have done some of the
1692 * unaccounting, now delete_from_swap_cache() will do
1693 * the rest.
1694 */
1695 if (error) {
1696 mem_cgroup_cancel_charge(page, memcg, false);
1697 delete_from_swap_cache(page);
1698 }
1699 }
1700 if (error)
1701 goto failed;
1702
1703 mem_cgroup_commit_charge(page, memcg, true, false);
1704
1705 spin_lock_irq(&info->lock);
1706 info->swapped--;
1707 shmem_recalc_inode(inode);
1708 spin_unlock_irq(&info->lock);
1709
1710 if (sgp == SGP_WRITE)
1711 mark_page_accessed(page);
1712
1713 delete_from_swap_cache(page);
1714 set_page_dirty(page);
1715 swap_free(swap);
1716
1717 *pagep = page;
1718 return 0;
1719failed:
1720 if (!shmem_confirm_swap(mapping, index, swap))
1721 error = -EEXIST;
1722unlock:
1723 if (page) {
1724 unlock_page(page);
1725 put_page(page);
1726 }
1727
1728 return error;
1729}
1730
1731/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001732 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 *
1734 * If we allocate a new one we do not mark it dirty. That's up to the
1735 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001736 * entry since a page cannot live in both the swap and page cache.
1737 *
Miles Chen28eb3c802019-09-23 15:39:34 -07001738 * vmf and fault_type are only supplied by shmem_fault:
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001739 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001741static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001742 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
Souptick Joarder2b740302018-08-23 17:01:36 -07001743 struct vm_area_struct *vma, struct vm_fault *vmf,
1744 vm_fault_t *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745{
1746 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann23f919d2016-12-12 16:42:28 -08001747 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001749 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001750 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001751 struct page *page;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001752 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001753 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001755 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001756 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001758 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001760 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1761 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762repeat:
Hugh Dickins75edd342016-05-19 17:12:44 -07001763 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001764 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001765 return -EINVAL;
1766 }
1767
1768 sbinfo = SHMEM_SB(inode->i_sb);
1769 charge_mm = vma ? vma->vm_mm : current->mm;
1770
1771 page = find_lock_entry(mapping, index);
1772 if (xa_is_value(page)) {
1773 error = shmem_swapin_page(inode, index, &page,
1774 sgp, gfp, vma, fault_type);
1775 if (error == -EEXIST)
1776 goto repeat;
1777
1778 *pagep = page;
1779 return error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001780 }
1781
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001782 if (page && sgp == SGP_WRITE)
1783 mark_page_accessed(page);
1784
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001785 /* fallocated page? */
1786 if (page && !PageUptodate(page)) {
1787 if (sgp != SGP_READ)
1788 goto clear;
1789 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001790 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001791 page = NULL;
1792 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001793 if (page || sgp == SGP_READ) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001794 *pagep = page;
1795 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001796 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001799 * Fast cache lookup did not find it:
1800 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001802
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001803 if (vma && userfaultfd_missing(vma)) {
1804 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1805 return 0;
1806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001808 /* shmem_symlink() */
1809 if (mapping->a_ops != &shmem_aops)
1810 goto alloc_nohuge;
1811 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1812 goto alloc_nohuge;
1813 if (shmem_huge == SHMEM_HUGE_FORCE)
1814 goto alloc_huge;
1815 switch (sbinfo->huge) {
1816 loff_t i_size;
1817 pgoff_t off;
1818 case SHMEM_HUGE_NEVER:
1819 goto alloc_nohuge;
1820 case SHMEM_HUGE_WITHIN_SIZE:
1821 off = round_up(index, HPAGE_PMD_NR);
1822 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1823 if (i_size >= HPAGE_PMD_SIZE &&
1824 i_size >> PAGE_SHIFT >= off)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001825 goto alloc_huge;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001826 /* fallthrough */
1827 case SHMEM_HUGE_ADVISE:
1828 if (sgp_huge == SGP_HUGE)
1829 goto alloc_huge;
1830 /* TODO: implement fadvise() hints */
1831 goto alloc_nohuge;
1832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001834alloc_huge:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001835 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1836 if (IS_ERR(page)) {
1837alloc_nohuge:
1838 page = shmem_alloc_and_acct_page(gfp, inode,
1839 index, false);
1840 }
1841 if (IS_ERR(page)) {
1842 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001843
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001844 error = PTR_ERR(page);
1845 page = NULL;
1846 if (error != -ENOSPC)
1847 goto unlock;
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001848 /*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001849 * Try to reclaim some space by splitting a huge page
1850 * beyond i_size on the filesystem.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001851 */
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001852 while (retry--) {
1853 int ret;
1854
1855 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1856 if (ret == SHRINK_STOP)
1857 break;
1858 if (ret)
1859 goto alloc_nohuge;
1860 }
1861 goto unlock;
1862 }
1863
1864 if (PageTransHuge(page))
1865 hindex = round_down(index, HPAGE_PMD_NR);
1866 else
1867 hindex = index;
1868
1869 if (sgp == SGP_WRITE)
1870 __SetPageReferenced(page);
1871
1872 error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1873 PageTransHuge(page));
1874 if (error)
1875 goto unacct;
1876 error = shmem_add_to_page_cache(page, mapping, hindex,
1877 NULL, gfp & GFP_RECLAIM_MASK);
1878 if (error) {
1879 mem_cgroup_cancel_charge(page, memcg,
1880 PageTransHuge(page));
1881 goto unacct;
1882 }
1883 mem_cgroup_commit_charge(page, memcg, false,
1884 PageTransHuge(page));
1885 lru_cache_add_anon(page);
1886
1887 spin_lock_irq(&info->lock);
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001888 info->alloced += compound_nr(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001889 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1890 shmem_recalc_inode(inode);
1891 spin_unlock_irq(&info->lock);
1892 alloced = true;
1893
1894 if (PageTransHuge(page) &&
1895 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1896 hindex + HPAGE_PMD_NR - 1) {
1897 /*
1898 * Part of the huge page is beyond i_size: subject
1899 * to shrink under memory pressure.
1900 */
1901 spin_lock(&sbinfo->shrinklist_lock);
1902 /*
1903 * _careful to defend against unlocked access to
1904 * ->shrink_list in shmem_unused_huge_shrink()
1905 */
1906 if (list_empty_careful(&info->shrinklist)) {
1907 list_add_tail(&info->shrinklist,
1908 &sbinfo->shrinklist);
1909 sbinfo->shrinklist_len++;
1910 }
1911 spin_unlock(&sbinfo->shrinklist_lock);
1912 }
1913
1914 /*
1915 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1916 */
1917 if (sgp == SGP_FALLOC)
1918 sgp = SGP_WRITE;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001919clear:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001920 /*
1921 * Let SGP_WRITE caller clear ends if write does not fill page;
1922 * but SGP_FALLOC on a page fallocated earlier must initialize
1923 * it now, lest undo on failure cancel our earlier guarantee.
1924 */
1925 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1926 struct page *head = compound_head(page);
1927 int i;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001928
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001929 for (i = 0; i < compound_nr(head); i++) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001930 clear_highpage(head + i);
1931 flush_dcache_page(head + i);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001932 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001933 SetPageUptodate(head);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001934 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001935
Hugh Dickins54af60422011-08-03 16:21:24 -07001936 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001937 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001938 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001939 if (alloced) {
1940 ClearPageDirty(page);
1941 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001942 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001943 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001944 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001945 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001946 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001947 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001948 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001949 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001950 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001951
Nick Piggind0217ac2007-07-19 01:47:03 -07001952 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001953 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001955unacct:
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001956 shmem_inode_unacct_blocks(inode, compound_nr(page));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001957
1958 if (PageTransHuge(page)) {
1959 unlock_page(page);
1960 put_page(page);
1961 goto alloc_nohuge;
1962 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001963unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001964 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001965 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001966 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001967 }
1968 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001969 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001970 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001971 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001972 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001973 }
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05001974 if (error == -EEXIST)
Hugh Dickins54af60422011-08-03 16:21:24 -07001975 goto repeat;
1976 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977}
1978
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001979/*
1980 * This is like autoremove_wake_function, but it removes the wait queue
1981 * entry unconditionally - even if something else had already woken the
1982 * target.
1983 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02001984static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001985{
1986 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02001987 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001988 return ret;
1989}
1990
Souptick Joarder20acce62018-06-07 17:09:17 -07001991static vm_fault_t shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992{
Dave Jiang11bac802017-02-24 14:56:41 -08001993 struct vm_area_struct *vma = vmf->vma;
Al Viro496ad9a2013-01-23 17:07:38 -05001994 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001995 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001996 enum sgp_type sgp;
Souptick Joarder20acce62018-06-07 17:09:17 -07001997 int err;
1998 vm_fault_t ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002000 /*
2001 * Trinity finds that probing a hole which tmpfs is punching can
2002 * prevent the hole-punch from ever completing: which in turn
2003 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07002004 * faulting pages into the hole while it's being punched. Although
2005 * shmem_undo_range() does remove the additions, it may be unable to
2006 * keep up, as each new page needs its own unmap_mapping_range() call,
2007 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2008 *
2009 * It does not matter if we sometimes reach this check just before the
2010 * hole-punch begins, so that one fault then races with the punch:
2011 * we just need to make racing faults a rare case.
