blob: 271548ca20f3149b90974224fd767cbec5e42cb2 [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
Mel Gormandd56b042015-11-06 16:28:43 -080086#include "internal.h"
87
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030088#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/* Pretend that each entry is of this size in directory's i_size */
92#define BOGO_DIRENT_SIZE 20
93
Hugh Dickins69f07ec2011-08-03 16:21:26 -070094/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95#define SHORT_SYMLINK_LEN 128
96
Hugh Dickins1aac1402012-05-29 15:06:42 -070097/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070098 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700101 */
102struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -0700103 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -0700104 pgoff_t start; /* start of range currently being fallocated */
105 pgoff_t next; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
108};
109
Al Viro0b5071d2019-09-08 18:49:18 -0400110struct shmem_options {
111 unsigned long long blocks;
112 unsigned long long inodes;
113 struct mempolicy *mpol;
114 kuid_t uid;
115 kgid_t gid;
116 umode_t mode;
Chris Downea3271f2020-08-06 23:20:25 -0700117 bool full_inums;
Al Viro0b5071d2019-09-08 18:49:18 -0400118 int huge;
119 int seen;
120#define SHMEM_SEEN_BLOCKS 1
121#define SHMEM_SEEN_INODES 2
122#define SHMEM_SEEN_HUGE 4
Chris Downea3271f2020-08-06 23:20:25 -0700123#define SHMEM_SEEN_INUMS 8
Al Viro0b5071d2019-09-08 18:49:18 -0400124};
125
Andrew Mortonb76db7352008-02-08 04:21:49 -0800126#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800127static unsigned long shmem_default_max_blocks(void)
128{
Arun KSca79b0c2018-12-28 00:34:29 -0800129 return totalram_pages() / 2;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800130}
131
132static unsigned long shmem_default_max_inodes(void)
133{
Arun KSca79b0c2018-12-28 00:34:29 -0800134 unsigned long nr_pages = totalram_pages();
135
136 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800137}
Andrew Mortonb76db7352008-02-08 04:21:49 -0800138#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800139
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700140static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
141static int shmem_replace_page(struct page **pagep, gfp_t gfp,
142 struct shmem_inode_info *info, pgoff_t index);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -0800143static int shmem_swapin_page(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 vm_fault_t *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700147static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700148 struct page **pagep, enum sgp_type sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800149 gfp_t gfp, struct vm_area_struct *vma,
Souptick Joarder2b740302018-08-23 17:01:36 -0700150 struct vm_fault *vmf, vm_fault_t *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700151
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700152int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700153 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700154{
155 return shmem_getpage_gfp(inode, index, pagep, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800156 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700157}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
160{
161 return sb->s_fs_info;
162}
163
164/*
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
169 */
170static inline int shmem_acct_size(unsigned long flags, loff_t size)
171{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000172 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000173 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174}
175
176static inline void shmem_unacct_size(unsigned long flags, loff_t size)
177{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000178 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 vm_unacct_memory(VM_ACCT(size));
180}
181
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700182static inline int shmem_reacct_size(unsigned long flags,
183 loff_t oldsize, loff_t newsize)
184{
185 if (!(flags & VM_NORESERVE)) {
186 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
187 return security_vm_enough_memory_mm(current->mm,
188 VM_ACCT(newsize) - VM_ACCT(oldsize));
189 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
190 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
191 }
192 return 0;
193}
194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195/*
196 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700197 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
200 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700201static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700203 if (!(flags & VM_NORESERVE))
204 return 0;
205
206 return security_vm_enough_memory_mm(current->mm,
207 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210static inline void shmem_unacct_blocks(unsigned long flags, long pages)
211{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000212 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300213 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Mike Rapoport0f079692017-09-06 16:22:59 -0700216static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
217{
218 struct shmem_inode_info *info = SHMEM_I(inode);
219 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
220
221 if (shmem_acct_block(info->flags, pages))
222 return false;
223
224 if (sbinfo->max_blocks) {
225 if (percpu_counter_compare(&sbinfo->used_blocks,
226 sbinfo->max_blocks - pages) > 0)
227 goto unacct;
228 percpu_counter_add(&sbinfo->used_blocks, pages);
229 }
230
231 return true;
232
233unacct:
234 shmem_unacct_blocks(info->flags, pages);
235 return false;
236}
237
238static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
239{
240 struct shmem_inode_info *info = SHMEM_I(inode);
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
242
243 if (sbinfo->max_blocks)
244 percpu_counter_sub(&sbinfo->used_blocks, pages);
245 shmem_unacct_blocks(info->flags, pages);
246}
247
Hugh Dickins759b9772007-03-05 00:30:28 -0800248static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700249static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800250static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800251static const struct inode_operations shmem_inode_operations;
252static const struct inode_operations shmem_dir_inode_operations;
253static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +0400254static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700255static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Mike Rapoportb0506e42017-02-22 15:43:28 -0800257bool vma_is_shmem(struct vm_area_struct *vma)
258{
259 return vma->vm_ops == &shmem_vm_ops;
260}
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800263static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Chris Downe809d5f2020-08-06 23:20:20 -0700265/*
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
268 *
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
273 */
274#define SHMEM_INO_BATCH 1024
275static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800276{
277 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -0700278 ino_t ino;
279
280 if (!(sb->s_flags & SB_KERNMOUNT)) {
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800281 spin_lock(&sbinfo->stat_lock);
282 if (!sbinfo->free_inodes) {
283 spin_unlock(&sbinfo->stat_lock);
284 return -ENOSPC;
285 }
286 sbinfo->free_inodes--;
Chris Downe809d5f2020-08-06 23:20:20 -0700287 if (inop) {
288 ino = sbinfo->next_ino++;
289 if (unlikely(is_zero_ino(ino)))
290 ino = sbinfo->next_ino++;
Chris Downea3271f2020-08-06 23:20:25 -0700291 if (unlikely(!sbinfo->full_inums &&
292 ino > UINT_MAX)) {
Chris Downe809d5f2020-08-06 23:20:20 -0700293 /*
294 * Emulate get_next_ino uint wraparound for
295 * compatibility
296 */
Chris Downea3271f2020-08-06 23:20:25 -0700297 if (IS_ENABLED(CONFIG_64BIT))
298 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
299 __func__, MINOR(sb->s_dev));
300 sbinfo->next_ino = 1;
301 ino = sbinfo->next_ino++;
Chris Downe809d5f2020-08-06 23:20:20 -0700302 }
303 *inop = ino;
304 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800305 spin_unlock(&sbinfo->stat_lock);
Chris Downe809d5f2020-08-06 23:20:20 -0700306 } else if (inop) {
307 /*
308 * __shmem_file_setup, one of our callers, is lock-free: it
309 * doesn't hold stat_lock in shmem_reserve_inode since
310 * max_inodes is always 0, and is called from potentially
311 * unknown contexts. As such, use a per-cpu batched allocator
312 * which doesn't require the per-sb stat_lock unless we are at
313 * the batch boundary.
Chris Downea3271f2020-08-06 23:20:25 -0700314 *
315 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
316 * shmem mounts are not exposed to userspace, so we don't need
317 * to worry about things like glibc compatibility.
Chris Downe809d5f2020-08-06 23:20:20 -0700318 */
319 ino_t *next_ino;
320 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
321 ino = *next_ino;
322 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
323 spin_lock(&sbinfo->stat_lock);
324 ino = sbinfo->next_ino;
325 sbinfo->next_ino += SHMEM_INO_BATCH;
326 spin_unlock(&sbinfo->stat_lock);
327 if (unlikely(is_zero_ino(ino)))
328 ino++;
329 }
330 *inop = ino;
331 *next_ino = ++ino;
332 put_cpu();
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800333 }
Chris Downe809d5f2020-08-06 23:20:20 -0700334
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800335 return 0;
336}
337
338static void shmem_free_inode(struct super_block *sb)
339{
340 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
341 if (sbinfo->max_inodes) {
342 spin_lock(&sbinfo->stat_lock);
343 sbinfo->free_inodes++;
344 spin_unlock(&sbinfo->stat_lock);
345 }
346}
347
Randy Dunlap46711812008-03-19 17:00:41 -0700348/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700349 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 * @inode: inode to recalc
351 *
352 * We have to calculate the free blocks since the mm can drop
353 * undirtied hole pages behind our back.
354 *
355 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
356 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
357 *
358 * It has to be called with the spinlock held.
359 */
360static void shmem_recalc_inode(struct inode *inode)
361{
362 struct shmem_inode_info *info = SHMEM_I(inode);
363 long freed;
364
365 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
366 if (freed > 0) {
367 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700368 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Mike Rapoport0f079692017-09-06 16:22:59 -0700369 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
371}
372
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700373bool shmem_charge(struct inode *inode, long pages)
374{
375 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700376 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700377
Mike Rapoport0f079692017-09-06 16:22:59 -0700378 if (!shmem_inode_acct_block(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700379 return false;
Mike Rapoportb1cc94a2017-09-06 16:22:56 -0700380
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800381 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
382 inode->i_mapping->nrpages += pages;
383
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700384 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700385 info->alloced += pages;
386 inode->i_blocks += pages * BLOCKS_PER_PAGE;
387 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700388 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700389
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700390 return true;
391}
392
393void shmem_uncharge(struct inode *inode, long pages)
394{
395 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700396 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700397
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800398 /* nrpages adjustment done by __delete_from_page_cache() or caller */
399
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700400 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700401 info->alloced -= pages;
402 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
403 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700404 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700405
Mike Rapoport0f079692017-09-06 16:22:59 -0700406 shmem_inode_unacct_blocks(inode, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700407}
408
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700409/*
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500410 * Replace item expected in xarray by a new item, while holding xa_lock.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700411 */
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500412static int shmem_replace_entry(struct address_space *mapping,
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700413 pgoff_t index, void *expected, void *replacement)
414{
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500415 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700416 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700417
418 VM_BUG_ON(!expected);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700419 VM_BUG_ON(!replacement);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500420 item = xas_load(&xas);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700421 if (item != expected)
422 return -ENOENT;
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500423 xas_store(&xas, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700424 return 0;
425}
426
427/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700428 * Sometimes, before we decide whether to proceed or to fail, we must check
429 * that an entry was not already brought back from swap by a racing thread.
430 *
431 * Checking page is not enough: by the time a SwapCache page is locked, it
432 * might be reused, and again be SwapCache, using the same swap as before.
433 */
434static bool shmem_confirm_swap(struct address_space *mapping,
435 pgoff_t index, swp_entry_t swap)
436{
Matthew Wilcoxa12831b2017-11-22 08:34:58 -0500437 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
Hugh Dickinsd1899222012-07-11 14:02:47 -0700438}
439
440/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700441 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
442 *
443 * SHMEM_HUGE_NEVER:
444 * disables huge pages for the mount;
445 * SHMEM_HUGE_ALWAYS:
446 * enables huge pages for the mount;
447 * SHMEM_HUGE_WITHIN_SIZE:
448 * only allocate huge pages if the page will be fully within i_size,
449 * also respect fadvise()/madvise() hints;
450 * SHMEM_HUGE_ADVISE:
451 * only allocate huge pages if requested with fadvise()/madvise();
452 */
453
454#define SHMEM_HUGE_NEVER 0
455#define SHMEM_HUGE_ALWAYS 1
456#define SHMEM_HUGE_WITHIN_SIZE 2
457#define SHMEM_HUGE_ADVISE 3
458
459/*
460 * Special values.
461 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
462 *
463 * SHMEM_HUGE_DENY:
464 * disables huge on shm_mnt and all mounts, for emergency use;
465 * SHMEM_HUGE_FORCE:
466 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
467 *
468 */
469#define SHMEM_HUGE_DENY (-1)
470#define SHMEM_HUGE_FORCE (-2)
471
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700472#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700473/* ifdef here to avoid bloating shmem.o when not necessary */
474
Mike Kravetz5b9c98f2018-06-07 17:05:53 -0700475static int shmem_huge __read_mostly;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700476
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700477#if defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700478static int shmem_parse_huge(const char *str)
479{
480 if (!strcmp(str, "never"))
481 return SHMEM_HUGE_NEVER;
482 if (!strcmp(str, "always"))
483 return SHMEM_HUGE_ALWAYS;
484 if (!strcmp(str, "within_size"))
485 return SHMEM_HUGE_WITHIN_SIZE;
486 if (!strcmp(str, "advise"))
487 return SHMEM_HUGE_ADVISE;
488 if (!strcmp(str, "deny"))
489 return SHMEM_HUGE_DENY;
490 if (!strcmp(str, "force"))
491 return SHMEM_HUGE_FORCE;
492 return -EINVAL;
493}
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700494#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700495
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700496#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700497static const char *shmem_format_huge(int huge)
498{
499 switch (huge) {
500 case SHMEM_HUGE_NEVER:
501 return "never";
502 case SHMEM_HUGE_ALWAYS:
503 return "always";
504 case SHMEM_HUGE_WITHIN_SIZE:
505 return "within_size";
506 case SHMEM_HUGE_ADVISE:
507 return "advise";
508 case SHMEM_HUGE_DENY:
509 return "deny";
510 case SHMEM_HUGE_FORCE:
511 return "force";
512 default:
513 VM_BUG_ON(1);
514 return "bad_val";
515 }
516}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800517#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700518
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700519static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
520 struct shrink_control *sc, unsigned long nr_to_split)
521{
522 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800523 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700524 struct inode *inode;
525 struct shmem_inode_info *info;
526 struct page *page;
527 unsigned long batch = sc ? sc->nr_to_scan : 128;
528 int removed = 0, split = 0;
529
530 if (list_empty(&sbinfo->shrinklist))
531 return SHRINK_STOP;
532
533 spin_lock(&sbinfo->shrinklist_lock);
534 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
535 info = list_entry(pos, struct shmem_inode_info, shrinklist);
536
537 /* pin the inode */
538 inode = igrab(&info->vfs_inode);
539
540 /* inode is about to be evicted */
541 if (!inode) {
542 list_del_init(&info->shrinklist);
543 removed++;
544 goto next;
545 }
546
547 /* Check if there's anything to gain */
548 if (round_up(inode->i_size, PAGE_SIZE) ==
549 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800550 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700551 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700552 goto next;
553 }
554
555 list_move(&info->shrinklist, &list);
556next:
557 if (!--batch)
558 break;
559 }
560 spin_unlock(&sbinfo->shrinklist_lock);
561
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800562 list_for_each_safe(pos, next, &to_remove) {
563 info = list_entry(pos, struct shmem_inode_info, shrinklist);
564 inode = &info->vfs_inode;
565 list_del_init(&info->shrinklist);
566 iput(inode);
567 }
568
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700569 list_for_each_safe(pos, next, &list) {
570 int ret;
571
572 info = list_entry(pos, struct shmem_inode_info, shrinklist);
573 inode = &info->vfs_inode;
574
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700575 if (nr_to_split && split >= nr_to_split)
576 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700577
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700578 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700579 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
580 if (!page)
581 goto drop;
582
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700583 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700584 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700585 put_page(page);
586 goto drop;
587 }
588
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700589 /*
590 * Leave the inode on the list if we failed to lock
591 * the page at this time.
592 *
593 * Waiting for the lock may lead to deadlock in the
594 * reclaim path.