2012 *
2013 * The implementation below would be much simpler if we just used a
2014 * standard mutex or completion: but we cannot take i_mutex in fault,
2015 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002016 */
2017 if (unlikely(inode->i_private)) {
2018 struct shmem_falloc *shmem_falloc;
2019
2020 spin_lock(&inode->i_lock);
2021 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002022 if (shmem_falloc &&
2023 shmem_falloc->waitq &&
2024 vmf->pgoff >= shmem_falloc->start &&
2025 vmf->pgoff < shmem_falloc->next) {
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002026 struct file *fpin;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002027 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002028 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002029
2030 ret = VM_FAULT_NOPAGE;
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002031 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2032 if (fpin)
Hugh Dickins8e205f72014-07-23 14:00:10 -07002033 ret = VM_FAULT_RETRY;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002034
2035 shmem_falloc_waitq = shmem_falloc->waitq;
2036 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2037 TASK_UNINTERRUPTIBLE);
2038 spin_unlock(&inode->i_lock);
2039 schedule();
2040
2041 /*
2042 * shmem_falloc_waitq points into the shmem_fallocate()
2043 * stack of the hole-punching task: shmem_falloc_waitq
2044 * is usually invalid by the time we reach here, but
2045 * finish_wait() does not dereference it in that case;
2046 * though i_lock needed lest racing with wake_up_all().
2047 */
2048 spin_lock(&inode->i_lock);
2049 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2050 spin_unlock(&inode->i_lock);
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002051
2052 if (fpin)
2053 fput(fpin);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002054 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002055 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07002056 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002057 }
2058
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002059 sgp = SGP_CACHE;
Michal Hocko18600332017-07-10 15:48:02 -07002060
2061 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2062 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002063 sgp = SGP_NOHUGE;
Michal Hocko18600332017-07-10 15:48:02 -07002064 else if (vma->vm_flags & VM_HUGEPAGE)
2065 sgp = SGP_HUGE;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002066
Souptick Joarder20acce62018-06-07 17:09:17 -07002067 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08002068 gfp, vma, vmf, &ret);
Souptick Joarder20acce62018-06-07 17:09:17 -07002069 if (err)
2070 return vmf_error(err);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002071 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072}
2073
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002074unsigned long shmem_get_unmapped_area(struct file *file,
2075 unsigned long uaddr, unsigned long len,
2076 unsigned long pgoff, unsigned long flags)
2077{
2078 unsigned long (*get_area)(struct file *,
2079 unsigned long, unsigned long, unsigned long, unsigned long);
2080 unsigned long addr;
2081 unsigned long offset;
2082 unsigned long inflated_len;
2083 unsigned long inflated_addr;
2084 unsigned long inflated_offset;
2085
2086 if (len > TASK_SIZE)
2087 return -ENOMEM;
2088
2089 get_area = current->mm->get_unmapped_area;
2090 addr = get_area(file, uaddr, len, pgoff, flags);
2091
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002092 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002093 return addr;
2094 if (IS_ERR_VALUE(addr))
2095 return addr;
2096 if (addr & ~PAGE_MASK)
2097 return addr;
2098 if (addr > TASK_SIZE - len)
2099 return addr;
2100
2101 if (shmem_huge == SHMEM_HUGE_DENY)
2102 return addr;
2103 if (len < HPAGE_PMD_SIZE)
2104 return addr;
2105 if (flags & MAP_FIXED)
2106 return addr;
2107 /*
2108 * Our priority is to support MAP_SHARED mapped hugely;
2109 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002110 * But if caller specified an address hint and we allocated area there
2111 * successfully, respect that as before.
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002112 */
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002113 if (uaddr == addr)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002114 return addr;
2115
2116 if (shmem_huge != SHMEM_HUGE_FORCE) {
2117 struct super_block *sb;
2118
2119 if (file) {
2120 VM_BUG_ON(file->f_op != &shmem_file_operations);
2121 sb = file_inode(file)->i_sb;
2122 } else {
2123 /*
2124 * Called directly from mm/mmap.c, or drivers/char/mem.c
2125 * for "/dev/zero", to create a shared anonymous object.
2126 */
2127 if (IS_ERR(shm_mnt))
2128 return addr;
2129 sb = shm_mnt->mnt_sb;
2130 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002131 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002132 return addr;
2133 }
2134
2135 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2136 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2137 return addr;
2138 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2139 return addr;
2140
2141 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2142 if (inflated_len > TASK_SIZE)
2143 return addr;
2144 if (inflated_len < len)
2145 return addr;
2146
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002147 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002148 if (IS_ERR_VALUE(inflated_addr))
2149 return addr;
2150 if (inflated_addr & ~PAGE_MASK)
2151 return addr;
2152
2153 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2154 inflated_addr += offset - inflated_offset;
2155 if (inflated_offset > offset)
2156 inflated_addr += HPAGE_PMD_SIZE;
2157
2158 if (inflated_addr > TASK_SIZE - len)
2159 return addr;
2160 return inflated_addr;
2161}
2162
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002164static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165{
Al Viro496ad9a2013-01-23 17:07:38 -05002166 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002167 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168}
2169
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002170static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2171 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172{
Al Viro496ad9a2013-01-23 17:07:38 -05002173 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002174 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002176 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2177 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178}
2179#endif
2180
2181int shmem_lock(struct file *file, int lock, struct user_struct *user)
2182{
Al Viro496ad9a2013-01-23 17:07:38 -05002183 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 struct shmem_inode_info *info = SHMEM_I(inode);
2185 int retval = -ENOMEM;
2186
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002187 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (lock && !(info->flags & VM_LOCKED)) {
2189 if (!user_shm_lock(inode->i_size, user))
2190 goto out_nomem;
2191 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002192 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 }
2194 if (!lock && (info->flags & VM_LOCKED) && user) {
2195 user_shm_unlock(inode->i_size, user);
2196 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002197 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 }
2199 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002202 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 return retval;
2204}
2205
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002206static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002208 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2209
2210 if (info->seals & F_SEAL_FUTURE_WRITE) {
2211 /*
2212 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2213 * "future write" seal active.
2214 */
2215 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2216 return -EPERM;
2217
2218 /*
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002219 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2220 * MAP_SHARED and read-only, take care to not allow mprotect to
2221 * revert protections on such mappings. Do this only for shared
2222 * mappings. For private mappings, don't need to mask
2223 * VM_MAYWRITE as we still want them to be COW-writable.
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002224 */
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002225 if (vma->vm_flags & VM_SHARED)
2226 vma->vm_flags &= ~(VM_MAYWRITE);
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002227 }
2228
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 file_accessed(file);
2230 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002231 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002232 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2233 (vma->vm_end & HPAGE_PMD_MASK)) {
2234 khugepaged_enter(vma, vma->vm_flags);
2235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 return 0;
2237}
2238
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002239static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002240 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241{
2242 struct inode *inode;
2243 struct shmem_inode_info *info;
2244 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2245
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002246 if (shmem_reserve_inode(sb))
2247 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
2249 inode = new_inode(sb);
2250 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002251 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002252 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002254 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
Arnd Bergmann46c9a942018-08-17 15:45:09 -07002255 inode->i_generation = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 info = SHMEM_I(inode);
2257 memset(info, 0, (char *)inode - (char *)info);
2258 spin_lock_init(&info->lock);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07002259 atomic_set(&info->stop_eviction, 0);
David Herrmann40e041a2014-08-08 14:25:27 -07002260 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002261 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002262 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002264 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002265 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
2267 switch (mode & S_IFMT) {
2268 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002269 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 init_special_inode(inode, mode, dev);
2271 break;
2272 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002273 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 inode->i_op = &shmem_inode_operations;
2275 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002276 mpol_shared_policy_init(&info->policy,
2277 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 break;
2279 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002280 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 /* Some things misbehave if size == 0 on a directory */
2282 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2283 inode->i_op = &shmem_dir_inode_operations;
2284 inode->i_fop = &simple_dir_operations;
2285 break;
2286 case S_IFLNK:
2287 /*
2288 * Must not load anything in the rbtree,
2289 * mpol_free_shared_policy will not be called.
2290 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002291 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 break;
2293 }
Joel Fernandes (Google)b45d71f2018-09-20 12:22:39 -07002294
2295 lockdep_annotate_inode_mutex_key(inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002296 } else
2297 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 return inode;
2299}
2300
Johannes Weiner0cd61442014-04-03 14:47:46 -07002301bool shmem_mapping(struct address_space *mapping)
2302{
Hugh Dickinsf8005452017-02-22 15:41:42 -08002303 return mapping->a_ops == &shmem_aops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002304}
2305
Mike Rapoport8d103962017-09-06 16:23:02 -07002306static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2307 pmd_t *dst_pmd,
2308 struct vm_area_struct *dst_vma,
2309 unsigned long dst_addr,
2310 unsigned long src_addr,
2311 bool zeropage,
2312 struct page **pagep)
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002313{
2314 struct inode *inode = file_inode(dst_vma->vm_file);
2315 struct shmem_inode_info *info = SHMEM_I(inode);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002316 struct address_space *mapping = inode->i_mapping;
2317 gfp_t gfp = mapping_gfp_mask(mapping);
2318 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2319 struct mem_cgroup *memcg;
2320 spinlock_t *ptl;
2321 void *page_kaddr;
2322 struct page *page;
2323 pte_t _dst_pte, *dst_pte;
2324 int ret;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002325 pgoff_t offset, max_off;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002326
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002327 ret = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07002328 if (!shmem_inode_acct_block(inode, 1))
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002329 goto out;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002330
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002331 if (!*pagep) {
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002332 page = shmem_alloc_page(gfp, info, pgoff);
2333 if (!page)
Mike Rapoport0f079692017-09-06 16:22:59 -07002334 goto out_unacct_blocks;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002335
Mike Rapoport8d103962017-09-06 16:23:02 -07002336 if (!zeropage) { /* mcopy_atomic */
2337 page_kaddr = kmap_atomic(page);
2338 ret = copy_from_user(page_kaddr,
2339 (const void __user *)src_addr,
2340 PAGE_SIZE);
2341 kunmap_atomic(page_kaddr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002342
Mike Rapoport8d103962017-09-06 16:23:02 -07002343 /* fallback to copy_from_user outside mmap_sem */
2344 if (unlikely(ret)) {
2345 *pagep = page;
2346 shmem_inode_unacct_blocks(inode, 1);
2347 /* don't free the page */
Andrea Arcangeli9e368252018-11-30 14:09:25 -08002348 return -ENOENT;
Mike Rapoport8d103962017-09-06 16:23:02 -07002349 }
2350 } else { /* mfill_zeropage_atomic */
2351 clear_highpage(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002352 }
2353 } else {
2354 page = *pagep;
2355 *pagep = NULL;
2356 }
2357
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002358 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2359 __SetPageLocked(page);
2360 __SetPageSwapBacked(page);
Andrea Arcangelia425d352017-02-22 15:43:52 -08002361 __SetPageUptodate(page);
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002362
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002363 ret = -EFAULT;
2364 offset = linear_page_index(dst_vma, dst_addr);
2365 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2366 if (unlikely(offset >= max_off))
2367 goto out_release;
2368
Tejun Heo2cf85582018-07-03 11:14:56 -04002369 ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002370 if (ret)
2371 goto out_release;
2372
Matthew Wilcox552446a2017-12-01 13:25:14 -05002373 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2374 gfp & GFP_RECLAIM_MASK);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002375 if (ret)
2376 goto out_release_uncharge;
2377
2378 mem_cgroup_commit_charge(page, memcg, false, false);
2379
2380 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2381 if (dst_vma->vm_flags & VM_WRITE)
2382 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002383 else {
2384 /*
2385 * We don't set the pte dirty if the vma has no
2386 * VM_WRITE permission, so mark the page dirty or it
2387 * could be freed from under us. We could do it
2388 * unconditionally before unlock_page(), but doing it
2389 * only if VM_WRITE is not set is faster.
2390 */
2391 set_page_dirty(page);
2392 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002393
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002394 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002395
2396 ret = -EFAULT;
2397 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2398 if (unlikely(offset >= max_off))
2399 goto out_release_uncharge_unlock;
2400
2401 ret = -EEXIST;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002402 if (!pte_none(*dst_pte))
2403 goto out_release_uncharge_unlock;
2404
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002405 lru_cache_add_anon(page);
2406
2407 spin_lock(&info->lock);
2408 info->alloced++;
2409 inode->i_blocks += BLOCKS_PER_PAGE;
2410 shmem_recalc_inode(inode);
2411 spin_unlock(&info->lock);
2412
2413 inc_mm_counter(dst_mm, mm_counter_file(page));
2414 page_add_file_rmap(page, false);
2415 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2416
2417 /* No need to invalidate - it was non-present before */
2418 update_mmu_cache(dst_vma, dst_addr, dst_pte);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002419 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002420 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002421 ret = 0;
2422out:
2423 return ret;
2424out_release_uncharge_unlock:
2425 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002426 ClearPageDirty(page);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002427 delete_from_page_cache(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002428out_release_uncharge:
2429 mem_cgroup_cancel_charge(page, memcg, false);
2430out_release:
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002431 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002432 put_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002433out_unacct_blocks:
Mike Rapoport0f079692017-09-06 16:22:59 -07002434 shmem_inode_unacct_blocks(inode, 1);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002435 goto out;
2436}
2437
Mike Rapoport8d103962017-09-06 16:23:02 -07002438int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2439 pmd_t *dst_pmd,
2440 struct vm_area_struct *dst_vma,
2441 unsigned long dst_addr,
2442 unsigned long src_addr,
2443 struct page **pagep)
2444{
2445 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2446 dst_addr, src_addr, false, pagep);
2447}
2448
2449int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2450 pmd_t *dst_pmd,
2451 struct vm_area_struct *dst_vma,
2452 unsigned long dst_addr)
2453{
2454 struct page *page = NULL;
2455
2456 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2457 dst_addr, 0, true, &page);
2458}
2459
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002461static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002462static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002464#ifdef CONFIG_TMPFS_XATTR
2465static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2466#else
2467#define shmem_initxattrs NULL
2468#endif
2469
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002471shmem_write_begin(struct file *file, struct address_space *mapping,
2472 loff_t pos, unsigned len, unsigned flags,
2473 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474{
Nick Piggin800d15a2007-10-16 01:25:03 -07002475 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002476 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002477 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002478
2479 /* i_mutex is held by caller */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002480 if (unlikely(info->seals & (F_SEAL_GROW |
2481 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2482 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
David Herrmann40e041a2014-08-08 14:25:27 -07002483 return -EPERM;
2484 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2485 return -EPERM;
2486 }
2487
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002488 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002489}
2490
2491static int
2492shmem_write_end(struct file *file, struct address_space *mapping,
2493 loff_t pos, unsigned len, unsigned copied,
2494 struct page *page, void *fsdata)
2495{
2496 struct inode *inode = mapping->host;
2497
Hugh Dickinsd3602442008-02-04 22:28:44 -08002498 if (pos + copied > inode->i_size)
2499 i_size_write(inode, pos + copied);
2500
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002501 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002502 struct page *head = compound_head(page);
2503 if (PageTransCompound(page)) {
2504 int i;
2505
2506 for (i = 0; i < HPAGE_PMD_NR; i++) {
2507 if (head + i == page)
2508 continue;
2509 clear_highpage(head + i);
2510 flush_dcache_page(head + i);
2511 }
2512 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002513 if (copied < PAGE_SIZE) {
2514 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002515 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002516 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002517 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002518 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002519 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002520 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002521 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002522 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002523
Nick Piggin800d15a2007-10-16 01:25:03 -07002524 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525}
2526
Al Viro2ba5bbe2014-04-02 20:00:02 -04002527static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528{
Al Viro6e58e792014-02-03 17:07:03 -05002529 struct file *file = iocb->ki_filp;
2530 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002532 pgoff_t index;
2533 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002534 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002535 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002536 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002537 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002538
2539 /*
2540 * Might this read be for a stacking filesystem? Then when reading
2541 * holes of a sparse file, we actually need to allocate those pages,
2542 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2543 */
Al Viro777eda22014-12-17 04:46:46 -05002544 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002545 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002547 index = *ppos >> PAGE_SHIFT;
2548 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
2550 for (;;) {
2551 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002552 pgoff_t end_index;
2553 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 loff_t i_size = i_size_read(inode);
2555
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002556 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 if (index > end_index)
2558 break;
2559 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002560 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 if (nr <= offset)
2562 break;
2563 }
2564
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002565 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002566 if (error) {
2567 if (error == -EINVAL)
2568 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 break;
2570 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002571 if (page) {
2572 if (sgp == SGP_CACHE)
2573 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002574 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002575 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576
2577 /*
2578 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002579 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002581 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002583 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002585 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 if (nr <= offset) {
2587 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002588 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 break;
2590 }
2591 }
2592 nr -= offset;
2593
2594 if (page) {
2595 /*
2596 * If users can be writing to this page using arbitrary
2597 * virtual addresses, take care about potential aliasing
2598 * before reading the page on the kernel side.
2599 */
2600 if (mapping_writably_mapped(mapping))
2601 flush_dcache_page(page);
2602 /*
2603 * Mark the page accessed if we read the beginning.