595 */
596 if (!trylock_page(page)) {
597 put_page(page);
598 goto leave;
599 }
600
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700601 ret = split_huge_page(page);
602 unlock_page(page);
603 put_page(page);
604
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700605 /* If split failed leave the inode on the list */
606 if (ret)
607 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700608
609 split++;
610drop:
611 list_del_init(&info->shrinklist);
612 removed++;
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700613leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700614 iput(inode);
615 }
616
617 spin_lock(&sbinfo->shrinklist_lock);
618 list_splice_tail(&list, &sbinfo->shrinklist);
619 sbinfo->shrinklist_len -= removed;
620 spin_unlock(&sbinfo->shrinklist_lock);
621
622 return split;
623}
624
625static long shmem_unused_huge_scan(struct super_block *sb,
626 struct shrink_control *sc)
627{
628 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
629
630 if (!READ_ONCE(sbinfo->shrinklist_len))
631 return SHRINK_STOP;
632
633 return shmem_unused_huge_shrink(sbinfo, sc, 0);
634}
635
636static long shmem_unused_huge_count(struct super_block *sb,
637 struct shrink_control *sc)
638{
639 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
640 return READ_ONCE(sbinfo->shrinklist_len);
641}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700642#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700643
644#define shmem_huge SHMEM_HUGE_DENY
645
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700646static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
647 struct shrink_control *sc, unsigned long nr_to_split)
648{
649 return 0;
650}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700651#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700652
Yang Shi89fdcd22018-06-07 17:06:59 -0700653static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
654{
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700655 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Yang Shi89fdcd22018-06-07 17:06:59 -0700656 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
657 shmem_huge != SHMEM_HUGE_DENY)
658 return true;
659 return false;
660}
661
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700662/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700663 * Like add_to_page_cache_locked, but error if expected item has gone.
664 */
665static int shmem_add_to_page_cache(struct page *page,
666 struct address_space *mapping,
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700667 pgoff_t index, void *expected, gfp_t gfp,
668 struct mm_struct *charge_mm)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700669{
Matthew Wilcox552446a2017-12-01 13:25:14 -0500670 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
671 unsigned long i = 0;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -0700672 unsigned long nr = compound_nr(page);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700673 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700674
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700675 VM_BUG_ON_PAGE(PageTail(page), page);
676 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800677 VM_BUG_ON_PAGE(!PageLocked(page), page);
678 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700679 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700680
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700681 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700682 page->mapping = mapping;
683 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700684
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700685 if (!PageSwapCache(page)) {
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -0700686 error = mem_cgroup_charge(page, charge_mm, gfp);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700687 if (error) {
688 if (PageTransHuge(page)) {
689 count_vm_event(THP_FILE_FALLBACK);
690 count_vm_event(THP_FILE_FALLBACK_CHARGE);
691 }
692 goto error;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700693 }
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700694 }
695 cgroup_throttle_swaprate(page, gfp);
696
Matthew Wilcox552446a2017-12-01 13:25:14 -0500697 do {
698 void *entry;
699 xas_lock_irq(&xas);
700 entry = xas_find_conflict(&xas);
701 if (entry != expected)
702 xas_set_err(&xas, -EEXIST);
703 xas_create_range(&xas);
704 if (xas_error(&xas))
705 goto unlock;
706next:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700707 xas_store(&xas, page);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500708 if (++i < nr) {
709 xas_next(&xas);
710 goto next;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700711 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500712 if (PageTransHuge(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700713 count_vm_event(THP_FILE_ALLOC);
Mel Gorman11fb9982016-07-28 15:46:20 -0700714 __inc_node_page_state(page, NR_SHMEM_THPS);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500715 }
716 mapping->nrpages += nr;
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700717 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
718 __mod_lruvec_page_state(page, NR_SHMEM, nr);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500719unlock:
720 xas_unlock_irq(&xas);
721 } while (xas_nomem(&xas, gfp));
722
723 if (xas_error(&xas)) {
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700724 error = xas_error(&xas);
725 goto error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700726 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500727
728 return 0;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700729error:
730 page->mapping = NULL;
731 page_ref_sub(page, nr);
732 return error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700733}
734
735/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700736 * Like delete_from_page_cache, but substitutes swap for page.
737 */
738static void shmem_delete_from_page_cache(struct page *page, void *radswap)
739{
740 struct address_space *mapping = page->mapping;
741 int error;
742
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700743 VM_BUG_ON_PAGE(PageCompound(page), page);
744
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700745 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500746 error = shmem_replace_entry(mapping, page->index, page, radswap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700747 page->mapping = NULL;
748 mapping->nrpages--;
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700749 __dec_lruvec_page_state(page, NR_FILE_PAGES);
750 __dec_lruvec_page_state(page, NR_SHMEM);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700751 xa_unlock_irq(&mapping->i_pages);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300752 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700753 BUG_ON(error);
754}
755
756/*
Matthew Wilcoxc121d3b2017-12-04 03:13:54 -0500757 * Remove swap entry from page cache, free the swap and its page cache.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700758 */
759static int shmem_free_swap(struct address_space *mapping,
760 pgoff_t index, void *radswap)
761{
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700762 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700763
Matthew Wilcox55f3f7e2018-11-26 16:08:43 -0500764 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700765 if (old != radswap)
766 return -ENOENT;
767 free_swap_and_cache(radix_to_swp_entry(radswap));
768 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700769}
770
771/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800772 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800773 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800774 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700775 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800776 * as long as the inode doesn't go away and racy results are not a problem.
777 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800778unsigned long shmem_partial_swap_usage(struct address_space *mapping,
779 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800780{
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500781 XA_STATE(xas, &mapping->i_pages, start);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800782 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800783 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800784
785 rcu_read_lock();
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500786 xas_for_each(&xas, page, end - 1) {
787 if (xas_retry(&xas, page))
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700788 continue;
Matthew Wilcox3159f942017-11-03 13:30:42 -0400789 if (xa_is_value(page))
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800790 swapped++;
791
792 if (need_resched()) {
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500793 xas_pause(&xas);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800794 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800795 }
796 }
797
798 rcu_read_unlock();
799
800 return swapped << PAGE_SHIFT;
801}
802
803/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800804 * Determine (in bytes) how many of the shmem object's pages mapped by the
805 * given vma is swapped out.
806 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700807 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka48131e02016-01-14 15:19:23 -0800808 * as long as the inode doesn't go away and racy results are not a problem.
809 */
810unsigned long shmem_swap_usage(struct vm_area_struct *vma)
811{
812 struct inode *inode = file_inode(vma->vm_file);
813 struct shmem_inode_info *info = SHMEM_I(inode);
814 struct address_space *mapping = inode->i_mapping;
815 unsigned long swapped;
816
817 /* Be careful as we don't hold info->lock */
818 swapped = READ_ONCE(info->swapped);
819
820 /*
821 * The easier cases are when the shmem object has nothing in swap, or
822 * the vma maps it whole. Then we can simply use the stats that we
823 * already track.
824 */
825 if (!swapped)
826 return 0;
827
828 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
829 return swapped << PAGE_SHIFT;
830
831 /* Here comes the more involved part */
832 return shmem_partial_swap_usage(mapping,
833 linear_page_index(vma, vma->vm_start),
834 linear_page_index(vma, vma->vm_end));
835}
836
837/*
Hugh Dickins24513262012-01-20 14:34:21 -0800838 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
839 */
840void shmem_unlock_mapping(struct address_space *mapping)
841{
842 struct pagevec pvec;
843 pgoff_t indices[PAGEVEC_SIZE];
844 pgoff_t index = 0;
845
Mel Gorman86679822017-11-15 17:37:52 -0800846 pagevec_init(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800847 /*
848 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
849 */
850 while (!mapping_unevictable(mapping)) {
851 /*
852 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
853 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
854 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700855 pvec.nr = find_get_entries(mapping, index,
856 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800857 if (!pvec.nr)
858 break;
859 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700860 pagevec_remove_exceptionals(&pvec);
Kuo-Hsin Yang64e3d122018-11-06 13:23:24 +0000861 check_move_unevictable_pages(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800862 pagevec_release(&pvec);
863 cond_resched();
864 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700865}
866
867/*
Hugh Dickins71725ed2020-04-06 20:07:57 -0700868 * Check whether a hole-punch or truncation needs to split a huge page,
869 * returning true if no split was required, or the split has been successful.
870 *
871 * Eviction (or truncation to 0 size) should never need to split a huge page;
872 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
873 * head, and then succeeded to trylock on tail.
874 *
875 * A split can only succeed when there are no additional references on the
876 * huge page: so the split below relies upon find_get_entries() having stopped
877 * when it found a subpage of the huge page, without getting further references.
878 */
879static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
880{
881 if (!PageTransCompound(page))
882 return true;
883
884 /* Just proceed to delete a huge page wholly within the range punched */
885 if (PageHead(page) &&
886 page->index >= start && page->index + HPAGE_PMD_NR <= end)
887 return true;
888
889 /* Try to split huge page, so we can truly punch the hole or truncate */
890 return split_huge_page(page) >= 0;
891}
892
893/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -0500894 * Remove range of pages and swap entries from page cache, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700895 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700896 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700897static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
898 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700900 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300902 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
903 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
904 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
905 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700906 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700907 pgoff_t indices[PAGEVEC_SIZE];
908 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700909 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700910 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700912 if (lend == -1)
913 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700914
Mel Gorman86679822017-11-15 17:37:52 -0800915 pagevec_init(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700916 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700917 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700918 pvec.nr = find_get_entries(mapping, index,
919 min(end - index, (pgoff_t)PAGEVEC_SIZE),
920 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700921 if (!pvec.nr)
922 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700923 for (i = 0; i < pagevec_count(&pvec); i++) {
924 struct page *page = pvec.pages[i];
925
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700926 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700927 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700928 break;
929
Matthew Wilcox3159f942017-11-03 13:30:42 -0400930 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700931 if (unfalloc)
932 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700933 nr_swaps_freed += !shmem_free_swap(mapping,
934 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700935 continue;
936 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700937
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700938 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
939
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700940 if (!trylock_page(page))
941 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700942
Hugh Dickins71725ed2020-04-06 20:07:57 -0700943 if ((!unfalloc || !PageUptodate(page)) &&
944 page_mapping(page) == mapping) {
945 VM_BUG_ON_PAGE(PageWriteback(page), page);
946 if (shmem_punch_compound(page, start, end))
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700947 truncate_inode_page(mapping, page);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700948 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700949 unlock_page(page);
950 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700951 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800952 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700953 cond_resched();
954 index++;
955 }
956
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700957 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700958 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700959 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700960 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300961 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700962 if (start > end) {
963 top = partial_end;
964 partial_end = 0;
965 }
966 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700967 set_page_dirty(page);
968 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300969 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700970 }
971 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700972 if (partial_end) {
973 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700974 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700975 if (page) {
976 zero_user_segment(page, 0, partial_end);
977 set_page_dirty(page);
978 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300979 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700980 }
981 }
982 if (start >= end)
983 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700984
985 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700986 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700987 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700988
989 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700990 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700991 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700992 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700993 /* If all gone or hole-punch or unfalloc, we're done */
994 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700995 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700996 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700997 index = start;
998 continue;
999 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001000 for (i = 0; i < pagevec_count(&pvec); i++) {
1001 struct page *page = pvec.pages[i];
1002
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001003 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001004 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001005 break;
1006
Matthew Wilcox3159f942017-11-03 13:30:42 -04001007 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001008 if (unfalloc)
1009 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001010 if (shmem_free_swap(mapping, index, page)) {
1011 /* Swap was replaced by page: retry */
1012 index--;
1013 break;
1014 }
1015 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001016 continue;
1017 }
1018
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001019 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001020
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001021 if (!unfalloc || !PageUptodate(page)) {
Hugh Dickins71725ed2020-04-06 20:07:57 -07001022 if (page_mapping(page) != mapping) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001023 /* Page was replaced by swap: retry */
1024 unlock_page(page);
1025 index--;
1026 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001027 }
Hugh Dickins71725ed2020-04-06 20:07:57 -07001028 VM_BUG_ON_PAGE(PageWriteback(page), page);
1029 if (shmem_punch_compound(page, start, end))
1030 truncate_inode_page(mapping, page);
Hugh Dickins0783ac92020-04-20 18:14:07 -07001031 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
Hugh Dickins71725ed2020-04-06 20:07:57 -07001032 /* Wipe the page and don't get stuck */
1033 clear_highpage(page);
1034 flush_dcache_page(page);
1035 set_page_dirty(page);
1036 if (index <
1037 round_up(start, HPAGE_PMD_NR))
1038 start = index + 1;
1039 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001040 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001041 unlock_page(page);
1042 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001043 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -08001044 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001045 index++;
1046 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001047
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001048 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001049 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001051 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001052}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001054void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1055{
1056 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001057 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058}
Hugh Dickins94c1e622011-06-27 16:18:03 -07001059EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
David Howellsa528d352017-01-31 16:46:22 +00001061static int shmem_getattr(const struct path *path, struct kstat *stat,
1062 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -07001063{
David Howellsa528d352017-01-31 16:46:22 +00001064 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -07001065 struct shmem_inode_info *info = SHMEM_I(inode);
Yang Shi89fdcd22018-06-07 17:06:59 -07001066 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
Yu Zhao44a30222015-09-08 15:03:33 -07001067
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001068 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001069 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001070 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001071 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001072 }
Yu Zhao44a30222015-09-08 15:03:33 -07001073 generic_fillattr(inode, stat);
Yang Shi89fdcd22018-06-07 17:06:59 -07001074
1075 if (is_huge_enabled(sb_info))
1076 stat->blksize = HPAGE_PMD_SIZE;
1077
Yu Zhao44a30222015-09-08 15:03:33 -07001078 return 0;
1079}
1080
Hugh Dickins94c1e622011-06-27 16:18:03 -07001081static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082{
David Howells75c3cfa2015-03-17 22:26:12 +00001083 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -07001084 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001085 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 int error;
1087
Jan Kara31051c82016-05-26 16:55:18 +02001088 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b8772010-06-04 11:30:03 +02001089 if (error)
1090 return error;
1091
Hugh Dickins94c1e622011-06-27 16:18:03 -07001092 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1093 loff_t oldsize = inode->i_size;
1094 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001095
David Herrmann40e041a2014-08-08 14:25:27 -07001096 /* protected by i_mutex */
1097 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1098 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1099 return -EPERM;
1100
Hugh Dickins94c1e622011-06-27 16:18:03 -07001101 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001102 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1103 oldsize, newsize);
1104 if (error)
1105 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001106 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001107 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001108 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001109 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001110 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001111 if (oldsize > holebegin)
1112 unmap_mapping_range(inode->i_mapping,
1113 holebegin, 0, 1);
1114 if (info->alloced)
1115 shmem_truncate_range(inode,
1116 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001117 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001118 if (oldsize > holebegin)
1119 unmap_mapping_range(inode->i_mapping,
1120 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001121
1122 /*
1123 * Part of the huge page can be beyond i_size: subject
1124 * to shrink under memory pressure.