2604 */
2605 if (!offset)
2606 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002607 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002609 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002610 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
2612 /*
2613 * Ok, we have the page, and it's up-to-date, so
2614 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002616 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002617 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002619 index += offset >> PAGE_SHIFT;
2620 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002622 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002623 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 break;
Al Viro6e58e792014-02-03 17:07:03 -05002625 if (ret < nr) {
2626 error = -EFAULT;
2627 break;
2628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 cond_resched();
2630 }
2631
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002632 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002633 file_accessed(file);
2634 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635}
2636
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002637/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002638 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002639 */
2640static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002641 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002642{
2643 struct page *page;
2644 struct pagevec pvec;
2645 pgoff_t indices[PAGEVEC_SIZE];
2646 bool done = false;
2647 int i;
2648
Mel Gorman86679822017-11-15 17:37:52 -08002649 pagevec_init(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002650 pvec.nr = 1; /* start small: we may be there already */
2651 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002652 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002653 pvec.nr, pvec.pages, indices);
2654 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002655 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002656 index = end;
2657 break;
2658 }
2659 for (i = 0; i < pvec.nr; i++, index++) {
2660 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002661 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002662 done = true;
2663 break;
2664 }
2665 index = indices[i];
2666 }
2667 page = pvec.pages[i];
Matthew Wilcox3159f942017-11-03 13:30:42 -04002668 if (page && !xa_is_value(page)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002669 if (!PageUptodate(page))
2670 page = NULL;
2671 }
2672 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002673 (page && whence == SEEK_DATA) ||
2674 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002675 done = true;
2676 break;
2677 }
2678 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002679 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002680 pagevec_release(&pvec);
2681 pvec.nr = PAGEVEC_SIZE;
2682 cond_resched();
2683 }
2684 return index;
2685}
2686
Andrew Morton965c8e52012-12-17 15:59:39 -08002687static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002688{
2689 struct address_space *mapping = file->f_mapping;
2690 struct inode *inode = mapping->host;
2691 pgoff_t start, end;
2692 loff_t new_offset;
2693
Andrew Morton965c8e52012-12-17 15:59:39 -08002694 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2695 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002696 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002697 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002698 /* We're holding i_mutex so we can access i_size directly */
2699
Yufen Yu1a413642018-11-16 15:08:39 -08002700 if (offset < 0 || offset >= inode->i_size)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002701 offset = -ENXIO;
2702 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002703 start = offset >> PAGE_SHIFT;
2704 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002705 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002706 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002707 if (new_offset > offset) {
2708 if (new_offset < inode->i_size)
2709 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002710 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002711 offset = -ENXIO;
2712 else
2713 offset = inode->i_size;
2714 }
2715 }
2716
Hugh Dickins387aae62013-08-04 11:30:25 -07002717 if (offset >= 0)
2718 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002719 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002720 return offset;
2721}
2722
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002723static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2724 loff_t len)
2725{
Al Viro496ad9a2013-01-23 17:07:38 -05002726 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002727 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002728 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002729 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002730 pgoff_t start, index, end;
2731 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002732
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002733 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2734 return -EOPNOTSUPP;
2735
Al Viro59551022016-01-22 15:40:57 -05002736 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002737
2738 if (mode & FALLOC_FL_PUNCH_HOLE) {
2739 struct address_space *mapping = file->f_mapping;
2740 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2741 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002742 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002743
David Herrmann40e041a2014-08-08 14:25:27 -07002744 /* protected by i_mutex */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002745 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
David Herrmann40e041a2014-08-08 14:25:27 -07002746 error = -EPERM;
2747 goto out;
2748 }
2749
Hugh Dickins8e205f72014-07-23 14:00:10 -07002750 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junaa71ecd2019-11-30 17:58:11 -08002751 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002752 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2753 spin_lock(&inode->i_lock);
2754 inode->i_private = &shmem_falloc;
2755 spin_unlock(&inode->i_lock);
2756
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002757 if ((u64)unmap_end > (u64)unmap_start)
2758 unmap_mapping_range(mapping, unmap_start,
2759 1 + unmap_end - unmap_start, 0);
2760 shmem_truncate_range(inode, offset, offset + len - 1);
2761 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002762
2763 spin_lock(&inode->i_lock);
2764 inode->i_private = NULL;
2765 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02002766 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002767 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002768 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002769 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002770 }
2771
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002772 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2773 error = inode_newsize_ok(inode, offset + len);
2774 if (error)
2775 goto out;
2776
David Herrmann40e041a2014-08-08 14:25:27 -07002777 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2778 error = -EPERM;
2779 goto out;
2780 }
2781
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002782 start = offset >> PAGE_SHIFT;
2783 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002784 /* Try to avoid a swapstorm if len is impossible to satisfy */
2785 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2786 error = -ENOSPC;
2787 goto out;
2788 }
2789
Hugh Dickins8e205f72014-07-23 14:00:10 -07002790 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002791 shmem_falloc.start = start;
2792 shmem_falloc.next = start;
2793 shmem_falloc.nr_falloced = 0;
2794 shmem_falloc.nr_unswapped = 0;
2795 spin_lock(&inode->i_lock);
2796 inode->i_private = &shmem_falloc;
2797 spin_unlock(&inode->i_lock);
2798
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002799 for (index = start; index < end; index++) {
2800 struct page *page;
2801
2802 /*
2803 * Good, the fallocate(2) manpage permits EINTR: we may have
2804 * been interrupted because we are using up too much memory.
2805 */
2806 if (signal_pending(current))
2807 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002808 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2809 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002810 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002811 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002812 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002813 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002814 if (index > start) {
2815 shmem_undo_range(inode,
2816 (loff_t)start << PAGE_SHIFT,
2817 ((loff_t)index << PAGE_SHIFT) - 1, true);
2818 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002819 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002820 }
2821
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002822 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002823 * Inform shmem_writepage() how far we have reached.
2824 * No need for lock or barrier: we have the page lock.
2825 */
2826 shmem_falloc.next++;
2827 if (!PageUptodate(page))
2828 shmem_falloc.nr_falloced++;
2829
2830 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002831 * If !PageUptodate, leave it that way so that freeable pages
2832 * can be recognized if we need to rollback on error later.
2833 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002834 * than free the pages we are allocating (and SGP_CACHE pages
2835 * might still be clean: we now need to mark those dirty too).
2836 */
2837 set_page_dirty(page);
2838 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002839 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002840 cond_resched();
2841 }
2842
2843 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2844 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002845 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002846undone:
2847 spin_lock(&inode->i_lock);
2848 inode->i_private = NULL;
2849 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002850out:
Al Viro59551022016-01-22 15:40:57 -05002851 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002852 return error;
2853}
2854
David Howells726c3342006-06-23 02:02:58 -07002855static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856{
David Howells726c3342006-06-23 02:02:58 -07002857 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858
2859 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002860 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002862 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002864 buf->f_bavail =
2865 buf->f_bfree = sbinfo->max_blocks -
2866 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002867 }
2868 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 buf->f_files = sbinfo->max_inodes;
2870 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 }
2872 /* else leave those fields 0 like simple_statfs */
2873 return 0;
2874}
2875
2876/*
2877 * File creation. Allocate an inode, and we're done..
2878 */
2879static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002880shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002882 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 int error = -ENOSPC;
2884
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002885 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002887 error = simple_acl_create(dir, inode);
2888 if (error)
2889 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002890 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002891 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002892 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002893 if (error && error != -EOPNOTSUPP)
2894 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002895
Al Viro718deb62009-12-16 19:35:36 -05002896 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002898 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 d_instantiate(dentry, inode);
2900 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 }
2902 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002903out_iput:
2904 iput(inode);
2905 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906}
2907
Al Viro60545d02013-06-07 01:20:27 -04002908static int
2909shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2910{
2911 struct inode *inode;
2912 int error = -ENOSPC;
2913
2914 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2915 if (inode) {
2916 error = security_inode_init_security(inode, dir,
2917 NULL,
2918 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002919 if (error && error != -EOPNOTSUPP)
2920 goto out_iput;
2921 error = simple_acl_create(dir, inode);
2922 if (error)
2923 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002924 d_tmpfile(dentry, inode);
2925 }
2926 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002927out_iput:
2928 iput(inode);
2929 return error;
Al Viro60545d02013-06-07 01:20:27 -04002930}
2931
Al Viro18bb1db2011-07-26 01:41:39 -04002932static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933{
2934 int error;
2935
2936 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2937 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002938 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 return 0;
2940}
2941
Al Viro4acdaf22011-07-26 01:42:34 -04002942static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002943 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944{
2945 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2946}
2947
2948/*
2949 * Link a file..
2950 */
2951static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2952{
David Howells75c3cfa2015-03-17 22:26:12 +00002953 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong29b00e62019-02-22 22:35:32 -08002954 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955
2956 /*
2957 * No ordinary (disk based) filesystem counts links as inodes;
2958 * but each new link needs a new dentry, pinning lowmem, and
2959 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wong1062af92019-02-21 08:48:09 -08002960 * But if an O_TMPFILE file is linked into the tmpfs, the
2961 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 */
Darrick J. Wong1062af92019-02-21 08:48:09 -08002963 if (inode->i_nlink) {
2964 ret = shmem_reserve_inode(inode->i_sb);
2965 if (ret)
2966 goto out;
2967 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968
2969 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002970 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07002971 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002972 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 dget(dentry); /* Extra pinning count for the created dentry */
2974 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002975out:
2976 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977}
2978
2979static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2980{
David Howells75c3cfa2015-03-17 22:26:12 +00002981 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002983 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2984 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985
2986 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002987 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002988 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 dput(dentry); /* Undo the count from "create" - this does all the work */
2990 return 0;
2991}
2992
2993static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2994{
2995 if (!simple_empty(dentry))
2996 return -ENOTEMPTY;
2997
David Howells75c3cfa2015-03-17 22:26:12 +00002998 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002999 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 return shmem_unlink(dir, dentry);
3001}
3002
Miklos Szeredi37456772014-07-23 15:15:34 +02003003static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3004{
David Howellse36cb0b2015-01-29 12:02:35 +00003005 bool old_is_dir = d_is_dir(old_dentry);
3006 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003007
3008 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3009 if (old_is_dir) {
3010 drop_nlink(old_dir);
3011 inc_nlink(new_dir);
3012 } else {
3013 drop_nlink(new_dir);
3014 inc_nlink(old_dir);
3015 }
3016 }
3017 old_dir->i_ctime = old_dir->i_mtime =
3018 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003019 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003020 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02003021
3022 return 0;
3023}
3024
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003025static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3026{
3027 struct dentry *whiteout;
3028 int error;
3029
3030 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3031 if (!whiteout)
3032 return -ENOMEM;
3033
3034 error = shmem_mknod(old_dir, whiteout,
3035 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3036 dput(whiteout);
3037 if (error)
3038 return error;
3039
3040 /*
3041 * Cheat and hash the whiteout while the old dentry is still in
3042 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3043 *
3044 * d_lookup() will consistently find one of them at this point,
3045 * not sure which one, but that isn't even important.
3046 */
3047 d_rehash(whiteout);
3048 return 0;
3049}
3050
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051/*
3052 * The VFS layer already does all the dentry stuff for rename,
3053 * we just have to decrement the usage count for the target if
3054 * it exists so that the VFS layer correctly free's it when it
3055 * gets overwritten.
3056 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003057static 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 -07003058{
David Howells75c3cfa2015-03-17 22:26:12 +00003059 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 int they_are_dirs = S_ISDIR(inode->i_mode);
3061
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003062 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003063 return -EINVAL;
3064
Miklos Szeredi37456772014-07-23 15:15:34 +02003065 if (flags & RENAME_EXCHANGE)
3066 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3067
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 if (!simple_empty(new_dentry))
3069 return -ENOTEMPTY;
3070
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003071 if (flags & RENAME_WHITEOUT) {
3072 int error;
3073
3074 error = shmem_whiteout(old_dir, old_dentry);
3075 if (error)
3076 return error;
3077 }
3078
David Howells75c3cfa2015-03-17 22:26:12 +00003079 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003081 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003082 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003083 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003086 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003087 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 }
3089
3090 old_dir->i_size -= BOGO_DIRENT_SIZE;
3091 new_dir->i_size += BOGO_DIRENT_SIZE;
3092 old_dir->i_ctime = old_dir->i_mtime =
3093 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003094 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 return 0;
3096}
3097
3098static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3099{
3100 int error;
3101 int len;
3102 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003103 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104
3105 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003106 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 return -ENAMETOOLONG;
3108
Joe Perches0825a6f2018-06-14 15:27:58 -07003109 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3110 VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 if (!inode)
3112 return -ENOSPC;
3113
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003114 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003115 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003116 if (error) {
3117 if (error != -EOPNOTSUPP) {
3118 iput(inode);
3119 return error;
3120 }
3121 error = 0;
3122 }
3123
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003125 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003126 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3127 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003128 iput(inode);
3129 return -ENOMEM;
3130 }
3131 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003133 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003134 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 if (error) {
3136 iput(inode);
3137 return error;
3138 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003139 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003141 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003142 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003144 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003145 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003148 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 d_instantiate(dentry, inode);
3150 dget(dentry);
3151 return 0;
3152}
3153
Al Virofceef392015-12-29 15:58:39 -05003154static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155{
Al Virofceef392015-12-29 15:58:39 -05003156 mark_page_accessed(arg);
3157 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158}
3159
Al Viro6b255392015-11-17 10:20:54 -05003160static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003161 struct inode *inode,
3162 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003165 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003166 if (!dentry) {
3167 page = find_get_page(inode->i_mapping, 0);
3168 if (!page)
3169 return ERR_PTR(-ECHILD);
3170 if (!PageUptodate(page)) {
3171 put_page(page);
3172 return ERR_PTR(-ECHILD);
3173 }
3174 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003175 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003176 if (error)
3177 return ERR_PTR(error);
3178 unlock_page(page);
3179 }
Al Virofceef392015-12-29 15:58:39 -05003180 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003181 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182}
3183
Eric Parisb09e0fa2011-05-24 17:12:39 -07003184#ifdef CONFIG_TMPFS_XATTR
3185/*
3186 * Superblocks without xattr inode operations may get some security.* xattr
3187 * support from the LSM "for free". As soon as we have any other xattrs
3188 * like ACLs, we also need to implement the security.* handlers at
3189 * filesystem level, though.
3190 */
3191
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003192/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003193 * Callback for security_inode_init_security() for acquiring xattrs.
3194 */
3195static int shmem_initxattrs(struct inode *inode,
3196 const struct xattr *xattr_array,
3197 void *fs_info)
3198{
3199 struct shmem_inode_info *info = SHMEM_I(inode);
3200 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003201 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003202 size_t len;
3203
3204 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003205 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003206 if (!new_xattr)
3207 return -ENOMEM;
3208
3209 len = strlen(xattr->name) + 1;
3210 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3211 GFP_KERNEL);
3212 if (!new_xattr->name) {
3213 kfree(new_xattr);
3214 return -ENOMEM;
3215 }
3216
3217 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3218 XATTR_SECURITY_PREFIX_LEN);
3219 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3220 xattr->name, len);
3221
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003222 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003223 }
3224
3225 return 0;
3226}
3227
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003228static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003229 struct dentry *unused, struct inode *inode,
3230 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003231{
Al Virob2968212016-04-10 20:48:24 -04003232 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003233
3234 name = xattr_full_name(handler, name);
3235 return simple_xattr_get(&info->xattrs, name, buffer, size);
3236}
3237
3238static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003239 struct dentry *unused, struct inode *inode,
3240 const char *name, const void *value,
3241 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003242{
Al Viro59301222016-05-27 10:19:30 -04003243 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003244
3245 name = xattr_full_name(handler, name);
3246 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3247}
3248
3249static const struct xattr_handler shmem_security_xattr_handler = {
3250 .prefix = XATTR_SECURITY_PREFIX,
3251 .get = shmem_xattr_handler_get,
3252 .set = shmem_xattr_handler_set,
3253};
3254
3255static const struct xattr_handler shmem_trusted_xattr_handler = {
3256 .prefix = XATTR_TRUSTED_PREFIX,
3257 .get = shmem_xattr_handler_get,
3258 .set = shmem_xattr_handler_set,
3259};
3260
Eric Parisb09e0fa2011-05-24 17:12:39 -07003261static const struct xattr_handler *shmem_xattr_handlers[] = {
3262#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003263 &posix_acl_access_xattr_handler,
3264 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003265#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003266 &shmem_security_xattr_handler,
3267 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003268 NULL
3269};
3270
Eric Parisb09e0fa2011-05-24 17:12:39 -07003271static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3272{
David Howells75c3cfa2015-03-17 22:26:12 +00003273 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003274 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003275}
3276#endif /* CONFIG_TMPFS_XATTR */
3277
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003278static const struct inode_operations shmem_short_symlink_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003279 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003280#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003281 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003282#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283};
3284
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003285static const struct inode_operations shmem_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003286 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003287#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003288 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003289#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003290};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003291
David M. Grimes91828a42006-10-17 00:09:45 -07003292static struct dentry *shmem_get_parent(struct dentry *child)
3293{
3294 return ERR_PTR(-ESTALE);
3295}
3296
3297static int shmem_match(struct inode *ino, void *vfh)
3298{
3299 __u32 *fh = vfh;
3300 __u64 inum = fh[2];
3301 inum = (inum << 32) | fh[1];
3302 return ino->i_ino == inum && fh[0] == ino->i_generation;
3303}
3304
Amir Goldstein12ba7802018-06-07 17:07:15 -07003305/* Find any alias of inode, but prefer a hashed alias */
3306static struct dentry *shmem_find_alias(struct inode *inode)
3307{
3308 struct dentry *alias = d_find_alias(inode);
3309
3310 return alias ?: d_find_any_alias(inode);
3311}
3312
3313
Christoph Hellwig480b1162007-10-21 16:42:13 -07003314static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3315 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003316{
David M. Grimes91828a42006-10-17 00:09:45 -07003317 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003318 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003319 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003320
Christoph Hellwig480b1162007-10-21 16:42:13 -07003321 if (fh_len < 3)
3322 return NULL;
3323
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003324 inum = fid->raw[2];