1125 */
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001126 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001127 spin_lock(&sbinfo->shrinklist_lock);
Cong Wangd0413532017-08-10 15:24:24 -07001128 /*
1129 * _careful to defend against unlocked access to
1130 * ->shrink_list in shmem_unused_huge_shrink()
1131 */
1132 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001133 list_add_tail(&info->shrinklist,
1134 &sbinfo->shrinklist);
1135 sbinfo->shrinklist_len++;
1136 }
1137 spin_unlock(&sbinfo->shrinklist_lock);
1138 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 }
1141
Christoph Hellwigdb78b8772010-06-04 11:30:03 +02001142 setattr_copy(inode, attr);
Christoph Hellwigdb78b8772010-06-04 11:30:03 +02001143 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001144 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 return error;
1146}
1147
Al Viro1f895f72010-06-05 19:10:41 -04001148static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001151 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001153 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 shmem_unacct_size(info->flags, inode->i_size);
1155 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001156 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001157 if (!list_empty(&info->shrinklist)) {
1158 spin_lock(&sbinfo->shrinklist_lock);
1159 if (!list_empty(&info->shrinklist)) {
1160 list_del_init(&info->shrinklist);
1161 sbinfo->shrinklist_len--;
1162 }
1163 spin_unlock(&sbinfo->shrinklist_lock);
1164 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001165 while (!list_empty(&info->swaplist)) {
1166 /* Wait while shmem_unuse() is scanning this inode... */
1167 wait_var_event(&info->stop_eviction,
1168 !atomic_read(&info->stop_eviction));
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001169 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001170 /* ...but beware of the race if we peeked too early */
1171 if (!atomic_read(&info->stop_eviction))
1172 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001173 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
Al Viro3ed47db2016-01-22 18:08:52 -05001175 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001176
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001177 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001178 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001179 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001180 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001183extern struct swap_info_struct *swap_info[];
1184
1185static int shmem_find_swap_entries(struct address_space *mapping,
1186 pgoff_t start, unsigned int nr_entries,
1187 struct page **entries, pgoff_t *indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001188 unsigned int type, bool frontswap)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001189{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001190 XA_STATE(xas, &mapping->i_pages, start);
1191 struct page *page;
Hugh Dickins87039542019-04-18 17:49:58 -07001192 swp_entry_t entry;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001193 unsigned int ret = 0;
1194
1195 if (!nr_entries)
1196 return 0;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001197
1198 rcu_read_lock();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001199 xas_for_each(&xas, page, ULONG_MAX) {
1200 if (xas_retry(&xas, page))
Mike Kravetz5b9c98f2018-06-07 17:05:53 -07001201 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001202
1203 if (!xa_is_value(page))
1204 continue;
1205
Hugh Dickins87039542019-04-18 17:49:58 -07001206 entry = radix_to_swp_entry(page);
1207 if (swp_type(entry) != type)
1208 continue;
1209 if (frontswap &&
1210 !frontswap_test(swap_info[type], swp_offset(entry)))
1211 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001212
1213 indices[ret] = xas.xa_index;
1214 entries[ret] = page;
1215
1216 if (need_resched()) {
1217 xas_pause(&xas);
1218 cond_resched_rcu();
1219 }
1220 if (++ret == nr_entries)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001221 break;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001222 }
Matthew Wilcox478922e2016-12-14 15:08:52 -08001223 rcu_read_unlock();
Matthew Wilcoxe21a2952017-11-22 08:36:00 -05001224
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001225 return ret;
1226}
1227
1228/*
1229 * Move the swapped pages for an inode to page cache. Returns the count
1230 * of pages swapped in, or the error in case of failure.
1231 */
1232static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1233 pgoff_t *indices)
1234{
1235 int i = 0;
1236 int ret = 0;
1237 int error = 0;
1238 struct address_space *mapping = inode->i_mapping;
1239
1240 for (i = 0; i < pvec.nr; i++) {
1241 struct page *page = pvec.pages[i];
1242
1243 if (!xa_is_value(page))
1244 continue;
1245 error = shmem_swapin_page(inode, indices[i],
1246 &page, SGP_CACHE,
1247 mapping_gfp_mask(mapping),
1248 NULL, NULL);
1249 if (error == 0) {
1250 unlock_page(page);
1251 put_page(page);
1252 ret++;
1253 }
1254 if (error == -ENOMEM)
1255 break;
1256 error = 0;
1257 }
1258 return error ? error : ret;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001259}
1260
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001261/*
1262 * If swap found in inode, free it and move page from swapcache to filecache.
1263 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001264static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1265 bool frontswap, unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001267 struct address_space *mapping = inode->i_mapping;
1268 pgoff_t start = 0;
1269 struct pagevec pvec;
1270 pgoff_t indices[PAGEVEC_SIZE];
1271 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1272 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001274 pagevec_init(&pvec);
1275 do {
1276 unsigned int nr_entries = PAGEVEC_SIZE;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001277
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001278 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1279 nr_entries = *fs_pages_to_unuse;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001280
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001281 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1282 pvec.pages, indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001283 type, frontswap);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001284 if (pvec.nr == 0) {
1285 ret = 0;
1286 break;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001287 }
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001288
1289 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1290 if (ret < 0)
1291 break;
1292
1293 if (frontswap_partial) {
1294 *fs_pages_to_unuse -= ret;
1295 if (*fs_pages_to_unuse == 0) {
1296 ret = FRONTSWAP_PAGES_UNUSED;
1297 break;
1298 }
1299 }
1300
1301 start = indices[pvec.nr - 1];
1302 } while (true);
1303
1304 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305}
1306
1307/*
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001308 * Read all the shared memory data that resides in the swap
1309 * device 'type' back into memory, so the swap device can be
1310 * unused.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001312int shmem_unuse(unsigned int type, bool frontswap,
1313 unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001315 struct shmem_inode_info *info, *next;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001316 int error = 0;
1317
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001318 if (list_empty(&shmem_swaplist))
1319 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001321 mutex_lock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001322 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1323 if (!info->swapped) {
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001324 list_del_init(&info->swaplist);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001325 continue;
1326 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001327 /*
1328 * Drop the swaplist mutex while searching the inode for swap;
1329 * but before doing so, make sure shmem_evict_inode() will not
1330 * remove placeholder inode from swaplist, nor let it be freed
1331 * (igrab() would protect from unlink, but not from unmount).
1332 */
1333 atomic_inc(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001334 mutex_unlock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001335
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001336 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001337 fs_pages_to_unuse);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001338 cond_resched();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001339
1340 mutex_lock(&shmem_swaplist_mutex);
1341 next = list_next_entry(info, swaplist);
1342 if (!info->swapped)
1343 list_del_init(&info->swaplist);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001344 if (atomic_dec_and_test(&info->stop_eviction))
1345 wake_up_var(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001346 if (error)
Hugh Dickins778dd892011-05-11 15:13:37 -07001347 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001349 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001350
Hugh Dickins778dd892011-05-11 15:13:37 -07001351 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352}
1353
1354/*
1355 * Move the page from the page cache to the swap cache.
1356 */
1357static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1358{
1359 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001362 swp_entry_t swap;
1363 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001365 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 mapping = page->mapping;
1368 index = page->index;
1369 inode = mapping->host;
1370 info = SHMEM_I(inode);
1371 if (info->flags & VM_LOCKED)
1372 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001373 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 goto redirty;
1375
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001376 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001377 * Our capabilities prevent regular writeback or sync from ever calling
1378 * shmem_writepage; but a stacking filesystem might use ->writepage of
1379 * its underlying filesystem, in which case tmpfs should write out to
1380 * swap only in response to memory pressure, and not for the writeback
1381 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001382 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001383 if (!wbc->for_reclaim) {
1384 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1385 goto redirty;
1386 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001387
1388 /*
1389 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1390 * value into swapfile.c, the only way we can correctly account for a
1391 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001392 *
1393 * That's okay for a page already fallocated earlier, but if we have
1394 * not yet completed the fallocation, then (a) we want to keep track
1395 * of this page in case we have to undo it, and (b) it may not be a
1396 * good idea to continue anyway, once we're pushing into swap. So
1397 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001398 */
1399 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001400 if (inode->i_private) {
1401 struct shmem_falloc *shmem_falloc;
1402 spin_lock(&inode->i_lock);
1403 shmem_falloc = inode->i_private;
1404 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001405 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001406 index >= shmem_falloc->start &&
1407 index < shmem_falloc->next)
1408 shmem_falloc->nr_unswapped++;
1409 else
1410 shmem_falloc = NULL;
1411 spin_unlock(&inode->i_lock);
1412 if (shmem_falloc)
1413 goto redirty;
1414 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001415 clear_highpage(page);
1416 flush_dcache_page(page);
1417 SetPageUptodate(page);
1418 }
1419
Huang Ying38d8b4e2017-07-06 15:37:18 -07001420 swap = get_swap_page(page);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001421 if (!swap.val)
1422 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001423
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001424 /*
1425 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001426 * if it's not already there. Do it now before the page is
1427 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001428 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001429 * we've incremented swapped, because shmem_unuse_inode() will
1430 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001431 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001432 mutex_lock(&shmem_swaplist_mutex);
1433 if (list_empty(&info->swaplist))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001434 list_add(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001435
Yang Shi4afab1c2019-11-30 17:58:07 -08001436 if (add_to_swap_cache(page, swap,
Joonsoo Kim3852f672020-08-11 18:30:47 -07001437 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1438 NULL) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001439 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001440 shmem_recalc_inode(inode);
1441 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001442 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001443
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001444 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001445 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1446
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001447 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001448 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001449 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return 0;
1451 }
1452
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001453 mutex_unlock(&shmem_swaplist_mutex);
Minchan Kim75f6d6d2017-07-06 15:37:21 -07001454 put_swap_page(page, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455redirty:
1456 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001457 if (wbc->for_reclaim)
1458 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1459 unlock_page(page);
1460 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
Hugh Dickins75edd342016-05-19 17:12:44 -07001463#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001464static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001465{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001466 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001467
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001468 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001469 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001470
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001471 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001472
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001473 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001474}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001475
1476static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1477{
1478 struct mempolicy *mpol = NULL;
1479 if (sbinfo->mpol) {
1480 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1481 mpol = sbinfo->mpol;
1482 mpol_get(mpol);
1483 spin_unlock(&sbinfo->stat_lock);
1484 }
1485 return mpol;
1486}
Hugh Dickins75edd342016-05-19 17:12:44 -07001487#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1488static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1489{
1490}
1491static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1492{
1493 return NULL;
1494}
1495#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1496#ifndef CONFIG_NUMA
1497#define vm_policy vm_private_data
1498#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001499
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001500static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1501 struct shmem_inode_info *info, pgoff_t index)
1502{
1503 /* Create a pseudo vma that just contains the policy */
Kirill A. Shutemov2c4541e2018-07-26 16:37:30 -07001504 vma_init(vma, NULL);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001505 /* Bias interleave by inode number to distribute better across nodes */
1506 vma->vm_pgoff = index + info->vfs_inode.i_ino;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001507 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1508}
1509
1510static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1511{
1512 /* Drop reference taken by mpol_shared_policy_lookup() */
1513 mpol_cond_put(vma->vm_policy);
1514}
1515
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001516static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1517 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001520 struct page *page;
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001521 struct vm_fault vmf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001523 shmem_pseudo_vma_init(&pvma, info, index);
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001524 vmf.vma = &pvma;
1525 vmf.address = 0;
1526 page = swap_cluster_readahead(swap, gfp, &vmf);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001527 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001528
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001529 return page;
1530}
Mel Gorman18a2f372012-12-05 14:01:41 -08001531
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001532static struct page *shmem_alloc_hugepage(gfp_t gfp,
1533 struct shmem_inode_info *info, pgoff_t index)
1534{
1535 struct vm_area_struct pvma;
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001536 struct address_space *mapping = info->vfs_inode.i_mapping;
1537 pgoff_t hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001538 struct page *page;
1539
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001540 hindex = round_down(index, HPAGE_PMD_NR);
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001541 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1542 XA_PRESENT))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001543 return NULL;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001544
1545 shmem_pseudo_vma_init(&pvma, info, hindex);
1546 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
David Rientjes19deb762019-09-04 12:54:20 -07001547 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001548 shmem_pseudo_vma_destroy(&pvma);
1549 if (page)
1550 prep_transhuge_page(page);
David Rientjesdcdf11e2020-04-06 20:04:25 -07001551 else
1552 count_vm_event(THP_FILE_FALLBACK);
Mel Gorman18a2f372012-12-05 14:01:41 -08001553 return page;
1554}
1555
1556static struct page *shmem_alloc_page(gfp_t gfp,
1557 struct shmem_inode_info *info, pgoff_t index)
1558{
1559 struct vm_area_struct pvma;
1560 struct page *page;
1561
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001562 shmem_pseudo_vma_init(&pvma, info, index);
1563 page = alloc_page_vma(gfp, &pvma, 0);
1564 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001565
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001566 return page;
1567}
1568
1569static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
Mike Rapoport0f079692017-09-06 16:22:59 -07001570 struct inode *inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001571 pgoff_t index, bool huge)
1572{
Mike Rapoport0f079692017-09-06 16:22:59 -07001573 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001574 struct page *page;
1575 int nr;
1576 int err = -ENOSPC;
1577
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001578 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001579 huge = false;
1580 nr = huge ? HPAGE_PMD_NR : 1;
1581
Mike Rapoport0f079692017-09-06 16:22:59 -07001582 if (!shmem_inode_acct_block(inode, nr))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001583 goto failed;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001584
1585 if (huge)
1586 page = shmem_alloc_hugepage(gfp, info, index);
1587 else
1588 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001589 if (page) {
1590 __SetPageLocked(page);
1591 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001592 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001593 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001594
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001595 err = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07001596 shmem_inode_unacct_blocks(inode, nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001597failed:
1598 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001600
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001602 * When a page is moved from swapcache to shmem filecache (either by the
1603 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1604 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1605 * ignorance of the mapping it belongs to. If that mapping has special
1606 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1607 * we may need to copy to a suitable page before moving to filecache.
1608 *
1609 * In a future release, this may well be extended to respect cpuset and
1610 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1611 * but for now it is a simple matter of zone.
1612 */
1613static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1614{
1615 return page_zonenum(page) > gfp_zone(gfp);
1616}
1617
1618static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1619 struct shmem_inode_info *info, pgoff_t index)
1620{
1621 struct page *oldpage, *newpage;
1622 struct address_space *swap_mapping;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001623 swp_entry_t entry;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001624 pgoff_t swap_index;
1625 int error;
1626
1627 oldpage = *pagep;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001628 entry.val = page_private(oldpage);
1629 swap_index = swp_offset(entry);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001630 swap_mapping = page_mapping(oldpage);
1631
1632 /*
1633 * We have arrived here because our zones are constrained, so don't
1634 * limit chance of success by further cpuset and node constraints.
1635 */
1636 gfp &= ~GFP_CONSTRAINT_MASK;
1637 newpage = shmem_alloc_page(gfp, info, index);
1638 if (!newpage)
1639 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001640
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001641 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001642 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001643 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001644
Hugh Dickins9956edf2016-11-10 10:46:11 -08001645 __SetPageLocked(newpage);
1646 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001647 SetPageUptodate(newpage);
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001648 set_page_private(newpage, entry.val);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001649 SetPageSwapCache(newpage);
1650
1651 /*
1652 * Our caller will very soon move newpage out of swapcache, but it's
1653 * a nice clean interface for us to replace oldpage by newpage there.
1654 */
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001655 xa_lock_irq(&swap_mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -05001656 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001657 if (!error) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07001658 mem_cgroup_migrate(oldpage, newpage);
1659 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1660 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001661 }
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001662 xa_unlock_irq(&swap_mapping->i_pages);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001663
Hugh Dickins0142ef62012-06-07 14:21:09 -07001664 if (unlikely(error)) {
1665 /*
1666 * Is this possible? I think not, now that our callers check
1667 * both PageSwapCache and page_private after getting page lock;
1668 * but be defensive. Reverse old to newpage for clear and free.
1669 */
1670 oldpage = newpage;
1671 } else {
Johannes Weiner6058eae2020-06-03 16:02:40 -07001672 lru_cache_add(newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001673 *pagep = newpage;
1674 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001675
1676 ClearPageSwapCache(oldpage);
1677 set_page_private(oldpage, 0);
1678
1679 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001680 put_page(oldpage);
1681 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001682 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001683}
1684
1685/*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001686 * Swap in the page pointed to by *pagep.
1687 * Caller has to make sure that *pagep contains a valid swapped page.
1688 * Returns 0 and the page in pagep if success. On failure, returns the
Randy Dunlapaf44c122020-08-11 18:33:17 -07001689 * error code and NULL in *pagep.
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001690 */
1691static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1692 struct page **pagep, enum sgp_type sgp,
1693 gfp_t gfp, struct vm_area_struct *vma,
1694 vm_fault_t *fault_type)
1695{
1696 struct address_space *mapping = inode->i_mapping;
1697 struct shmem_inode_info *info = SHMEM_I(inode);
1698 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001699 struct page *page;
1700 swp_entry_t swap;
1701 int error;
1702
1703 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1704 swap = radix_to_swp_entry(*pagep);
1705 *pagep = NULL;
1706
1707 /* Look it up and read it in.. */
1708 page = lookup_swap_cache(swap, NULL, 0);
1709 if (!page) {
1710 /* Or update major stats only when swapin succeeds?? */
1711 if (fault_type) {
1712 *fault_type |= VM_FAULT_MAJOR;
1713 count_vm_event(PGMAJFAULT);
1714 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1715 }
1716 /* Here we actually start the io */
1717 page = shmem_swapin(swap, gfp, info, index);
1718 if (!page) {
1719 error = -ENOMEM;
1720 goto failed;
1721 }
1722 }
1723
1724 /* We have to do this with page locked to prevent races */
1725 lock_page(page);
1726 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1727 !shmem_confirm_swap(mapping, index, swap)) {
1728 error = -EEXIST;
1729 goto unlock;
1730 }
1731 if (!PageUptodate(page)) {
1732 error = -EIO;
1733 goto failed;
1734 }
1735 wait_on_page_writeback(page);
1736
1737 if (shmem_should_replace_page(page, gfp)) {
1738 error = shmem_replace_page(&page, gfp, info, index);
1739 if (error)
1740 goto failed;
1741 }
1742
Johannes Weiner3fea5a42020-06-03 16:01:41 -07001743 error = shmem_add_to_page_cache(page, mapping, index,
1744 swp_to_radix_entry(swap), gfp,
1745 charge_mm);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001746 if (error)
1747 goto failed;
1748
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001749 spin_lock_irq(&info->lock);
1750 info->swapped--;
1751 shmem_recalc_inode(inode);
1752 spin_unlock_irq(&info->lock);
1753
1754 if (sgp == SGP_WRITE)
1755 mark_page_accessed(page);
1756
1757 delete_from_swap_cache(page);
1758 set_page_dirty(page);
1759 swap_free(swap);
1760
1761 *pagep = page;
1762 return 0;
1763failed:
1764 if (!shmem_confirm_swap(mapping, index, swap))
1765 error = -EEXIST;
1766unlock:
1767 if (page) {
1768 unlock_page(page);
1769 put_page(page);
1770 }
1771
1772 return error;
1773}
1774
1775/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001776 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 *
1778 * If we allocate a new one we do not mark it dirty. That's up to the
1779 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001780 * entry since a page cannot live in both the swap and page cache.
1781 *
Miles Chen28eb3c802019-09-23 15:39:34 -07001782 * vmf and fault_type are only supplied by shmem_fault:
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001783 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001785static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001786 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
Souptick Joarder2b740302018-08-23 17:01:36 -07001787 struct vm_area_struct *vma, struct vm_fault *vmf,
1788 vm_fault_t *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789{
1790 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann23f919d2016-12-12 16:42:28 -08001791 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001793 struct mm_struct *charge_mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001794 struct page *page;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001795 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001796 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001798 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001799 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001801 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001803 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1804 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805repeat:
Hugh Dickins75edd342016-05-19 17:12:44 -07001806 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001807 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001808 return -EINVAL;
1809 }
1810
1811 sbinfo = SHMEM_SB(inode->i_sb);
1812 charge_mm = vma ? vma->vm_mm : current->mm;
1813
1814 page = find_lock_entry(mapping, index);
1815 if (xa_is_value(page)) {
1816 error = shmem_swapin_page(inode, index, &page,
1817 sgp, gfp, vma, fault_type);
1818 if (error == -EEXIST)
1819 goto repeat;
1820
1821 *pagep = page;
1822 return error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001823 }
1824
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001825 if (page && sgp == SGP_WRITE)
1826 mark_page_accessed(page);
1827
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001828 /* fallocated page? */
1829 if (page && !PageUptodate(page)) {
1830 if (sgp != SGP_READ)
1831 goto clear;
1832 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001833 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001834 page = NULL;
1835 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001836 if (page || sgp == SGP_READ) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001837 *pagep = page;
1838 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
1841 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001842 * Fast cache lookup did not find it:
1843 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001845
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001846 if (vma && userfaultfd_missing(vma)) {
1847 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1848 return 0;
1849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001851 /* shmem_symlink() */
1852 if (mapping->a_ops != &shmem_aops)
1853 goto alloc_nohuge;
1854 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1855 goto alloc_nohuge;
1856 if (shmem_huge == SHMEM_HUGE_FORCE)
1857 goto alloc_huge;
1858 switch (sbinfo->huge) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001859 case SHMEM_HUGE_NEVER:
1860 goto alloc_nohuge;
Kees Cook27d80fa22020-04-06 20:07:51 -07001861 case SHMEM_HUGE_WITHIN_SIZE: {
1862 loff_t i_size;
1863 pgoff_t off;
1864
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001865 off = round_up(index, HPAGE_PMD_NR);
1866 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1867 if (i_size >= HPAGE_PMD_SIZE &&
1868 i_size >> PAGE_SHIFT >= off)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001869 goto alloc_huge;
Kees Cook27d80fa22020-04-06 20:07:51 -07001870
1871 fallthrough;
1872 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001873 case SHMEM_HUGE_ADVISE:
1874 if (sgp_huge == SGP_HUGE)
1875 goto alloc_huge;
1876 /* TODO: implement fadvise() hints */
1877 goto alloc_nohuge;
1878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001880alloc_huge:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001881 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1882 if (IS_ERR(page)) {
1883alloc_nohuge:
1884 page = shmem_alloc_and_acct_page(gfp, inode,
1885 index, false);
1886 }
1887 if (IS_ERR(page)) {
1888 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001889
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001890 error = PTR_ERR(page);
1891 page = NULL;
1892 if (error != -ENOSPC)
1893 goto unlock;
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001894 /*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001895 * Try to reclaim some space by splitting a huge page
1896 * beyond i_size on the filesystem.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001897 */
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001898 while (retry--) {
1899 int ret;
1900
1901 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1902 if (ret == SHRINK_STOP)
1903 break;
1904 if (ret)
1905 goto alloc_nohuge;
1906 }
1907 goto unlock;
1908 }
1909
1910 if (PageTransHuge(page))
1911 hindex = round_down(index, HPAGE_PMD_NR);
1912 else
1913 hindex = index;
1914
1915 if (sgp == SGP_WRITE)
1916 __SetPageReferenced(page);
1917
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001918 error = shmem_add_to_page_cache(page, mapping, hindex,
Johannes Weiner3fea5a42020-06-03 16:01:41 -07001919 NULL, gfp & GFP_RECLAIM_MASK,
1920 charge_mm);
1921 if (error)
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001922 goto unacct;
Johannes Weiner6058eae2020-06-03 16:02:40 -07001923 lru_cache_add(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001924
1925 spin_lock_irq(&info->lock);
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001926 info->alloced += compound_nr(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001927 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1928 shmem_recalc_inode(inode);
1929 spin_unlock_irq(&info->lock);
1930 alloced = true;
1931
1932 if (PageTransHuge(page) &&
1933 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1934 hindex + HPAGE_PMD_NR - 1) {
1935 /*
1936 * Part of the huge page is beyond i_size: subject
1937 * to shrink under memory pressure.
1938 */
1939 spin_lock(&sbinfo->shrinklist_lock);
1940 /*
1941 * _careful to defend against unlocked access to
1942 * ->shrink_list in shmem_unused_huge_shrink()
1943 */
1944 if (list_empty_careful(&info->shrinklist)) {
1945 list_add_tail(&info->shrinklist,
1946 &sbinfo->shrinklist);
1947 sbinfo->shrinklist_len++;
1948 }
1949 spin_unlock(&sbinfo->shrinklist_lock);
1950 }
1951
1952 /*
1953 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1954 */
1955 if (sgp == SGP_FALLOC)
1956 sgp = SGP_WRITE;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001957clear:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001958 /*
1959 * Let SGP_WRITE caller clear ends if write does not fill page;
1960 * but SGP_FALLOC on a page fallocated earlier must initialize
1961 * it now, lest undo on failure cancel our earlier guarantee.
1962 */
1963 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1964 struct page *head = compound_head(page);
1965 int i;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001966
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001967 for (i = 0; i < compound_nr(head); i++) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001968 clear_highpage(head + i);
1969 flush_dcache_page(head + i);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001970 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001971 SetPageUptodate(head);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001972 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001973
Hugh Dickins54af60422011-08-03 16:21:24 -07001974 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001975 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001976 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001977 if (alloced) {
1978 ClearPageDirty(page);
1979 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001980 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001981 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001982 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001983 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001984 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001985 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001986 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001987 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001988 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001989
Nick Piggind0217ac2007-07-19 01:47:03 -07001990 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001991 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001993unacct:
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001994 shmem_inode_unacct_blocks(inode, compound_nr(page));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001995
1996 if (PageTransHuge(page)) {
1997 unlock_page(page);
1998 put_page(page);
1999 goto alloc_nohuge;
2000 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07002001unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07002002 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07002003 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002004 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07002005 }
2006 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002007 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07002008 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002009 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07002010 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002011 }
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002012 if (error == -EEXIST)
Hugh Dickins54af60422011-08-03 16:21:24 -07002013 goto repeat;
2014 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015}
2016
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002017/*
2018 * This is like autoremove_wake_function, but it removes the wait queue
2019 * entry unconditionally - even if something else had already woken the
2020 * target.
2021 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02002022static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002023{
2024 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02002025 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002026 return ret;
2027}
2028
Souptick Joarder20acce62018-06-07 17:09:17 -07002029static vm_fault_t shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030{
Dave Jiang11bac802017-02-24 14:56:41 -08002031 struct vm_area_struct *vma = vmf->vma;
Al Viro496ad9a2013-01-23 17:07:38 -05002032 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002033 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002034 enum sgp_type sgp;
Souptick Joarder20acce62018-06-07 17:09:17 -07002035 int err;
2036 vm_fault_t ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002038 /*
2039 * Trinity finds that probing a hole which tmpfs is punching can
2040 * prevent the hole-punch from ever completing: which in turn
2041 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07002042 * faulting pages into the hole while it's being punched. Although
2043 * shmem_undo_range() does remove the additions, it may be unable to
2044 * keep up, as each new page needs its own unmap_mapping_range() call,
2045 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2046 *
2047 * It does not matter if we sometimes reach this check just before the
2048 * hole-punch begins, so that one fault then races with the punch:
2049 * we just need to make racing faults a rare case.
2050 *
2051 * The implementation below would be much simpler if we just used a
2052 * standard mutex or completion: but we cannot take i_mutex in fault,
2053 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002054 */
2055 if (unlikely(inode->i_private)) {
2056 struct shmem_falloc *shmem_falloc;
2057
2058 spin_lock(&inode->i_lock);
2059 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002060 if (shmem_falloc &&
2061 shmem_falloc->waitq &&
2062 vmf->pgoff >= shmem_falloc->start &&
2063 vmf->pgoff < shmem_falloc->next) {
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002064 struct file *fpin;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002065 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002066 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002067
2068 ret = VM_FAULT_NOPAGE;
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002069 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2070 if (fpin)
Hugh Dickins8e205f72014-07-23 14:00:10 -07002071 ret = VM_FAULT_RETRY;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002072
2073 shmem_falloc_waitq = shmem_falloc->waitq;
2074 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2075 TASK_UNINTERRUPTIBLE);
2076 spin_unlock(&inode->i_lock);
2077 schedule();
2078
2079 /*
2080 * shmem_falloc_waitq points into the shmem_fallocate()
2081 * stack of the hole-punching task: shmem_falloc_waitq
2082 * is usually invalid by the time we reach here, but
2083 * finish_wait() does not dereference it in that case;
2084 * though i_lock needed lest racing with wake_up_all().
2085 */
2086 spin_lock(&inode->i_lock);
2087 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2088 spin_unlock(&inode->i_lock);
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002089
2090 if (fpin)
2091 fput(fpin);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002092 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002093 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07002094 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002095 }
2096
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002097 sgp = SGP_CACHE;
Michal Hocko18600332017-07-10 15:48:02 -07002098
2099 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2100 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002101 sgp = SGP_NOHUGE;
Michal Hocko18600332017-07-10 15:48:02 -07002102 else if (vma->vm_flags & VM_HUGEPAGE)
2103 sgp = SGP_HUGE;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002104
Souptick Joarder20acce62018-06-07 17:09:17 -07002105 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08002106 gfp, vma, vmf, &ret);
Souptick Joarder20acce62018-06-07 17:09:17 -07002107 if (err)
2108 return vmf_error(err);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002109 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110}
2111
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002112unsigned long shmem_get_unmapped_area(struct file *file,
2113 unsigned long uaddr, unsigned long len,
2114 unsigned long pgoff, unsigned long flags)
2115{
2116 unsigned long (*get_area)(struct file *,
2117 unsigned long, unsigned long, unsigned long, unsigned long);
2118 unsigned long addr;
2119 unsigned long offset;
2120 unsigned long inflated_len;
2121 unsigned long inflated_addr;
2122 unsigned long inflated_offset;
2123
2124 if (len > TASK_SIZE)
2125 return -ENOMEM;
2126
2127 get_area = current->mm->get_unmapped_area;
2128 addr = get_area(file, uaddr, len, pgoff, flags);
2129
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002130 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002131 return addr;
2132 if (IS_ERR_VALUE(addr))
2133 return addr;
2134 if (addr & ~PAGE_MASK)
2135 return addr;
2136 if (addr > TASK_SIZE - len)
2137 return addr;
2138
2139 if (shmem_huge == SHMEM_HUGE_DENY)
2140 return addr;
2141 if (len < HPAGE_PMD_SIZE)
2142 return addr;
2143 if (flags & MAP_FIXED)
2144 return addr;
2145 /*
2146 * Our priority is to support MAP_SHARED mapped hugely;
2147 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002148 * But if caller specified an address hint and we allocated area there
2149 * successfully, respect that as before.
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002150 */
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002151 if (uaddr == addr)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002152 return addr;
2153
2154 if (shmem_huge != SHMEM_HUGE_FORCE) {
2155 struct super_block *sb;
2156
2157 if (file) {
2158 VM_BUG_ON(file->f_op != &shmem_file_operations);
2159 sb = file_inode(file)->i_sb;
2160 } else {
2161 /*
2162 * Called directly from mm/mmap.c, or drivers/char/mem.c
2163 * for "/dev/zero", to create a shared anonymous object.
2164 */
2165 if (IS_ERR(shm_mnt))
2166 return addr;
2167 sb = shm_mnt->mnt_sb;
2168 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002169 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002170 return addr;
2171 }
2172
2173 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2174 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2175 return addr;
2176 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2177 return addr;
2178
2179 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2180 if (inflated_len > TASK_SIZE)
2181 return addr;
2182 if (inflated_len < len)
2183 return addr;
2184
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002185 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002186 if (IS_ERR_VALUE(inflated_addr))
2187 return addr;
2188 if (inflated_addr & ~PAGE_MASK)
2189 return addr;
2190
2191 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2192 inflated_addr += offset - inflated_offset;
2193 if (inflated_offset > offset)
2194 inflated_addr += HPAGE_PMD_SIZE;
2195
2196 if (inflated_addr > TASK_SIZE - len)
2197 return addr;
2198 return inflated_addr;
2199}
2200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002202static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203{
Al Viro496ad9a2013-01-23 17:07:38 -05002204 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002205 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206}
2207
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002208static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2209 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210{
Al Viro496ad9a2013-01-23 17:07:38 -05002211 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002212 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002214 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2215 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216}
2217#endif
2218
2219int shmem_lock(struct file *file, int lock, struct user_struct *user)
2220{
Al Viro496ad9a2013-01-23 17:07:38 -05002221 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 struct shmem_inode_info *info = SHMEM_I(inode);
2223 int retval = -ENOMEM;
2224
Hugh Dickinsea0dfeb2020-04-20 18:14:14 -07002225 /*
2226 * What serializes the accesses to info->flags?