3325 inum = (inum << 32) | fid->raw[1];
3326
Christoph Hellwig480b1162007-10-21 16:42:13 -07003327 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3328 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003329 if (inode) {
Amir Goldstein12ba7802018-06-07 17:07:15 -07003330 dentry = shmem_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003331 iput(inode);
3332 }
3333
Christoph Hellwig480b1162007-10-21 16:42:13 -07003334 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003335}
3336
Al Virob0b03822012-04-02 14:34:06 -04003337static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3338 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003339{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303340 if (*len < 3) {
3341 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003342 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303343 }
David M. Grimes91828a42006-10-17 00:09:45 -07003344
Al Viro1d3382cb2010-10-23 15:19:20 -04003345 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003346 /* Unfortunately insert_inode_hash is not idempotent,
3347 * so as we hash inodes here rather than at creation
3348 * time, we need a lock to ensure we only try
3349 * to do it once
3350 */
3351 static DEFINE_SPINLOCK(lock);
3352 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003353 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003354 __insert_inode_hash(inode,
3355 inode->i_ino + inode->i_generation);
3356 spin_unlock(&lock);
3357 }
3358
3359 fh[0] = inode->i_generation;
3360 fh[1] = inode->i_ino;
3361 fh[2] = ((__u64)inode->i_ino) >> 32;
3362
3363 *len = 3;
3364 return 1;
3365}
3366
Christoph Hellwig39655162007-10-21 16:42:17 -07003367static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003368 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003369 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003370 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003371};
3372
Al Viro626c3922019-09-08 20:28:06 -04003373enum shmem_param {
3374 Opt_gid,
3375 Opt_huge,
3376 Opt_mode,
3377 Opt_mpol,
3378 Opt_nr_blocks,
3379 Opt_nr_inodes,
3380 Opt_size,
3381 Opt_uid,
3382};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
Al Viro5eede622019-12-16 13:33:32 -05003384static const struct constant_table shmem_param_enums_huge[] = {
Al Viro2710c957a2019-09-06 22:12:08 -04003385 {"never", SHMEM_HUGE_NEVER },
3386 {"always", SHMEM_HUGE_ALWAYS },
3387 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3388 {"advise", SHMEM_HUGE_ADVISE },
Al Viro2710c957a2019-09-06 22:12:08 -04003389 {}
3390};
3391
Al Virod7167b12019-09-07 07:23:15 -04003392const struct fs_parameter_spec shmem_fs_parameters[] = {
Al Viro626c3922019-09-08 20:28:06 -04003393 fsparam_u32 ("gid", Opt_gid),
Al Viro2710c957a2019-09-06 22:12:08 -04003394 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
Al Viro626c3922019-09-08 20:28:06 -04003395 fsparam_u32oct("mode", Opt_mode),
3396 fsparam_string("mpol", Opt_mpol),
3397 fsparam_string("nr_blocks", Opt_nr_blocks),
3398 fsparam_string("nr_inodes", Opt_nr_inodes),
3399 fsparam_string("size", Opt_size),
3400 fsparam_u32 ("uid", Opt_uid),
3401 {}
3402};
3403
David Howellsf3235622019-03-25 16:38:31 +00003404static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
Al Viro626c3922019-09-08 20:28:06 -04003405{
David Howellsf3235622019-03-25 16:38:31 +00003406 struct shmem_options *ctx = fc->fs_private;
Al Viro626c3922019-09-08 20:28:06 -04003407 struct fs_parse_result result;
3408 unsigned long long size;
3409 char *rest;
3410 int opt;
3411
Al Virod7167b12019-09-07 07:23:15 -04003412 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
David Howellsf3235622019-03-25 16:38:31 +00003413 if (opt < 0)
Al Viro626c3922019-09-08 20:28:06 -04003414 return opt;
Al Viro626c3922019-09-08 20:28:06 -04003415
3416 switch (opt) {
3417 case Opt_size:
3418 size = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003419 if (*rest == '%') {
3420 size <<= PAGE_SHIFT;
3421 size *= totalram_pages();
3422 do_div(size, 100);
3423 rest++;
3424 }
3425 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003426 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003427 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3428 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003429 break;
3430 case Opt_nr_blocks:
3431 ctx->blocks = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003432 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003433 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003434 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003435 break;
3436 case Opt_nr_inodes:
3437 ctx->inodes = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003438 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003439 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003440 ctx->seen |= SHMEM_SEEN_INODES;
Al Viro626c3922019-09-08 20:28:06 -04003441 break;
3442 case Opt_mode:
3443 ctx->mode = result.uint_32 & 07777;
3444 break;
3445 case Opt_uid:
3446 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003447 if (!uid_valid(ctx->uid))
Al Viro626c3922019-09-08 20:28:06 -04003448 goto bad_value;
3449 break;
3450 case Opt_gid:
3451 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003452 if (!gid_valid(ctx->gid))
Al Viro626c3922019-09-08 20:28:06 -04003453 goto bad_value;
3454 break;
3455 case Opt_huge:
3456 ctx->huge = result.uint_32;
3457 if (ctx->huge != SHMEM_HUGE_NEVER &&
3458 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
3459 has_transparent_hugepage()))
3460 goto unsupported_parameter;
Al Viroe04dc422019-09-08 19:20:12 -04003461 ctx->seen |= SHMEM_SEEN_HUGE;
Al Viro626c3922019-09-08 20:28:06 -04003462 break;
3463 case Opt_mpol:
3464 if (IS_ENABLED(CONFIG_NUMA)) {
3465 mpol_put(ctx->mpol);
3466 ctx->mpol = NULL;
3467 if (mpol_parse_str(param->string, &ctx->mpol))
3468 goto bad_value;
3469 break;
3470 }
3471 goto unsupported_parameter;
Al Viroe04dc422019-09-08 19:20:12 -04003472 }
3473 return 0;
3474
Al Viro626c3922019-09-08 20:28:06 -04003475unsupported_parameter:
Al Virof35aa2b2019-12-21 21:35:55 -05003476 return invalfc(fc, "Unsupported parameter '%s'", param->key);
Al Viro626c3922019-09-08 20:28:06 -04003477bad_value:
Al Virof35aa2b2019-12-21 21:35:55 -05003478 return invalfc(fc, "Bad value for '%s'", param->key);
Al Viroe04dc422019-09-08 19:20:12 -04003479}
3480
David Howellsf3235622019-03-25 16:38:31 +00003481static int shmem_parse_options(struct fs_context *fc, void *data)
Al Viroe04dc422019-09-08 19:20:12 -04003482{
David Howellsf3235622019-03-25 16:38:31 +00003483 char *options = data;
3484
Al Viro33f37c62019-10-09 22:48:01 -04003485 if (options) {
3486 int err = security_sb_eat_lsm_opts(options, &fc->security);
3487 if (err)
3488 return err;
3489 }
3490
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003491 while (options != NULL) {
Al Viro626c3922019-09-08 20:28:06 -04003492 char *this_char = options;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003493 for (;;) {
3494 /*
3495 * NUL-terminate this option: unfortunately,
3496 * mount options form a comma-separated list,
3497 * but mpol's nodelist may also contain commas.
3498 */
3499 options = strchr(options, ',');
3500 if (options == NULL)
3501 break;
3502 options++;
3503 if (!isdigit(*options)) {
3504 options[-1] = '\0';
3505 break;
3506 }
3507 }
Al Viro626c3922019-09-08 20:28:06 -04003508 if (*this_char) {
3509 char *value = strchr(this_char,'=');
David Howellsf3235622019-03-25 16:38:31 +00003510 size_t len = 0;
Al Viro626c3922019-09-08 20:28:06 -04003511 int err;
3512
3513 if (value) {
3514 *value++ = '\0';
David Howellsf3235622019-03-25 16:38:31 +00003515 len = strlen(value);
Al Viro626c3922019-09-08 20:28:06 -04003516 }
David Howellsf3235622019-03-25 16:38:31 +00003517 err = vfs_parse_fs_string(fc, this_char, value, len);
3518 if (err < 0)
3519 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521 }
3522 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523}
3524
David Howellsf3235622019-03-25 16:38:31 +00003525/*
3526 * Reconfigure a shmem filesystem.
3527 *
3528 * Note that we disallow change from limited->unlimited blocks/inodes while any
3529 * are in use; but we must separately disallow unlimited->limited, because in
3530 * that case we have no record of how much is already in use.
3531 */
3532static int shmem_reconfigure(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533{
David Howellsf3235622019-03-25 16:38:31 +00003534 struct shmem_options *ctx = fc->fs_private;
3535 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003536 unsigned long inodes;
David Howellsf3235622019-03-25 16:38:31 +00003537 const char *err;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003538
3539 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003540 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
David Howellsf3235622019-03-25 16:38:31 +00003541 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3542 if (!sbinfo->max_blocks) {
3543 err = "Cannot retroactively limit size";
Al Viro0b5071d2019-09-08 18:49:18 -04003544 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003545 }
Al Viro0b5071d2019-09-08 18:49:18 -04003546 if (percpu_counter_compare(&sbinfo->used_blocks,
David Howellsf3235622019-03-25 16:38:31 +00003547 ctx->blocks) > 0) {
3548 err = "Too small a size for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003549 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003550 }
Al Viro0b5071d2019-09-08 18:49:18 -04003551 }
David Howellsf3235622019-03-25 16:38:31 +00003552 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3553 if (!sbinfo->max_inodes) {
3554 err = "Cannot retroactively limit inodes";
Al Viro0b5071d2019-09-08 18:49:18 -04003555 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003556 }
3557 if (ctx->inodes < inodes) {
3558 err = "Too few inodes for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003559 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003560 }
Al Viro0b5071d2019-09-08 18:49:18 -04003561 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003562
David Howellsf3235622019-03-25 16:38:31 +00003563 if (ctx->seen & SHMEM_SEEN_HUGE)
3564 sbinfo->huge = ctx->huge;
3565 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3566 sbinfo->max_blocks = ctx->blocks;
3567 if (ctx->seen & SHMEM_SEEN_INODES) {
3568 sbinfo->max_inodes = ctx->inodes;
3569 sbinfo->free_inodes = ctx->inodes - inodes;
Al Viro0b5071d2019-09-08 18:49:18 -04003570 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003571
Greg Thelen5f001102013-02-22 16:36:01 -08003572 /*
3573 * Preserve previous mempolicy unless mpol remount option was specified.
3574 */
David Howellsf3235622019-03-25 16:38:31 +00003575 if (ctx->mpol) {
Greg Thelen5f001102013-02-22 16:36:01 -08003576 mpol_put(sbinfo->mpol);
David Howellsf3235622019-03-25 16:38:31 +00003577 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3578 ctx->mpol = NULL;
Greg Thelen5f001102013-02-22 16:36:01 -08003579 }
David Howellsf3235622019-03-25 16:38:31 +00003580 spin_unlock(&sbinfo->stat_lock);
3581 return 0;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003582out:
3583 spin_unlock(&sbinfo->stat_lock);
Al Virof35aa2b2019-12-21 21:35:55 -05003584 return invalfc(fc, "%s", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003586
Al Viro34c80b12011-12-08 21:32:45 -05003587static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003588{
Al Viro34c80b12011-12-08 21:32:45 -05003589 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003590
3591 if (sbinfo->max_blocks != shmem_default_max_blocks())
3592 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003593 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003594 if (sbinfo->max_inodes != shmem_default_max_inodes())
3595 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
Joe Perches0825a6f2018-06-14 15:27:58 -07003596 if (sbinfo->mode != (0777 | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003597 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003598 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3599 seq_printf(seq, ",uid=%u",
3600 from_kuid_munged(&init_user_ns, sbinfo->uid));
3601 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3602 seq_printf(seq, ",gid=%u",
3603 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003604#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003605 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3606 if (sbinfo->huge)
3607 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3608#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003609 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003610 return 0;
3611}
David Herrmann9183df22014-08-08 14:25:29 -07003612
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003613#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614
3615static void shmem_put_super(struct super_block *sb)
3616{
Hugh Dickins602586a2010-08-17 15:23:56 -07003617 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3618
3619 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003620 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003621 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 sb->s_fs_info = NULL;
3623}
3624
David Howellsf3235622019-03-25 16:38:31 +00003625static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626{
David Howellsf3235622019-03-25 16:38:31 +00003627 struct shmem_options *ctx = fc->fs_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003629 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003630 int err = -ENOMEM;
3631
3632 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003633 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003634 L1_CACHE_BYTES), GFP_KERNEL);
3635 if (!sbinfo)
3636 return -ENOMEM;
3637
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003638 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003640#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 /*
3642 * Per default we only allow half of the physical ram per
3643 * tmpfs instance, limiting inodes to one per page of lowmem;
3644 * but the internal instance is left unlimited.
3645 */
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003646 if (!(sb->s_flags & SB_KERNMOUNT)) {
David Howellsf3235622019-03-25 16:38:31 +00003647 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3648 ctx->blocks = shmem_default_max_blocks();
3649 if (!(ctx->seen & SHMEM_SEEN_INODES))
3650 ctx->inodes = shmem_default_max_inodes();
Al Viroca4e0512013-08-31 12:57:10 -04003651 } else {
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003652 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653 }
David M. Grimes91828a42006-10-17 00:09:45 -07003654 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003655 sb->s_flags |= SB_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656#else
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003657 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658#endif
David Howellsf3235622019-03-25 16:38:31 +00003659 sbinfo->max_blocks = ctx->blocks;
3660 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3661 sbinfo->uid = ctx->uid;
3662 sbinfo->gid = ctx->gid;
3663 sbinfo->mode = ctx->mode;
3664 sbinfo->huge = ctx->huge;
3665 sbinfo->mpol = ctx->mpol;
3666 ctx->mpol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003668 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003669 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003670 goto failed;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003671 spin_lock_init(&sbinfo->shrinklist_lock);
3672 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003673
Hugh Dickins285b2c42011-08-03 16:21:20 -07003674 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003675 sb->s_blocksize = PAGE_SIZE;
3676 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677 sb->s_magic = TMPFS_MAGIC;
3678 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003679 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003680#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003681 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003682#endif
3683#ifdef CONFIG_TMPFS_POSIX_ACL
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003684 sb->s_flags |= SB_POSIXACL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003685#endif
Amir Goldstein2b4db792017-05-18 15:29:33 +03003686 uuid_gen(&sb->s_uuid);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003687
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003688 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 if (!inode)
3690 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003691 inode->i_uid = sbinfo->uid;
3692 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003693 sb->s_root = d_make_root(inode);
3694 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003695 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 return 0;
3697
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698failed:
3699 shmem_put_super(sb);
3700 return err;
3701}
3702
David Howellsf3235622019-03-25 16:38:31 +00003703static int shmem_get_tree(struct fs_context *fc)
3704{
3705 return get_tree_nodev(fc, shmem_fill_super);
3706}
3707
3708static void shmem_free_fc(struct fs_context *fc)
3709{
3710 struct shmem_options *ctx = fc->fs_private;
3711
3712 if (ctx) {
3713 mpol_put(ctx->mpol);
3714 kfree(ctx);
3715 }
3716}
3717
3718static const struct fs_context_operations shmem_fs_context_ops = {
3719 .free = shmem_free_fc,
3720 .get_tree = shmem_get_tree,
3721#ifdef CONFIG_TMPFS
3722 .parse_monolithic = shmem_parse_options,
3723 .parse_param = shmem_parse_one,
3724 .reconfigure = shmem_reconfigure,
3725#endif
3726};
3727
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003728static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729
3730static struct inode *shmem_alloc_inode(struct super_block *sb)
3731{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003732 struct shmem_inode_info *info;
3733 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3734 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003736 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737}
3738
Al Viro74b1da52019-04-15 23:19:05 -04003739static void shmem_free_in_core_inode(struct inode *inode)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003740{
Al Viro84e710d2016-04-15 00:58:55 -04003741 if (S_ISLNK(inode->i_mode))
3742 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003743 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3744}
3745
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746static void shmem_destroy_inode(struct inode *inode)
3747{
Al Viro09208d12011-07-26 03:15:03 -04003748 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750}
3751
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003752static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003754 struct shmem_inode_info *info = foo;
3755 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756}
3757
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003758static void shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759{
3760 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3761 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003762 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763}
3764
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003765static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003767 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768}
3769
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003770static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003772 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003774 .write_begin = shmem_write_begin,
3775 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003777#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003778 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003779#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003780 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781};
3782
Helge Deller15ad7cd2006-12-06 20:40:36 -08003783static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003785 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003787 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003788 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003789 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003790 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003791 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003792 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003793 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794#endif
3795};
3796
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003797static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003798 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003799 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003800#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003801 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003802 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003803#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804};
3805
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003806static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807#ifdef CONFIG_TMPFS
3808 .create = shmem_create,
3809 .lookup = simple_lookup,
3810 .link = shmem_link,
3811 .unlink = shmem_unlink,
3812 .symlink = shmem_symlink,
3813 .mkdir = shmem_mkdir,
3814 .rmdir = shmem_rmdir,
3815 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003816 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003817 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003819#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003820 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003821#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003822#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003823 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003824 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003825#endif
3826};
3827
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003828static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003829#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003830 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003831#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003832#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003833 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003834 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003835#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836};
3837
Hugh Dickins759b9772007-03-05 00:30:28 -08003838static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 .alloc_inode = shmem_alloc_inode,
Al Viro74b1da52019-04-15 23:19:05 -04003840 .free_inode = shmem_free_in_core_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 .destroy_inode = shmem_destroy_inode,
3842#ifdef CONFIG_TMPFS
3843 .statfs = shmem_statfs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003844 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845#endif
Al Viro1f895f72010-06-05 19:10:41 -04003846 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847 .drop_inode = generic_delete_inode,
3848 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003849#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3850 .nr_cached_objects = shmem_unused_huge_count,
3851 .free_cached_objects = shmem_unused_huge_scan,
3852#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853};
3854
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003855static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003856 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003857 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858#ifdef CONFIG_NUMA
3859 .set_policy = shmem_set_policy,
3860 .get_policy = shmem_get_policy,
3861#endif
3862};
3863
David Howellsf3235622019-03-25 16:38:31 +00003864int shmem_init_fs_context(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865{
David Howellsf3235622019-03-25 16:38:31 +00003866 struct shmem_options *ctx;
3867
3868 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3869 if (!ctx)
3870 return -ENOMEM;
3871
3872 ctx->mode = 0777 | S_ISVTX;
3873 ctx->uid = current_fsuid();
3874 ctx->gid = current_fsgid();
3875
3876 fc->fs_private = ctx;
3877 fc->ops = &shmem_fs_context_ops;
3878 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879}
3880
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003881static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 .owner = THIS_MODULE,
3883 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00003884 .init_fs_context = shmem_init_fs_context,
3885#ifdef CONFIG_TMPFS
Al Virod7167b12019-09-07 07:23:15 -04003886 .parameters = shmem_fs_parameters,