2227 * ipc_lock_object() when called from shmctl_do_lock(),
2228 * no serialization needed when called from shm_destroy().
2229 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 if (lock && !(info->flags & VM_LOCKED)) {
2231 if (!user_shm_lock(inode->i_size, user))
2232 goto out_nomem;
2233 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002234 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
2236 if (!lock && (info->flags & VM_LOCKED) && user) {
2237 user_shm_unlock(inode->i_size, user);
2238 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002239 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 }
2241 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002242
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243out_nomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 return retval;
2245}
2246
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002247static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248{
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002249 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2250
2251 if (info->seals & F_SEAL_FUTURE_WRITE) {
2252 /*
2253 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2254 * "future write" seal active.
2255 */
2256 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2257 return -EPERM;
2258
2259 /*
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002260 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2261 * MAP_SHARED and read-only, take care to not allow mprotect to
2262 * revert protections on such mappings. Do this only for shared
2263 * mappings. For private mappings, don't need to mask
2264 * VM_MAYWRITE as we still want them to be COW-writable.
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002265 */
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002266 if (vma->vm_flags & VM_SHARED)
2267 vma->vm_flags &= ~(VM_MAYWRITE);
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002268 }
2269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 file_accessed(file);
2271 vma->vm_ops = &shmem_vm_ops;
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002272 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002273 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2274 (vma->vm_end & HPAGE_PMD_MASK)) {
2275 khugepaged_enter(vma, vma->vm_flags);
2276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 return 0;
2278}
2279
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002280static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002281 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282{
2283 struct inode *inode;
2284 struct shmem_inode_info *info;
2285 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -07002286 ino_t ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
Chris Downe809d5f2020-08-06 23:20:20 -07002288 if (shmem_reserve_inode(sb, &ino))
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002289 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 inode = new_inode(sb);
2292 if (inode) {
Chris Downe809d5f2020-08-06 23:20:20 -07002293 inode->i_ino = ino;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002294 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002296 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
Arnd Bergmann46c9a942018-08-17 15:45:09 -07002297 inode->i_generation = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 info = SHMEM_I(inode);
2299 memset(info, 0, (char *)inode - (char *)info);
2300 spin_lock_init(&info->lock);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07002301 atomic_set(&info->stop_eviction, 0);
David Herrmann40e041a2014-08-08 14:25:27 -07002302 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002303 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002304 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002306 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002307 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309 switch (mode & S_IFMT) {
2310 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002311 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 init_special_inode(inode, mode, dev);
2313 break;
2314 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002315 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 inode->i_op = &shmem_inode_operations;
2317 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002318 mpol_shared_policy_init(&info->policy,
2319 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 break;
2321 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002322 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 /* Some things misbehave if size == 0 on a directory */
2324 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2325 inode->i_op = &shmem_dir_inode_operations;
2326 inode->i_fop = &simple_dir_operations;
2327 break;
2328 case S_IFLNK:
2329 /*
2330 * Must not load anything in the rbtree,
2331 * mpol_free_shared_policy will not be called.
2332 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002333 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 break;
2335 }
Joel Fernandes (Google)b45d71f2018-09-20 12:22:39 -07002336
2337 lockdep_annotate_inode_mutex_key(inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002338 } else
2339 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 return inode;
2341}
2342
Johannes Weiner0cd61442014-04-03 14:47:46 -07002343bool shmem_mapping(struct address_space *mapping)
2344{
Hugh Dickinsf8005452017-02-22 15:41:42 -08002345 return mapping->a_ops == &shmem_aops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002346}
2347
Mike Rapoport8d103962017-09-06 16:23:02 -07002348static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2349 pmd_t *dst_pmd,
2350 struct vm_area_struct *dst_vma,
2351 unsigned long dst_addr,
2352 unsigned long src_addr,
2353 bool zeropage,
2354 struct page **pagep)
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002355{
2356 struct inode *inode = file_inode(dst_vma->vm_file);
2357 struct shmem_inode_info *info = SHMEM_I(inode);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002358 struct address_space *mapping = inode->i_mapping;
2359 gfp_t gfp = mapping_gfp_mask(mapping);
2360 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002361 spinlock_t *ptl;
2362 void *page_kaddr;
2363 struct page *page;
2364 pte_t _dst_pte, *dst_pte;
2365 int ret;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002366 pgoff_t offset, max_off;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002367
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002368 ret = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07002369 if (!shmem_inode_acct_block(inode, 1))
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002370 goto out;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002371
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002372 if (!*pagep) {
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002373 page = shmem_alloc_page(gfp, info, pgoff);
2374 if (!page)
Mike Rapoport0f079692017-09-06 16:22:59 -07002375 goto out_unacct_blocks;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002376
Mike Rapoport8d103962017-09-06 16:23:02 -07002377 if (!zeropage) { /* mcopy_atomic */
2378 page_kaddr = kmap_atomic(page);
2379 ret = copy_from_user(page_kaddr,
2380 (const void __user *)src_addr,
2381 PAGE_SIZE);
2382 kunmap_atomic(page_kaddr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002383
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002384 /* fallback to copy_from_user outside mmap_lock */
Mike Rapoport8d103962017-09-06 16:23:02 -07002385 if (unlikely(ret)) {
2386 *pagep = page;
2387 shmem_inode_unacct_blocks(inode, 1);
2388 /* don't free the page */
Andrea Arcangeli9e368252018-11-30 14:09:25 -08002389 return -ENOENT;
Mike Rapoport8d103962017-09-06 16:23:02 -07002390 }
2391 } else { /* mfill_zeropage_atomic */
2392 clear_highpage(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002393 }
2394 } else {
2395 page = *pagep;
2396 *pagep = NULL;
2397 }
2398
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002399 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2400 __SetPageLocked(page);
2401 __SetPageSwapBacked(page);
Andrea Arcangelia425d352017-02-22 15:43:52 -08002402 __SetPageUptodate(page);
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002403
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002404 ret = -EFAULT;
2405 offset = linear_page_index(dst_vma, dst_addr);
2406 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2407 if (unlikely(offset >= max_off))
2408 goto out_release;
2409
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002410 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2411 gfp & GFP_RECLAIM_MASK, dst_mm);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002412 if (ret)
2413 goto out_release;
2414
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002415 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2416 if (dst_vma->vm_flags & VM_WRITE)
2417 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002418 else {
2419 /*
2420 * We don't set the pte dirty if the vma has no
2421 * VM_WRITE permission, so mark the page dirty or it
2422 * could be freed from under us. We could do it
2423 * unconditionally before unlock_page(), but doing it
2424 * only if VM_WRITE is not set is faster.
2425 */
2426 set_page_dirty(page);
2427 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002428
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002429 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002430
2431 ret = -EFAULT;
2432 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2433 if (unlikely(offset >= max_off))
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002434 goto out_release_unlock;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002435
2436 ret = -EEXIST;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002437 if (!pte_none(*dst_pte))
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002438 goto out_release_unlock;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002439
Johannes Weiner6058eae2020-06-03 16:02:40 -07002440 lru_cache_add(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002441
Yang Shi94b7cc02020-04-20 18:14:17 -07002442 spin_lock_irq(&info->lock);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002443 info->alloced++;
2444 inode->i_blocks += BLOCKS_PER_PAGE;
2445 shmem_recalc_inode(inode);
Yang Shi94b7cc02020-04-20 18:14:17 -07002446 spin_unlock_irq(&info->lock);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002447
2448 inc_mm_counter(dst_mm, mm_counter_file(page));
2449 page_add_file_rmap(page, false);
2450 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2451
2452 /* No need to invalidate - it was non-present before */
2453 update_mmu_cache(dst_vma, dst_addr, dst_pte);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002454 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002455 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002456 ret = 0;
2457out:
2458 return ret;
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002459out_release_unlock:
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002460 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002461 ClearPageDirty(page);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002462 delete_from_page_cache(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002463out_release:
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002464 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002465 put_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002466out_unacct_blocks:
Mike Rapoport0f079692017-09-06 16:22:59 -07002467 shmem_inode_unacct_blocks(inode, 1);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002468 goto out;
2469}
2470
Mike Rapoport8d103962017-09-06 16:23:02 -07002471int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2472 pmd_t *dst_pmd,
2473 struct vm_area_struct *dst_vma,
2474 unsigned long dst_addr,
2475 unsigned long src_addr,
2476 struct page **pagep)
2477{
2478 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2479 dst_addr, src_addr, false, pagep);
2480}
2481
2482int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2483 pmd_t *dst_pmd,
2484 struct vm_area_struct *dst_vma,
2485 unsigned long dst_addr)
2486{
2487 struct page *page = NULL;
2488
2489 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2490 dst_addr, 0, true, &page);
2491}
2492
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002494static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002495static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002497#ifdef CONFIG_TMPFS_XATTR
2498static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2499#else
2500#define shmem_initxattrs NULL
2501#endif
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002504shmem_write_begin(struct file *file, struct address_space *mapping,
2505 loff_t pos, unsigned len, unsigned flags,
2506 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507{
Nick Piggin800d15a2007-10-16 01:25:03 -07002508 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002509 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002510 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002511
2512 /* i_mutex is held by caller */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002513 if (unlikely(info->seals & (F_SEAL_GROW |
2514 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2515 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
David Herrmann40e041a2014-08-08 14:25:27 -07002516 return -EPERM;
2517 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2518 return -EPERM;
2519 }
2520
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002521 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002522}
2523
2524static int
2525shmem_write_end(struct file *file, struct address_space *mapping,
2526 loff_t pos, unsigned len, unsigned copied,
2527 struct page *page, void *fsdata)
2528{
2529 struct inode *inode = mapping->host;
2530
Hugh Dickinsd3602442008-02-04 22:28:44 -08002531 if (pos + copied > inode->i_size)
2532 i_size_write(inode, pos + copied);
2533
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002534 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002535 struct page *head = compound_head(page);
2536 if (PageTransCompound(page)) {
2537 int i;
2538
2539 for (i = 0; i < HPAGE_PMD_NR; i++) {
2540 if (head + i == page)
2541 continue;
2542 clear_highpage(head + i);
2543 flush_dcache_page(head + i);
2544 }
2545 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002546 if (copied < PAGE_SIZE) {
2547 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002548 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002549 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002550 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002551 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002552 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002553 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002554 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002555 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002556
Nick Piggin800d15a2007-10-16 01:25:03 -07002557 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558}
2559
Al Viro2ba5bbe2014-04-02 20:00:02 -04002560static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561{
Al Viro6e58e792014-02-03 17:07:03 -05002562 struct file *file = iocb->ki_filp;
2563 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002565 pgoff_t index;
2566 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002567 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002568 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002569 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002570 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002571
2572 /*
2573 * Might this read be for a stacking filesystem? Then when reading
2574 * holes of a sparse file, we actually need to allocate those pages,
2575 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2576 */
Al Viro777eda22014-12-17 04:46:46 -05002577 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002578 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002580 index = *ppos >> PAGE_SHIFT;
2581 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
2583 for (;;) {
2584 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002585 pgoff_t end_index;
2586 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 loff_t i_size = i_size_read(inode);
2588
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002589 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 if (index > end_index)
2591 break;
2592 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002593 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 if (nr <= offset)
2595 break;
2596 }
2597
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002598 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002599 if (error) {
2600 if (error == -EINVAL)
2601 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 break;
2603 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002604 if (page) {
2605 if (sgp == SGP_CACHE)
2606 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002607 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
2610 /*
2611 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002612 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002614 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002616 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002618 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 if (nr <= offset) {
2620 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002621 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 break;
2623 }
2624 }
2625 nr -= offset;
2626
2627 if (page) {
2628 /*
2629 * If users can be writing to this page using arbitrary
2630 * virtual addresses, take care about potential aliasing
2631 * before reading the page on the kernel side.
2632 */
2633 if (mapping_writably_mapped(mapping))
2634 flush_dcache_page(page);
2635 /*
2636 * Mark the page accessed if we read the beginning.
2637 */
2638 if (!offset)
2639 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002640 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002642 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
2645 /*
2646 * Ok, we have the page, and it's up-to-date, so
2647 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002649 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002650 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002652 index += offset >> PAGE_SHIFT;
2653 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002655 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002656 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 break;
Al Viro6e58e792014-02-03 17:07:03 -05002658 if (ret < nr) {
2659 error = -EFAULT;
2660 break;
2661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 cond_resched();
2663 }
2664
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002665 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002666 file_accessed(file);
2667 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668}
2669
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002670/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002671 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002672 */
2673static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002674 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002675{
2676 struct page *page;
2677 struct pagevec pvec;
2678 pgoff_t indices[PAGEVEC_SIZE];
2679 bool done = false;
2680 int i;
2681
Mel Gorman86679822017-11-15 17:37:52 -08002682 pagevec_init(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002683 pvec.nr = 1; /* start small: we may be there already */
2684 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002685 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002686 pvec.nr, pvec.pages, indices);
2687 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002688 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002689 index = end;
2690 break;
2691 }
2692 for (i = 0; i < pvec.nr; i++, index++) {
2693 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002694 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002695 done = true;
2696 break;
2697 }
2698 index = indices[i];
2699 }
2700 page = pvec.pages[i];
Matthew Wilcox3159f942017-11-03 13:30:42 -04002701 if (page && !xa_is_value(page)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002702 if (!PageUptodate(page))
2703 page = NULL;
2704 }
2705 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002706 (page && whence == SEEK_DATA) ||
2707 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002708 done = true;
2709 break;
2710 }
2711 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002712 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002713 pagevec_release(&pvec);
2714 pvec.nr = PAGEVEC_SIZE;
2715 cond_resched();
2716 }
2717 return index;
2718}
2719
Andrew Morton965c8e52012-12-17 15:59:39 -08002720static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002721{
2722 struct address_space *mapping = file->f_mapping;
2723 struct inode *inode = mapping->host;
2724 pgoff_t start, end;
2725 loff_t new_offset;
2726
Andrew Morton965c8e52012-12-17 15:59:39 -08002727 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2728 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002729 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002730 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002731 /* We're holding i_mutex so we can access i_size directly */
2732
Yufen Yu1a413642018-11-16 15:08:39 -08002733 if (offset < 0 || offset >= inode->i_size)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002734 offset = -ENXIO;
2735 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002736 start = offset >> PAGE_SHIFT;
2737 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002738 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002739 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002740 if (new_offset > offset) {
2741 if (new_offset < inode->i_size)
2742 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002743 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002744 offset = -ENXIO;
2745 else
2746 offset = inode->i_size;
2747 }
2748 }
2749
Hugh Dickins387aae62013-08-04 11:30:25 -07002750 if (offset >= 0)
2751 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002752 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002753 return offset;
2754}
2755
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002756static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2757 loff_t len)
2758{
Al Viro496ad9a2013-01-23 17:07:38 -05002759 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002760 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002761 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002762 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002763 pgoff_t start, index, end;
2764 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002765
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002766 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2767 return -EOPNOTSUPP;
2768
Al Viro59551022016-01-22 15:40:57 -05002769 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002770
2771 if (mode & FALLOC_FL_PUNCH_HOLE) {
2772 struct address_space *mapping = file->f_mapping;
2773 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2774 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002775 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002776
David Herrmann40e041a2014-08-08 14:25:27 -07002777 /* protected by i_mutex */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002778 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
David Herrmann40e041a2014-08-08 14:25:27 -07002779 error = -EPERM;
2780 goto out;
2781 }
2782
Hugh Dickins8e205f72014-07-23 14:00:10 -07002783 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junaa71ecd2019-11-30 17:58:11 -08002784 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002785 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2786 spin_lock(&inode->i_lock);
2787 inode->i_private = &shmem_falloc;
2788 spin_unlock(&inode->i_lock);
2789
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002790 if ((u64)unmap_end > (u64)unmap_start)
2791 unmap_mapping_range(mapping, unmap_start,
2792 1 + unmap_end - unmap_start, 0);
2793 shmem_truncate_range(inode, offset, offset + len - 1);
2794 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002795
2796 spin_lock(&inode->i_lock);
2797 inode->i_private = NULL;
2798 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02002799 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002800 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002801 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002802 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002803 }
2804
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002805 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2806 error = inode_newsize_ok(inode, offset + len);
2807 if (error)
2808 goto out;
2809
David Herrmann40e041a2014-08-08 14:25:27 -07002810 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2811 error = -EPERM;
2812 goto out;
2813 }
2814
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002815 start = offset >> PAGE_SHIFT;
2816 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002817 /* Try to avoid a swapstorm if len is impossible to satisfy */
2818 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2819 error = -ENOSPC;
2820 goto out;
2821 }
2822
Hugh Dickins8e205f72014-07-23 14:00:10 -07002823 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002824 shmem_falloc.start = start;
2825 shmem_falloc.next = start;
2826 shmem_falloc.nr_falloced = 0;
2827 shmem_falloc.nr_unswapped = 0;
2828 spin_lock(&inode->i_lock);
2829 inode->i_private = &shmem_falloc;
2830 spin_unlock(&inode->i_lock);
2831
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002832 for (index = start; index < end; index++) {
2833 struct page *page;
2834
2835 /*
2836 * Good, the fallocate(2) manpage permits EINTR: we may have
2837 * been interrupted because we are using up too much memory.