David Howellsf3235622019-03-25 16:38:31 +00003887#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003889 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003892int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893{
3894 int error;
3895
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003896 shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003898 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003900 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 goto out2;
3902 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003903
Al Viroca4e0512013-08-31 12:57:10 -04003904 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 if (IS_ERR(shm_mnt)) {
3906 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003907 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908 goto out1;
3909 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003910
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003911#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07003912 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003913 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3914 else
3915 shmem_huge = 0; /* just in case it was patched */
3916#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917 return 0;
3918
3919out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003920 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003922 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923 shm_mnt = ERR_PTR(error);
3924 return error;
3925}
Matt Mackall853ac432009-01-06 14:40:20 -08003926
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003927#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003928static ssize_t shmem_enabled_show(struct kobject *kobj,
3929 struct kobj_attribute *attr, char *buf)
3930{
Colin Ian King26083eb62019-11-30 17:58:04 -08003931 static const int values[] = {
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003932 SHMEM_HUGE_ALWAYS,
3933 SHMEM_HUGE_WITHIN_SIZE,
3934 SHMEM_HUGE_ADVISE,
3935 SHMEM_HUGE_NEVER,
3936 SHMEM_HUGE_DENY,
3937 SHMEM_HUGE_FORCE,
3938 };
3939 int i, count;
3940
3941 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3942 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3943
3944 count += sprintf(buf + count, fmt,
3945 shmem_format_huge(values[i]));
3946 }
3947 buf[count - 1] = '\n';
3948 return count;
3949}
3950
3951static ssize_t shmem_enabled_store(struct kobject *kobj,
3952 struct kobj_attribute *attr, const char *buf, size_t count)
3953{
3954 char tmp[16];
3955 int huge;
3956
3957 if (count + 1 > sizeof(tmp))
3958 return -EINVAL;
3959 memcpy(tmp, buf, count);
3960 tmp[count] = '\0';
3961 if (count && tmp[count - 1] == '\n')
3962 tmp[count - 1] = '\0';
3963
3964 huge = shmem_parse_huge(tmp);
3965 if (huge == -EINVAL)
3966 return -EINVAL;
3967 if (!has_transparent_hugepage() &&
3968 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3969 return -EINVAL;
3970
3971 shmem_huge = huge;
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07003972 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003973 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3974 return count;
3975}
3976
3977struct kobj_attribute shmem_enabled_attr =
3978 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07003979#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003980
Arnd Bergmann3b337192016-08-10 16:27:44 -07003981#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003982bool shmem_huge_enabled(struct vm_area_struct *vma)
3983{
3984 struct inode *inode = file_inode(vma->vm_file);
3985 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3986 loff_t i_size;
3987 pgoff_t off;
3988
Yang Shic0630662019-07-18 15:57:27 -07003989 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
3990 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
3991 return false;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07003992 if (shmem_huge == SHMEM_HUGE_FORCE)
3993 return true;
3994 if (shmem_huge == SHMEM_HUGE_DENY)
3995 return false;
3996 switch (sbinfo->huge) {
3997 case SHMEM_HUGE_NEVER:
3998 return false;
3999 case SHMEM_HUGE_ALWAYS:
4000 return true;
4001 case SHMEM_HUGE_WITHIN_SIZE:
4002 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4003 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4004 if (i_size >= HPAGE_PMD_SIZE &&
4005 i_size >> PAGE_SHIFT >= off)
4006 return true;
Gustavo A. R. Silvac8402872017-11-15 17:38:45 -08004007 /* fall through */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004008 case SHMEM_HUGE_ADVISE:
4009 /* TODO: implement fadvise() hints */
4010 return (vma->vm_flags & VM_HUGEPAGE);
4011 default:
4012 VM_BUG_ON(1);
4013 return false;
4014 }
4015}
Arnd Bergmann3b337192016-08-10 16:27:44 -07004016#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004017
Matt Mackall853ac432009-01-06 14:40:20 -08004018#else /* !CONFIG_SHMEM */
4019
4020/*
4021 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4022 *
4023 * This is intended for small system where the benefits of the full
4024 * shmem code (swap-backed and resource-limited) are outweighed by
4025 * their complexity. On systems without swap this code should be
4026 * effectively equivalent, but much lighter weight.
4027 */
4028
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004029static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004030 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00004031 .init_fs_context = ramfs_init_fs_context,
Al Virod7167b12019-09-07 07:23:15 -04004032 .parameters = ramfs_fs_parameters,
Matt Mackall853ac432009-01-06 14:40:20 -08004033 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004034 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004035};
4036
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004037int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004038{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004039 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004040
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004041 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004042 BUG_ON(IS_ERR(shm_mnt));
4043
4044 return 0;
4045}
4046
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08004047int shmem_unuse(unsigned int type, bool frontswap,
4048 unsigned long *fs_pages_to_unuse)
Matt Mackall853ac432009-01-06 14:40:20 -08004049{
4050 return 0;
4051}
4052
Hugh Dickins3f96b792009-09-21 17:03:37 -07004053int shmem_lock(struct file *file, int lock, struct user_struct *user)
4054{
4055 return 0;
4056}
4057
Hugh Dickins24513262012-01-20 14:34:21 -08004058void shmem_unlock_mapping(struct address_space *mapping)
4059{
4060}
4061
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004062#ifdef CONFIG_MMU
4063unsigned long shmem_get_unmapped_area(struct file *file,
4064 unsigned long addr, unsigned long len,
4065 unsigned long pgoff, unsigned long flags)
4066{
4067 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4068}
4069#endif
4070
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004071void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004072{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004073 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004074}
4075EXPORT_SYMBOL_GPL(shmem_truncate_range);
4076
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004077#define shmem_vm_ops generic_file_vm_ops
4078#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004079#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004080#define shmem_acct_size(flags, size) 0
4081#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004082
4083#endif /* CONFIG_SHMEM */
4084
4085/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086
Matthew Auld703321b2017-10-06 23:18:13 +01004087static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
Eric Parisc7277092013-12-02 11:24:19 +00004088 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 struct inode *inode;
Al Viro93dec2d2018-07-08 23:02:03 -04004091 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092
Matthew Auld703321b2017-10-06 23:18:13 +01004093 if (IS_ERR(mnt))
4094 return ERR_CAST(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095
Hugh Dickins285b2c42011-08-03 16:21:20 -07004096 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097 return ERR_PTR(-EINVAL);
4098
4099 if (shmem_acct_size(flags, size))
4100 return ERR_PTR(-ENOMEM);
4101
Al Viro93dec2d2018-07-08 23:02:03 -04004102 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4103 flags);
Al Virodac2d1f2018-06-09 09:27:41 -04004104 if (unlikely(!inode)) {
4105 shmem_unacct_size(flags, size);
4106 return ERR_PTR(-ENOSPC);
4107 }
Eric Parisc7277092013-12-02 11:24:19 +00004108 inode->i_flags |= i_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004110 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004111 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
Al Viro93dec2d2018-07-08 23:02:03 -04004112 if (!IS_ERR(res))
4113 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4114 &shmem_file_operations);
Al Viro26567cd2013-03-01 20:22:53 -05004115 if (IS_ERR(res))
Al Viro93dec2d2018-07-08 23:02:03 -04004116 iput(inode);
Al Viro6b4d0b22013-02-14 21:37:26 -05004117 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118}
Eric Parisc7277092013-12-02 11:24:19 +00004119
4120/**
4121 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4122 * kernel internal. There will be NO LSM permission checks against the
4123 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004124 * higher layer. The users are the big_key and shm implementations. LSM
4125 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004126 * @name: name for dentry (to be seen in /proc/<pid>/maps
4127 * @size: size to be set for the file
4128 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4129 */
4130struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4131{
Matthew Auld703321b2017-10-06 23:18:13 +01004132 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
Eric Parisc7277092013-12-02 11:24:19 +00004133}
4134
4135/**
4136 * shmem_file_setup - get an unlinked file living in tmpfs
4137 * @name: name for dentry (to be seen in /proc/<pid>/maps
4138 * @size: size to be set for the file
4139 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4140 */
4141struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4142{
Matthew Auld703321b2017-10-06 23:18:13 +01004143 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
Eric Parisc7277092013-12-02 11:24:19 +00004144}
Keith Packard395e0dd2008-06-20 00:08:06 -07004145EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146
Randy Dunlap46711812008-03-19 17:00:41 -07004147/**
Matthew Auld703321b2017-10-06 23:18:13 +01004148 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4149 * @mnt: the tmpfs mount where the file will be created
4150 * @name: name for dentry (to be seen in /proc/<pid>/maps
4151 * @size: size to be set for the file
4152 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4153 */
4154struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4155 loff_t size, unsigned long flags)
4156{
4157 return __shmem_file_setup(mnt, name, size, flags, 0);
4158}
4159EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4160
4161/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4164 */
4165int shmem_zero_setup(struct vm_area_struct *vma)
4166{
4167 struct file *file;
4168 loff_t size = vma->vm_end - vma->vm_start;
4169
Hugh Dickins66fc1302015-06-14 09:48:09 -07004170 /*
4171 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4172 * between XFS directory reading and selinux: since this file is only
4173 * accessible to the user through its mapping, use S_PRIVATE flag to
4174 * bypass file security, in the same way as shmem_kernel_file_setup().
4175 */
Matthew Auld703321b2017-10-06 23:18:13 +01004176 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177 if (IS_ERR(file))
4178 return PTR_ERR(file);
4179
4180 if (vma->vm_file)
4181 fput(vma->vm_file);
4182 vma->vm_file = file;
4183 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004184
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004185 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004186 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4187 (vma->vm_end & HPAGE_PMD_MASK)) {
4188 khugepaged_enter(vma, vma->vm_flags);
4189 }
4190
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191 return 0;
4192}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004193
4194/**
4195 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4196 * @mapping: the page's address_space
4197 * @index: the page index
4198 * @gfp: the page allocator flags to use if allocating
4199 *
4200 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4201 * with any new page allocations done using the specified allocation flags.
4202 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4203 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4204 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4205 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004206 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4207 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004208 */
4209struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4210 pgoff_t index, gfp_t gfp)
4211{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004212#ifdef CONFIG_SHMEM
4213 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004214 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004215 int error;
4216
4217 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004218 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
Mike Rapoportcfda0522017-02-22 15:43:37 -08004219 gfp, NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004220 if (error)
4221 page = ERR_PTR(error);
4222 else
4223 unlock_page(page);
4224 return page;
4225#else
4226 /*
4227 * The tiny !SHMEM case uses ramfs without swap
4228 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004229 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004230#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004231}
4232EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);