2838 */
2839 if (signal_pending(current))
2840 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002841 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2842 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002843 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002844 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002845 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002846 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002847 if (index > start) {
2848 shmem_undo_range(inode,
2849 (loff_t)start << PAGE_SHIFT,
2850 ((loff_t)index << PAGE_SHIFT) - 1, true);
2851 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002852 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002853 }
2854
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002855 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002856 * Inform shmem_writepage() how far we have reached.
2857 * No need for lock or barrier: we have the page lock.
2858 */
2859 shmem_falloc.next++;
2860 if (!PageUptodate(page))
2861 shmem_falloc.nr_falloced++;
2862
2863 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002864 * If !PageUptodate, leave it that way so that freeable pages
2865 * can be recognized if we need to rollback on error later.
2866 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002867 * than free the pages we are allocating (and SGP_CACHE pages
2868 * might still be clean: we now need to mark those dirty too).
2869 */
2870 set_page_dirty(page);
2871 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002872 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002873 cond_resched();
2874 }
2875
2876 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2877 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002878 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002879undone:
2880 spin_lock(&inode->i_lock);
2881 inode->i_private = NULL;
2882 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002883out:
Al Viro59551022016-01-22 15:40:57 -05002884 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002885 return error;
2886}
2887
David Howells726c3342006-06-23 02:02:58 -07002888static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889{
David Howells726c3342006-06-23 02:02:58 -07002890 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
2892 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002893 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002895 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002897 buf->f_bavail =
2898 buf->f_bfree = sbinfo->max_blocks -
2899 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002900 }
2901 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 buf->f_files = sbinfo->max_inodes;
2903 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 }
2905 /* else leave those fields 0 like simple_statfs */
2906 return 0;
2907}
2908
2909/*
2910 * File creation. Allocate an inode, and we're done..
2911 */
2912static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002913shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002915 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 int error = -ENOSPC;
2917
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002918 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002920 error = simple_acl_create(dir, inode);
2921 if (error)
2922 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002923 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002924 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002925 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002926 if (error && error != -EOPNOTSUPP)
2927 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002928
Al Viro718deb62009-12-16 19:35:36 -05002929 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002931 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 d_instantiate(dentry, inode);
2933 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 }
2935 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002936out_iput:
2937 iput(inode);
2938 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939}
2940
Al Viro60545d02013-06-07 01:20:27 -04002941static int
2942shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2943{
2944 struct inode *inode;
2945 int error = -ENOSPC;
2946
2947 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2948 if (inode) {
2949 error = security_inode_init_security(inode, dir,
2950 NULL,
2951 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002952 if (error && error != -EOPNOTSUPP)
2953 goto out_iput;
2954 error = simple_acl_create(dir, inode);
2955 if (error)
2956 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002957 d_tmpfile(dentry, inode);
2958 }
2959 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002960out_iput:
2961 iput(inode);
2962 return error;
Al Viro60545d02013-06-07 01:20:27 -04002963}
2964
Al Viro18bb1db2011-07-26 01:41:39 -04002965static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966{
2967 int error;
2968
2969 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2970 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002971 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 return 0;
2973}
2974
Al Viro4acdaf22011-07-26 01:42:34 -04002975static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002976 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977{
2978 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2979}
2980
2981/*
2982 * Link a file..
2983 */
2984static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2985{
David Howells75c3cfa2015-03-17 22:26:12 +00002986 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong29b00e62019-02-22 22:35:32 -08002987 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
2989 /*
2990 * No ordinary (disk based) filesystem counts links as inodes;
2991 * but each new link needs a new dentry, pinning lowmem, and
2992 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wong1062af92019-02-21 08:48:09 -08002993 * But if an O_TMPFILE file is linked into the tmpfs, the
2994 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 */
Darrick J. Wong1062af92019-02-21 08:48:09 -08002996 if (inode->i_nlink) {
Chris Downe809d5f2020-08-06 23:20:20 -07002997 ret = shmem_reserve_inode(inode->i_sb, NULL);
Darrick J. Wong1062af92019-02-21 08:48:09 -08002998 if (ret)
2999 goto out;
3000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001
3002 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003003 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07003004 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04003005 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 dget(dentry); /* Extra pinning count for the created dentry */
3007 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003008out:
3009 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010}
3011
3012static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3013{
David Howells75c3cfa2015-03-17 22:26:12 +00003014 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003016 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3017 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018
3019 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003020 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003021 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 dput(dentry); /* Undo the count from "create" - this does all the work */
3023 return 0;
3024}
3025
3026static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3027{
3028 if (!simple_empty(dentry))
3029 return -ENOTEMPTY;
3030
David Howells75c3cfa2015-03-17 22:26:12 +00003031 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003032 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 return shmem_unlink(dir, dentry);
3034}
3035
Miklos Szeredi37456772014-07-23 15:15:34 +02003036static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3037{
David Howellse36cb0b2015-01-29 12:02:35 +00003038 bool old_is_dir = d_is_dir(old_dentry);
3039 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003040
3041 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3042 if (old_is_dir) {
3043 drop_nlink(old_dir);
3044 inc_nlink(new_dir);
3045 } else {
3046 drop_nlink(new_dir);
3047 inc_nlink(old_dir);
3048 }
3049 }
3050 old_dir->i_ctime = old_dir->i_mtime =
3051 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003052 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003053 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02003054
3055 return 0;
3056}
3057
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003058static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3059{
3060 struct dentry *whiteout;
3061 int error;
3062
3063 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3064 if (!whiteout)
3065 return -ENOMEM;
3066
3067 error = shmem_mknod(old_dir, whiteout,
3068 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3069 dput(whiteout);
3070 if (error)
3071 return error;
3072
3073 /*
3074 * Cheat and hash the whiteout while the old dentry is still in
3075 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3076 *
3077 * d_lookup() will consistently find one of them at this point,
3078 * not sure which one, but that isn't even important.
3079 */
3080 d_rehash(whiteout);
3081 return 0;
3082}
3083
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084/*
3085 * The VFS layer already does all the dentry stuff for rename,
3086 * we just have to decrement the usage count for the target if
3087 * it exists so that the VFS layer correctly free's it when it
3088 * gets overwritten.
3089 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003090static 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 -07003091{
David Howells75c3cfa2015-03-17 22:26:12 +00003092 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 int they_are_dirs = S_ISDIR(inode->i_mode);
3094
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003095 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003096 return -EINVAL;
3097
Miklos Szeredi37456772014-07-23 15:15:34 +02003098 if (flags & RENAME_EXCHANGE)
3099 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 if (!simple_empty(new_dentry))
3102 return -ENOTEMPTY;
3103
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003104 if (flags & RENAME_WHITEOUT) {
3105 int error;
3106
3107 error = shmem_whiteout(old_dir, old_dentry);
3108 if (error)
3109 return error;
3110 }
3111
David Howells75c3cfa2015-03-17 22:26:12 +00003112 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003114 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003115 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003116 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003119 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003120 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 }
3122
3123 old_dir->i_size -= BOGO_DIRENT_SIZE;
3124 new_dir->i_size += BOGO_DIRENT_SIZE;
3125 old_dir->i_ctime = old_dir->i_mtime =
3126 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003127 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 return 0;
3129}
3130
3131static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3132{
3133 int error;
3134 int len;
3135 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003136 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137
3138 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003139 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 return -ENAMETOOLONG;
3141
Joe Perches0825a6f2018-06-14 15:27:58 -07003142 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3143 VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 if (!inode)
3145 return -ENOSPC;
3146
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003147 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003148 shmem_initxattrs, NULL);
Mateusz Nosek343c3d72020-04-06 20:07:54 -07003149 if (error && error != -EOPNOTSUPP) {
3150 iput(inode);
3151 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003152 }
3153
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003155 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003156 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3157 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003158 iput(inode);
3159 return -ENOMEM;
3160 }
3161 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003163 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003164 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 if (error) {
3166 iput(inode);
3167 return error;
3168 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003169 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003171 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003172 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003174 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003175 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003178 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 d_instantiate(dentry, inode);
3180 dget(dentry);
3181 return 0;
3182}
3183
Al Virofceef392015-12-29 15:58:39 -05003184static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
Al Virofceef392015-12-29 15:58:39 -05003186 mark_page_accessed(arg);
3187 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188}
3189
Al Viro6b255392015-11-17 10:20:54 -05003190static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003191 struct inode *inode,
3192 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003195 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003196 if (!dentry) {
3197 page = find_get_page(inode->i_mapping, 0);
3198 if (!page)
3199 return ERR_PTR(-ECHILD);
3200 if (!PageUptodate(page)) {
3201 put_page(page);
3202 return ERR_PTR(-ECHILD);
3203 }
3204 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003205 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003206 if (error)
3207 return ERR_PTR(error);
3208 unlock_page(page);
3209 }
Al Virofceef392015-12-29 15:58:39 -05003210 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003211 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212}
3213
Eric Parisb09e0fa2011-05-24 17:12:39 -07003214#ifdef CONFIG_TMPFS_XATTR
3215/*
3216 * Superblocks without xattr inode operations may get some security.* xattr
3217 * support from the LSM "for free". As soon as we have any other xattrs
3218 * like ACLs, we also need to implement the security.* handlers at
3219 * filesystem level, though.
3220 */
3221
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003222/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003223 * Callback for security_inode_init_security() for acquiring xattrs.
3224 */
3225static int shmem_initxattrs(struct inode *inode,
3226 const struct xattr *xattr_array,
3227 void *fs_info)
3228{
3229 struct shmem_inode_info *info = SHMEM_I(inode);
3230 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003231 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003232 size_t len;
3233
3234 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003235 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003236 if (!new_xattr)
3237 return -ENOMEM;
3238
3239 len = strlen(xattr->name) + 1;
3240 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3241 GFP_KERNEL);
3242 if (!new_xattr->name) {
Chengguang Xu3bef7352020-07-23 21:15:14 -07003243 kvfree(new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003244 return -ENOMEM;
3245 }
3246
3247 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3248 XATTR_SECURITY_PREFIX_LEN);
3249 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3250 xattr->name, len);
3251
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003252 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003253 }
3254
3255 return 0;
3256}
3257
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003258static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003259 struct dentry *unused, struct inode *inode,
3260 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003261{
Al Virob2968212016-04-10 20:48:24 -04003262 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003263
3264 name = xattr_full_name(handler, name);
3265 return simple_xattr_get(&info->xattrs, name, buffer, size);
3266}
3267
3268static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003269 struct dentry *unused, struct inode *inode,
3270 const char *name, const void *value,
3271 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003272{
Al Viro59301222016-05-27 10:19:30 -04003273 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003274
3275 name = xattr_full_name(handler, name);
Daniel Xua46a2292020-03-12 13:03:15 -07003276 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003277}
3278
3279static const struct xattr_handler shmem_security_xattr_handler = {
3280 .prefix = XATTR_SECURITY_PREFIX,
3281 .get = shmem_xattr_handler_get,
3282 .set = shmem_xattr_handler_set,
3283};
3284
3285static const struct xattr_handler shmem_trusted_xattr_handler = {
3286 .prefix = XATTR_TRUSTED_PREFIX,
3287 .get = shmem_xattr_handler_get,
3288 .set = shmem_xattr_handler_set,
3289};
3290
Eric Parisb09e0fa2011-05-24 17:12:39 -07003291static const struct xattr_handler *shmem_xattr_handlers[] = {
3292#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003293 &posix_acl_access_xattr_handler,
3294 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003295#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003296 &shmem_security_xattr_handler,
3297 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003298 NULL
3299};
3300
Eric Parisb09e0fa2011-05-24 17:12:39 -07003301static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3302{
David Howells75c3cfa2015-03-17 22:26:12 +00003303 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003304 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003305}
3306#endif /* CONFIG_TMPFS_XATTR */
3307
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003308static const struct inode_operations shmem_short_symlink_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003309 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003310#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003311 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003312#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313};
3314
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003315static const struct inode_operations shmem_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003316 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003317#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003318 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003319#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003320};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003321
David M. Grimes91828a42006-10-17 00:09:45 -07003322static struct dentry *shmem_get_parent(struct dentry *child)
3323{
3324 return ERR_PTR(-ESTALE);
3325}
3326
3327static int shmem_match(struct inode *ino, void *vfh)
3328{
3329 __u32 *fh = vfh;
3330 __u64 inum = fh[2];
3331 inum = (inum << 32) | fh[1];
3332 return ino->i_ino == inum && fh[0] == ino->i_generation;
3333}
3334
Amir Goldstein12ba7802018-06-07 17:07:15 -07003335/* Find any alias of inode, but prefer a hashed alias */
3336static struct dentry *shmem_find_alias(struct inode *inode)
3337{
3338 struct dentry *alias = d_find_alias(inode);
3339
3340 return alias ?: d_find_any_alias(inode);
3341}
3342
3343
Christoph Hellwig480b1162007-10-21 16:42:13 -07003344static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3345 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003346{
David M. Grimes91828a42006-10-17 00:09:45 -07003347 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003348 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003349 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003350
Christoph Hellwig480b1162007-10-21 16:42:13 -07003351 if (fh_len < 3)
3352 return NULL;
3353
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003354 inum = fid->raw[2];
3355 inum = (inum << 32) | fid->raw[1];
3356
Christoph Hellwig480b1162007-10-21 16:42:13 -07003357 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3358 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003359 if (inode) {
Amir Goldstein12ba7802018-06-07 17:07:15 -07003360 dentry = shmem_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003361 iput(inode);
3362 }
3363
Christoph Hellwig480b1162007-10-21 16:42:13 -07003364 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003365}
3366
Al Virob0b03822012-04-02 14:34:06 -04003367static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3368 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003369{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303370 if (*len < 3) {
3371 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003372 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303373 }
David M. Grimes91828a42006-10-17 00:09:45 -07003374
Al Viro1d3382cb2010-10-23 15:19:20 -04003375 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003376 /* Unfortunately insert_inode_hash is not idempotent,
3377 * so as we hash inodes here rather than at creation
3378 * time, we need a lock to ensure we only try
3379 * to do it once
3380 */
3381 static DEFINE_SPINLOCK(lock);
3382 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003383 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003384 __insert_inode_hash(inode,
3385 inode->i_ino + inode->i_generation);
3386 spin_unlock(&lock);
3387 }
3388
3389 fh[0] = inode->i_generation;
3390 fh[1] = inode->i_ino;
3391 fh[2] = ((__u64)inode->i_ino) >> 32;
3392
3393 *len = 3;
3394 return 1;
3395}
3396
Christoph Hellwig39655162007-10-21 16:42:17 -07003397static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003398 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003399 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003400 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003401};
3402
Al Viro626c3922019-09-08 20:28:06 -04003403enum shmem_param {
3404 Opt_gid,
3405 Opt_huge,
3406 Opt_mode,
3407 Opt_mpol,
3408 Opt_nr_blocks,
3409 Opt_nr_inodes,
3410 Opt_size,
3411 Opt_uid,
Chris Downea3271f2020-08-06 23:20:25 -07003412 Opt_inode32,
3413 Opt_inode64,
Al Viro626c3922019-09-08 20:28:06 -04003414};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415
Al Viro5eede622019-12-16 13:33:32 -05003416static const struct constant_table shmem_param_enums_huge[] = {
Al Viro2710c957a2019-09-06 22:12:08 -04003417 {"never", SHMEM_HUGE_NEVER },
3418 {"always", SHMEM_HUGE_ALWAYS },
3419 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3420 {"advise", SHMEM_HUGE_ADVISE },
Al Viro2710c957a2019-09-06 22:12:08 -04003421 {}
3422};
3423
Al Virod7167b12019-09-07 07:23:15 -04003424const struct fs_parameter_spec shmem_fs_parameters[] = {
Al Viro626c3922019-09-08 20:28:06 -04003425 fsparam_u32 ("gid", Opt_gid),
Al Viro2710c957a2019-09-06 22:12:08 -04003426 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
Al Viro626c3922019-09-08 20:28:06 -04003427 fsparam_u32oct("mode", Opt_mode),
3428 fsparam_string("mpol", Opt_mpol),
3429 fsparam_string("nr_blocks", Opt_nr_blocks),
3430 fsparam_string("nr_inodes", Opt_nr_inodes),
3431 fsparam_string("size", Opt_size),
3432 fsparam_u32 ("uid", Opt_uid),
Chris Downea3271f2020-08-06 23:20:25 -07003433 fsparam_flag ("inode32", Opt_inode32),
3434 fsparam_flag ("inode64", Opt_inode64),
Al Viro626c3922019-09-08 20:28:06 -04003435 {}
3436};
3437
David Howellsf3235622019-03-25 16:38:31 +00003438static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
Al Viro626c3922019-09-08 20:28:06 -04003439{
David Howellsf3235622019-03-25 16:38:31 +00003440 struct shmem_options *ctx = fc->fs_private;
Al Viro626c3922019-09-08 20:28:06 -04003441 struct fs_parse_result result;
3442 unsigned long long size;
3443 char *rest;
3444 int opt;
3445
Al Virod7167b12019-09-07 07:23:15 -04003446 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
David Howellsf3235622019-03-25 16:38:31 +00003447 if (opt < 0)
Al Viro626c3922019-09-08 20:28:06 -04003448 return opt;
Al Viro626c3922019-09-08 20:28:06 -04003449
3450 switch (opt) {
3451 case Opt_size:
3452 size = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003453 if (*rest == '%') {
3454 size <<= PAGE_SHIFT;
3455 size *= totalram_pages();
3456 do_div(size, 100);
3457 rest++;
3458 }
3459 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003460 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003461 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3462 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003463 break;
3464 case Opt_nr_blocks:
3465 ctx->blocks = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003466 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003467 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003468 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003469 break;
3470 case Opt_nr_inodes:
3471 ctx->inodes = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003472 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003473 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003474 ctx->seen |= SHMEM_SEEN_INODES;
Al Viro626c3922019-09-08 20:28:06 -04003475 break;
3476 case Opt_mode:
3477 ctx->mode = result.uint_32 & 07777;
3478 break;
3479 case Opt_uid:
3480 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003481 if (!uid_valid(ctx->uid))
Al Viro626c3922019-09-08 20:28:06 -04003482 goto bad_value;
3483 break;
3484 case Opt_gid:
3485 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003486 if (!gid_valid(ctx->gid))
Al Viro626c3922019-09-08 20:28:06 -04003487 goto bad_value;
3488 break;
3489 case Opt_huge:
3490 ctx->huge = result.uint_32;
3491 if (ctx->huge != SHMEM_HUGE_NEVER &&
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003492 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Al Viro626c3922019-09-08 20:28:06 -04003493 has_transparent_hugepage()))
3494 goto unsupported_parameter;
Al Viroe04dc422019-09-08 19:20:12 -04003495 ctx->seen |= SHMEM_SEEN_HUGE;
Al Viro626c3922019-09-08 20:28:06 -04003496 break;
3497 case Opt_mpol:
3498 if (IS_ENABLED(CONFIG_NUMA)) {
3499 mpol_put(ctx->mpol);
3500 ctx->mpol = NULL;
3501 if (mpol_parse_str(param->string, &ctx->mpol))
3502 goto bad_value;
3503 break;
3504 }
3505 goto unsupported_parameter;
Chris Downea3271f2020-08-06 23:20:25 -07003506 case Opt_inode32:
3507 ctx->full_inums = false;
3508 ctx->seen |= SHMEM_SEEN_INUMS;
3509 break;
3510 case Opt_inode64:
3511 if (sizeof(ino_t) < 8) {
3512 return invalfc(fc,
3513 "Cannot use inode64 with <64bit inums in kernel\n");
3514 }
3515 ctx->full_inums = true;
3516 ctx->seen |= SHMEM_SEEN_INUMS;
3517 break;
Al Viroe04dc422019-09-08 19:20:12 -04003518 }
3519 return 0;
3520
Al Viro626c3922019-09-08 20:28:06 -04003521unsupported_parameter:
Al Virof35aa2b2019-12-21 21:35:55 -05003522 return invalfc(fc, "Unsupported parameter '%s'", param->key);
Al Viro626c3922019-09-08 20:28:06 -04003523bad_value:
Al Virof35aa2b2019-12-21 21:35:55 -05003524 return invalfc(fc, "Bad value for '%s'", param->key);
Al Viroe04dc422019-09-08 19:20:12 -04003525}
3526
David Howellsf3235622019-03-25 16:38:31 +00003527static int shmem_parse_options(struct fs_context *fc, void *data)
Al Viroe04dc422019-09-08 19:20:12 -04003528{
David Howellsf3235622019-03-25 16:38:31 +00003529 char *options = data;
3530
Al Viro33f37c62019-10-09 22:48:01 -04003531 if (options) {
3532 int err = security_sb_eat_lsm_opts(options, &fc->security);
3533 if (err)
3534 return err;
3535 }
3536
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003537 while (options != NULL) {
Al Viro626c3922019-09-08 20:28:06 -04003538 char *this_char = options;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003539 for (;;) {
3540 /*
3541 * NUL-terminate this option: unfortunately,
3542 * mount options form a comma-separated list,
3543 * but mpol's nodelist may also contain commas.
3544 */
3545 options = strchr(options, ',');
3546 if (options == NULL)
3547 break;
3548 options++;
3549 if (!isdigit(*options)) {
3550 options[-1] = '\0';
3551 break;
3552 }
3553 }
Al Viro626c3922019-09-08 20:28:06 -04003554 if (*this_char) {
3555 char *value = strchr(this_char,'=');
David Howellsf3235622019-03-25 16:38:31 +00003556 size_t len = 0;
Al Viro626c3922019-09-08 20:28:06 -04003557 int err;
3558
3559 if (value) {
3560 *value++ = '\0';
David Howellsf3235622019-03-25 16:38:31 +00003561 len = strlen(value);
Al Viro626c3922019-09-08 20:28:06 -04003562 }
David Howellsf3235622019-03-25 16:38:31 +00003563 err = vfs_parse_fs_string(fc, this_char, value, len);
3564 if (err < 0)
3565 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 }
3568 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569}
3570
David Howellsf3235622019-03-25 16:38:31 +00003571/*
3572 * Reconfigure a shmem filesystem.
3573 *
3574 * Note that we disallow change from limited->unlimited blocks/inodes while any
3575 * are in use; but we must separately disallow unlimited->limited, because in
3576 * that case we have no record of how much is already in use.
3577 */
3578static int shmem_reconfigure(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579{
David Howellsf3235622019-03-25 16:38:31 +00003580 struct shmem_options *ctx = fc->fs_private;
3581 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003582 unsigned long inodes;
David Howellsf3235622019-03-25 16:38:31 +00003583 const char *err;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003584
3585 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003586 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
David Howellsf3235622019-03-25 16:38:31 +00003587 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3588 if (!sbinfo->max_blocks) {
3589 err = "Cannot retroactively limit size";
Al Viro0b5071d2019-09-08 18:49:18 -04003590 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003591 }
Al Viro0b5071d2019-09-08 18:49:18 -04003592 if (percpu_counter_compare(&sbinfo->used_blocks,
David Howellsf3235622019-03-25 16:38:31 +00003593 ctx->blocks) > 0) {
3594 err = "Too small a size for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003595 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003596 }
Al Viro0b5071d2019-09-08 18:49:18 -04003597 }
David Howellsf3235622019-03-25 16:38:31 +00003598 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3599 if (!sbinfo->max_inodes) {
3600 err = "Cannot retroactively limit inodes";
Al Viro0b5071d2019-09-08 18:49:18 -04003601 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003602 }
3603 if (ctx->inodes < inodes) {
3604 err = "Too few inodes for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003605 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003606 }
Al Viro0b5071d2019-09-08 18:49:18 -04003607 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003608
Chris Downea3271f2020-08-06 23:20:25 -07003609 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3610 sbinfo->next_ino > UINT_MAX) {
3611 err = "Current inum too high to switch to 32-bit inums";
3612 goto out;
3613 }
3614
David Howellsf3235622019-03-25 16:38:31 +00003615 if (ctx->seen & SHMEM_SEEN_HUGE)
3616 sbinfo->huge = ctx->huge;
Chris Downea3271f2020-08-06 23:20:25 -07003617 if (ctx->seen & SHMEM_SEEN_INUMS)
3618 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00003619 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3620 sbinfo->max_blocks = ctx->blocks;
3621 if (ctx->seen & SHMEM_SEEN_INODES) {
3622 sbinfo->max_inodes = ctx->inodes;
3623 sbinfo->free_inodes = ctx->inodes - inodes;
Al Viro0b5071d2019-09-08 18:49:18 -04003624 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003625
Greg Thelen5f001102013-02-22 16:36:01 -08003626 /*
3627 * Preserve previous mempolicy unless mpol remount option was specified.
3628 */
David Howellsf3235622019-03-25 16:38:31 +00003629 if (ctx->mpol) {
Greg Thelen5f001102013-02-22 16:36:01 -08003630 mpol_put(sbinfo->mpol);
David Howellsf3235622019-03-25 16:38:31 +00003631 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3632 ctx->mpol = NULL;
Greg Thelen5f001102013-02-22 16:36:01 -08003633 }
David Howellsf3235622019-03-25 16:38:31 +00003634 spin_unlock(&sbinfo->stat_lock);
3635 return 0;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003636out:
3637 spin_unlock(&sbinfo->stat_lock);
Al Virof35aa2b2019-12-21 21:35:55 -05003638 return invalfc(fc, "%s", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003640
Al Viro34c80b12011-12-08 21:32:45 -05003641static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003642{
Al Viro34c80b12011-12-08 21:32:45 -05003643 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003644
3645 if (sbinfo->max_blocks != shmem_default_max_blocks())
3646 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003647 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003648 if (sbinfo->max_inodes != shmem_default_max_inodes())
3649 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
Joe Perches0825a6f2018-06-14 15:27:58 -07003650 if (sbinfo->mode != (0777 | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003651 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003652 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3653 seq_printf(seq, ",uid=%u",
3654 from_kuid_munged(&init_user_ns, sbinfo->uid));
3655 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3656 seq_printf(seq, ",gid=%u",
3657 from_kgid_munged(&init_user_ns, sbinfo->gid));
Chris Downea3271f2020-08-06 23:20:25 -07003658
3659 /*
3660 * Showing inode{64,32} might be useful even if it's the system default,
3661 * since then people don't have to resort to checking both here and
3662 * /proc/config.gz to confirm 64-bit inums were successfully applied
3663 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3664 *
3665 * We hide it when inode64 isn't the default and we are using 32-bit
3666 * inodes, since that probably just means the feature isn't even under
3667 * consideration.
3668 *
3669 * As such:
3670 *
3671 * +-----------------+-----------------+
3672 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3673 * +------------------+-----------------+-----------------+
3674 * | full_inums=true | show | show |
3675 * | full_inums=false | show | hide |
3676 * +------------------+-----------------+-----------------+
3677 *
3678 */
3679 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3680 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003681#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003682 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3683 if (sbinfo->huge)
3684 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3685#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003686 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003687 return 0;
3688}
David Herrmann9183df22014-08-08 14:25:29 -07003689
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003690#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
3692static void shmem_put_super(struct super_block *sb)
3693{
Hugh Dickins602586a2010-08-17 15:23:56 -07003694 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3695
Chris Downe809d5f2020-08-06 23:20:20 -07003696 free_percpu(sbinfo->ino_batch);
Hugh Dickins602586a2010-08-17 15:23:56 -07003697 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003698 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003699 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700 sb->s_fs_info = NULL;
3701}
3702
David Howellsf3235622019-03-25 16:38:31 +00003703static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704{
David Howellsf3235622019-03-25 16:38:31 +00003705 struct shmem_options *ctx = fc->fs_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003707 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003708 int err = -ENOMEM;
3709
3710 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003711 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003712 L1_CACHE_BYTES), GFP_KERNEL);
3713 if (!sbinfo)
3714 return -ENOMEM;
3715
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003716 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003718#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 /*
3720 * Per default we only allow half of the physical ram per
3721 * tmpfs instance, limiting inodes to one per page of lowmem;
3722 * but the internal instance is left unlimited.
3723 */
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003724 if (!(sb->s_flags & SB_KERNMOUNT)) {
David Howellsf3235622019-03-25 16:38:31 +00003725 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3726 ctx->blocks = shmem_default_max_blocks();
3727 if (!(ctx->seen & SHMEM_SEEN_INODES))
3728 ctx->inodes = shmem_default_max_inodes();
Chris Downea3271f2020-08-06 23:20:25 -07003729 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3730 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
Al Viroca4e0512013-08-31 12:57:10 -04003731 } else {
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003732 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733 }
David M. Grimes91828a42006-10-17 00:09:45 -07003734 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003735 sb->s_flags |= SB_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736#else
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003737 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738#endif
David Howellsf3235622019-03-25 16:38:31 +00003739 sbinfo->max_blocks = ctx->blocks;
3740 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
Chris Downe809d5f2020-08-06 23:20:20 -07003741 if (sb->s_flags & SB_KERNMOUNT) {
3742 sbinfo->ino_batch = alloc_percpu(ino_t);
3743 if (!sbinfo->ino_batch)
3744 goto failed;
3745 }
David Howellsf3235622019-03-25 16:38:31 +00003746 sbinfo->uid = ctx->uid;
3747 sbinfo->gid = ctx->gid;
Chris Downea3271f2020-08-06 23:20:25 -07003748 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00003749 sbinfo->mode = ctx->mode;
3750 sbinfo->huge = ctx->huge;
3751 sbinfo->mpol = ctx->mpol;
3752 ctx->mpol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003754 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003755 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003756 goto failed;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003757 spin_lock_init(&sbinfo->shrinklist_lock);
3758 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003759
Hugh Dickins285b2c42011-08-03 16:21:20 -07003760 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003761 sb->s_blocksize = PAGE_SIZE;
3762 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763 sb->s_magic = TMPFS_MAGIC;
3764 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003765 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003766#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003767 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003768#endif
3769#ifdef CONFIG_TMPFS_POSIX_ACL
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003770 sb->s_flags |= SB_POSIXACL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003771#endif
Amir Goldstein2b4db792017-05-18 15:29:33 +03003772 uuid_gen(&sb->s_uuid);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003773
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003774 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 if (!inode)
3776 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003777 inode->i_uid = sbinfo->uid;
3778 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003779 sb->s_root = d_make_root(inode);
3780 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003781 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782 return 0;
3783
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784failed:
3785 shmem_put_super(sb);
3786 return err;
3787}
3788
David Howellsf3235622019-03-25 16:38:31 +00003789static int shmem_get_tree(struct fs_context *fc)
3790{
3791 return get_tree_nodev(fc, shmem_fill_super);
3792}
3793
3794static void shmem_free_fc(struct fs_context *fc)
3795{
3796 struct shmem_options *ctx = fc->fs_private;
3797
3798 if (ctx) {
3799 mpol_put(ctx->mpol);
3800 kfree(ctx);
3801 }
3802}
3803
3804static const struct fs_context_operations shmem_fs_context_ops = {
3805 .free = shmem_free_fc,
3806 .get_tree = shmem_get_tree,
3807#ifdef CONFIG_TMPFS
3808 .parse_monolithic = shmem_parse_options,
3809 .parse_param = shmem_parse_one,
3810 .reconfigure = shmem_reconfigure,
3811#endif
3812};
3813
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003814static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815
3816static struct inode *shmem_alloc_inode(struct super_block *sb)
3817{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003818 struct shmem_inode_info *info;
3819 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3820 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003822 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823}
3824
Al Viro74b1da52019-04-15 23:19:05 -04003825static void shmem_free_in_core_inode(struct inode *inode)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003826{
Al Viro84e710d2016-04-15 00:58:55 -04003827 if (S_ISLNK(inode->i_mode))
3828 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003829 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3830}
3831
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832static void shmem_destroy_inode(struct inode *inode)
3833{
Al Viro09208d12011-07-26 03:15:03 -04003834 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836}
3837
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003838static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003840 struct shmem_inode_info *info = foo;
3841 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842}
3843
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003844static void shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845{
3846 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3847 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003848 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849}
3850
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003851static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003853 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854}
3855
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003856static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003858 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003860 .write_begin = shmem_write_begin,
3861 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003863#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003864 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003865#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003866 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867};
3868
Helge Deller15ad7cd2006-12-06 20:40:36 -08003869static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003871 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003873 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003874 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003875 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003876 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003877 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003878 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003879 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880#endif
3881};
3882
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003883static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003884 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003885 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003886#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003887 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003888 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003889#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890};
3891
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003892static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893#ifdef CONFIG_TMPFS
3894 .create = shmem_create,
3895 .lookup = simple_lookup,
3896 .link = shmem_link,
3897 .unlink = shmem_unlink,
3898 .symlink = shmem_symlink,
3899 .mkdir = shmem_mkdir,
3900 .rmdir = shmem_rmdir,
3901 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003902 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003903 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003905#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003906 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003907#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003908#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003909 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003910 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003911#endif
3912};
3913
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003914static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003915#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003916 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003917#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003918#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003919 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003920 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003921#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922};
3923
Hugh Dickins759b9772007-03-05 00:30:28 -08003924static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 .alloc_inode = shmem_alloc_inode,
Al Viro74b1da52019-04-15 23:19:05 -04003926 .free_inode = shmem_free_in_core_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927 .destroy_inode = shmem_destroy_inode,
3928#ifdef CONFIG_TMPFS
3929 .statfs = shmem_statfs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003930 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931#endif
Al Viro1f895f72010-06-05 19:10:41 -04003932 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 .drop_inode = generic_delete_inode,
3934 .put_super = shmem_put_super,
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003935#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003936 .nr_cached_objects = shmem_unused_huge_count,
3937 .free_cached_objects = shmem_unused_huge_scan,
3938#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939};
3940
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003941static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003942 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003943 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944#ifdef CONFIG_NUMA
3945 .set_policy = shmem_set_policy,
3946 .get_policy = shmem_get_policy,
3947#endif
3948};
3949
David Howellsf3235622019-03-25 16:38:31 +00003950int shmem_init_fs_context(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951{
David Howellsf3235622019-03-25 16:38:31 +00003952 struct shmem_options *ctx;
3953
3954 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3955 if (!ctx)
3956 return -ENOMEM;
3957
3958 ctx->mode = 0777 | S_ISVTX;
3959 ctx->uid = current_fsuid();
3960 ctx->gid = current_fsgid();
3961
3962 fc->fs_private = ctx;
3963 fc->ops = &shmem_fs_context_ops;
3964 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965}
3966
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003967static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 .owner = THIS_MODULE,
3969 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00003970 .init_fs_context = shmem_init_fs_context,
3971#ifdef CONFIG_TMPFS
Al Virod7167b12019-09-07 07:23:15 -04003972 .parameters = shmem_fs_parameters,
David Howellsf3235622019-03-25 16:38:31 +00003973#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003975 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003978int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979{
3980 int error;
3981
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003982 shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003984 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003986 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 goto out2;
3988 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003989
Al Viroca4e0512013-08-31 12:57:10 -04003990 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 if (IS_ERR(shm_mnt)) {
3992 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003993 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 goto out1;
3995 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003996
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003997#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07003998 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003999 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4000 else
4001 shmem_huge = 0; /* just in case it was patched */
4002#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 return 0;
4004
4005out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004006 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004008 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 shm_mnt = ERR_PTR(error);
4010 return error;
4011}
Matt Mackall853ac432009-01-06 14:40:20 -08004012
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004013#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004014static ssize_t shmem_enabled_show(struct kobject *kobj,
4015 struct kobj_attribute *attr, char *buf)
4016{
Colin Ian King26083eb62019-11-30 17:58:04 -08004017 static const int values[] = {
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004018 SHMEM_HUGE_ALWAYS,
4019 SHMEM_HUGE_WITHIN_SIZE,
4020 SHMEM_HUGE_ADVISE,
4021 SHMEM_HUGE_NEVER,
4022 SHMEM_HUGE_DENY,
4023 SHMEM_HUGE_FORCE,
4024 };
4025 int i, count;
4026
4027 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4028 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4029
4030 count += sprintf(buf + count, fmt,
4031 shmem_format_huge(values[i]));
4032 }
4033 buf[count - 1] = '\n';
4034 return count;
4035}
4036
4037static ssize_t shmem_enabled_store(struct kobject *kobj,
4038 struct kobj_attribute *attr, const char *buf, size_t count)
4039{
4040 char tmp[16];
4041 int huge;
4042
4043 if (count + 1 > sizeof(tmp))
4044 return -EINVAL;
4045 memcpy(tmp, buf, count);
4046 tmp[count] = '\0';
4047 if (count && tmp[count - 1] == '\n')
4048 tmp[count - 1] = '\0';
4049
4050 huge = shmem_parse_huge(tmp);
4051 if (huge == -EINVAL)
4052 return -EINVAL;
4053 if (!has_transparent_hugepage() &&
4054 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4055 return -EINVAL;
4056
4057 shmem_huge = huge;
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004058 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004059 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4060 return count;
4061}
4062
4063struct kobj_attribute shmem_enabled_attr =
4064 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004065#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004066
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004067#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004068bool shmem_huge_enabled(struct vm_area_struct *vma)
4069{
4070 struct inode *inode = file_inode(vma->vm_file);
4071 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4072 loff_t i_size;
4073 pgoff_t off;
4074
Yang Shic0630662019-07-18 15:57:27 -07004075 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4076 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4077 return false;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004078 if (shmem_huge == SHMEM_HUGE_FORCE)
4079 return true;
4080 if (shmem_huge == SHMEM_HUGE_DENY)
4081 return false;
4082 switch (sbinfo->huge) {
4083 case SHMEM_HUGE_NEVER:
4084 return false;
4085 case SHMEM_HUGE_ALWAYS:
4086 return true;
4087 case SHMEM_HUGE_WITHIN_SIZE:
4088 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4089 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4090 if (i_size >= HPAGE_PMD_SIZE &&
4091 i_size >> PAGE_SHIFT >= off)
4092 return true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07004093 fallthrough;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004094 case SHMEM_HUGE_ADVISE:
4095 /* TODO: implement fadvise() hints */
4096 return (vma->vm_flags & VM_HUGEPAGE);
4097 default:
4098 VM_BUG_ON(1);
4099 return false;
4100 }
4101}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004102#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004103
Matt Mackall853ac432009-01-06 14:40:20 -08004104#else /* !CONFIG_SHMEM */
4105
4106/*
4107 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4108 *
4109 * This is intended for small system where the benefits of the full
4110 * shmem code (swap-backed and resource-limited) are outweighed by
4111 * their complexity. On systems without swap this code should be
4112 * effectively equivalent, but much lighter weight.
4113 */
4114
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004115static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004116 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00004117 .init_fs_context = ramfs_init_fs_context,
Al Virod7167b12019-09-07 07:23:15 -04004118 .parameters = ramfs_fs_parameters,
Matt Mackall853ac432009-01-06 14:40:20 -08004119 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004120 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004121};
4122
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004123int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004124{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004125 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004126
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004127 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004128 BUG_ON(IS_ERR(shm_mnt));
4129
4130 return 0;
4131}
4132
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08004133int shmem_unuse(unsigned int type, bool frontswap,
4134 unsigned long *fs_pages_to_unuse)
Matt Mackall853ac432009-01-06 14:40:20 -08004135{
4136 return 0;
4137}
4138
Hugh Dickins3f96b792009-09-21 17:03:37 -07004139int shmem_lock(struct file *file, int lock, struct user_struct *user)
4140{
4141 return 0;
4142}
4143
Hugh Dickins24513262012-01-20 14:34:21 -08004144void shmem_unlock_mapping(struct address_space *mapping)
4145{
4146}
4147
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004148#ifdef CONFIG_MMU
4149unsigned long shmem_get_unmapped_area(struct file *file,
4150 unsigned long addr, unsigned long len,
4151 unsigned long pgoff, unsigned long flags)
4152{
4153 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4154}
4155#endif
4156
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004157void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004158{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004159 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004160}
4161EXPORT_SYMBOL_GPL(shmem_truncate_range);
4162
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004163#define shmem_vm_ops generic_file_vm_ops
4164#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004165#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004166#define shmem_acct_size(flags, size) 0
4167#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004168
4169#endif /* CONFIG_SHMEM */
4170
4171/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172
Matthew Auld703321b2017-10-06 23:18:13 +01004173static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
Eric Parisc7277092013-12-02 11:24:19 +00004174 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176 struct inode *inode;
Al Viro93dec2d2018-07-08 23:02:03 -04004177 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178
Matthew Auld703321b2017-10-06 23:18:13 +01004179 if (IS_ERR(mnt))
4180 return ERR_CAST(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181
Hugh Dickins285b2c42011-08-03 16:21:20 -07004182 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 return ERR_PTR(-EINVAL);
4184
4185 if (shmem_acct_size(flags, size))
4186 return ERR_PTR(-ENOMEM);
4187
Al Viro93dec2d2018-07-08 23:02:03 -04004188 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4189 flags);
Al Virodac2d1f2018-06-09 09:27:41 -04004190 if (unlikely(!inode)) {
4191 shmem_unacct_size(flags, size);
4192 return ERR_PTR(-ENOSPC);
4193 }
Eric Parisc7277092013-12-02 11:24:19 +00004194 inode->i_flags |= i_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004196 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004197 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
Al Viro93dec2d2018-07-08 23:02:03 -04004198 if (!IS_ERR(res))
4199 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4200 &shmem_file_operations);
Al Viro26567cd2013-03-01 20:22:53 -05004201 if (IS_ERR(res))
Al Viro93dec2d2018-07-08 23:02:03 -04004202 iput(inode);
Al Viro6b4d0b22013-02-14 21:37:26 -05004203 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204}
Eric Parisc7277092013-12-02 11:24:19 +00004205
4206/**
4207 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4208 * kernel internal. There will be NO LSM permission checks against the
4209 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004210 * higher layer. The users are the big_key and shm implementations. LSM
4211 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004212 * @name: name for dentry (to be seen in /proc/<pid>/maps
4213 * @size: size to be set for the file
4214 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4215 */
4216struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4217{
Matthew Auld703321b2017-10-06 23:18:13 +01004218 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
Eric Parisc7277092013-12-02 11:24:19 +00004219}
4220
4221/**
4222 * shmem_file_setup - get an unlinked file living in tmpfs
4223 * @name: name for dentry (to be seen in /proc/<pid>/maps
4224 * @size: size to be set for the file
4225 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4226 */
4227struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4228{
Matthew Auld703321b2017-10-06 23:18:13 +01004229 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
Eric Parisc7277092013-12-02 11:24:19 +00004230}
Keith Packard395e0dd2008-06-20 00:08:06 -07004231EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232
Randy Dunlap46711812008-03-19 17:00:41 -07004233/**
Matthew Auld703321b2017-10-06 23:18:13 +01004234 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4235 * @mnt: the tmpfs mount where the file will be created
4236 * @name: name for dentry (to be seen in /proc/<pid>/maps
4237 * @size: size to be set for the file
4238 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4239 */
4240struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4241 loff_t size, unsigned long flags)
4242{
4243 return __shmem_file_setup(mnt, name, size, flags, 0);
4244}
4245EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4246
4247/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 * shmem_zero_setup - setup a shared anonymous mapping
Peter Collingbourne45e55302020-08-06 23:23:37 -07004249 * @vma: the vma to be mmapped is prepared by do_mmap
Linus Torvalds1da177e2005-04-16 15:20:36 -07004250 */
4251int shmem_zero_setup(struct vm_area_struct *vma)
4252{
4253 struct file *file;
4254 loff_t size = vma->vm_end - vma->vm_start;
4255
Hugh Dickins66fc1302015-06-14 09:48:09 -07004256 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07004257 * Cloning a new file under mmap_lock leads to a lock ordering conflict
Hugh Dickins66fc1302015-06-14 09:48:09 -07004258 * between XFS directory reading and selinux: since this file is only
4259 * accessible to the user through its mapping, use S_PRIVATE flag to
4260 * bypass file security, in the same way as shmem_kernel_file_setup().
4261 */
Matthew Auld703321b2017-10-06 23:18:13 +01004262 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 if (IS_ERR(file))
4264 return PTR_ERR(file);
4265
4266 if (vma->vm_file)
4267 fput(vma->vm_file);
4268 vma->vm_file = file;
4269 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004270
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004271 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004272 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4273 (vma->vm_end & HPAGE_PMD_MASK)) {
4274 khugepaged_enter(vma, vma->vm_flags);
4275 }
4276
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277 return 0;
4278}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004279
4280/**
4281 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4282 * @mapping: the page's address_space
4283 * @index: the page index
4284 * @gfp: the page allocator flags to use if allocating
4285 *
4286 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4287 * with any new page allocations done using the specified allocation flags.
4288 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4289 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4290 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4291 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004292 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4293 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004294 */
4295struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4296 pgoff_t index, gfp_t gfp)
4297{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004298#ifdef CONFIG_SHMEM
4299 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004300 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004301 int error;
4302
4303 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004304 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
Mike Rapoportcfda0522017-02-22 15:43:37 -08004305 gfp, NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004306 if (error)
4307 page = ERR_PTR(error);
4308 else
4309 unlock_page(page);
4310 return page;
4311#else
4312 /*
4313 * The tiny !SHMEM case uses ramfs without swap
4314 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004315 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004316#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004317}
4318EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);