blob: 7c6b6d8f6c396d5c9f608a17983fbe7691c6b9cf [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;
Hui Su30e6a512020-12-14 19:06:14 -0800249const 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);
Byron Stanoszekbb3e96d2020-09-18 21:20:18 -0700282 if (sbinfo->max_inodes) {
283 if (!sbinfo->free_inodes) {
284 spin_unlock(&sbinfo->stat_lock);
285 return -ENOSPC;
286 }
287 sbinfo->free_inodes--;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800288 }
Chris Downe809d5f2020-08-06 23:20:20 -0700289 if (inop) {
290 ino = sbinfo->next_ino++;
291 if (unlikely(is_zero_ino(ino)))
292 ino = sbinfo->next_ino++;
Chris Downea3271f2020-08-06 23:20:25 -0700293 if (unlikely(!sbinfo->full_inums &&
294 ino > UINT_MAX)) {
Chris Downe809d5f2020-08-06 23:20:20 -0700295 /*
296 * Emulate get_next_ino uint wraparound for
297 * compatibility
298 */
Chris Downea3271f2020-08-06 23:20:25 -0700299 if (IS_ENABLED(CONFIG_64BIT))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__, MINOR(sb->s_dev));
302 sbinfo->next_ino = 1;
303 ino = sbinfo->next_ino++;
Chris Downe809d5f2020-08-06 23:20:20 -0700304 }
305 *inop = ino;
306 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800307 spin_unlock(&sbinfo->stat_lock);
Chris Downe809d5f2020-08-06 23:20:20 -0700308 } else if (inop) {
309 /*
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
Chris Downea3271f2020-08-06 23:20:25 -0700316 *
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
Chris Downe809d5f2020-08-06 23:20:20 -0700320 */
321 ino_t *next_ino;
322 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
323 ino = *next_ino;
324 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
325 spin_lock(&sbinfo->stat_lock);
326 ino = sbinfo->next_ino;
327 sbinfo->next_ino += SHMEM_INO_BATCH;
328 spin_unlock(&sbinfo->stat_lock);
329 if (unlikely(is_zero_ino(ino)))
330 ino++;
331 }
332 *inop = ino;
333 *next_ino = ++ino;
334 put_cpu();
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800335 }
Chris Downe809d5f2020-08-06 23:20:20 -0700336
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800337 return 0;
338}
339
340static void shmem_free_inode(struct super_block *sb)
341{
342 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
343 if (sbinfo->max_inodes) {
344 spin_lock(&sbinfo->stat_lock);
345 sbinfo->free_inodes++;
346 spin_unlock(&sbinfo->stat_lock);
347 }
348}
349
Randy Dunlap46711812008-03-19 17:00:41 -0700350/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700351 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * @inode: inode to recalc
353 *
354 * We have to calculate the free blocks since the mm can drop
355 * undirtied hole pages behind our back.
356 *
357 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
359 *
360 * It has to be called with the spinlock held.
361 */
362static void shmem_recalc_inode(struct inode *inode)
363{
364 struct shmem_inode_info *info = SHMEM_I(inode);
365 long freed;
366
367 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
368 if (freed > 0) {
369 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700370 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Mike Rapoport0f079692017-09-06 16:22:59 -0700371 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 }
373}
374
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700375bool shmem_charge(struct inode *inode, long pages)
376{
377 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700378 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700379
Mike Rapoport0f079692017-09-06 16:22:59 -0700380 if (!shmem_inode_acct_block(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700381 return false;
Mike Rapoportb1cc94a2017-09-06 16:22:56 -0700382
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800383 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
384 inode->i_mapping->nrpages += pages;
385
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700386 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700387 info->alloced += pages;
388 inode->i_blocks += pages * BLOCKS_PER_PAGE;
389 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700390 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700391
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700392 return true;
393}
394
395void shmem_uncharge(struct inode *inode, long pages)
396{
397 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700398 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700399
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800400 /* nrpages adjustment done by __delete_from_page_cache() or caller */
401
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700402 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700403 info->alloced -= pages;
404 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
405 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700406 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700407
Mike Rapoport0f079692017-09-06 16:22:59 -0700408 shmem_inode_unacct_blocks(inode, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700409}
410
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700411/*
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500412 * Replace item expected in xarray by a new item, while holding xa_lock.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700413 */
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500414static int shmem_replace_entry(struct address_space *mapping,
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700415 pgoff_t index, void *expected, void *replacement)
416{
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500417 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700418 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700419
420 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700421 VM_BUG_ON(!replacement);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500422 item = xas_load(&xas);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700423 if (item != expected)
424 return -ENOENT;
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500425 xas_store(&xas, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700426 return 0;
427}
428
429/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700430 * Sometimes, before we decide whether to proceed or to fail, we must check
431 * that an entry was not already brought back from swap by a racing thread.
432 *
433 * Checking page is not enough: by the time a SwapCache page is locked, it
434 * might be reused, and again be SwapCache, using the same swap as before.
435 */
436static bool shmem_confirm_swap(struct address_space *mapping,
437 pgoff_t index, swp_entry_t swap)
438{
Matthew Wilcoxa12831b2017-11-22 08:34:58 -0500439 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
Hugh Dickinsd1899222012-07-11 14:02:47 -0700440}
441
442/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
444 *
445 * SHMEM_HUGE_NEVER:
446 * disables huge pages for the mount;
447 * SHMEM_HUGE_ALWAYS:
448 * enables huge pages for the mount;
449 * SHMEM_HUGE_WITHIN_SIZE:
450 * only allocate huge pages if the page will be fully within i_size,
451 * also respect fadvise()/madvise() hints;
452 * SHMEM_HUGE_ADVISE:
453 * only allocate huge pages if requested with fadvise()/madvise();
454 */
455
456#define SHMEM_HUGE_NEVER 0
457#define SHMEM_HUGE_ALWAYS 1
458#define SHMEM_HUGE_WITHIN_SIZE 2
459#define SHMEM_HUGE_ADVISE 3
460
461/*
462 * Special values.
463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
464 *
465 * SHMEM_HUGE_DENY:
466 * disables huge on shm_mnt and all mounts, for emergency use;
467 * SHMEM_HUGE_FORCE:
468 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
469 *
470 */
471#define SHMEM_HUGE_DENY (-1)
472#define SHMEM_HUGE_FORCE (-2)
473
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700474#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700475/* ifdef here to avoid bloating shmem.o when not necessary */
476
Mike Kravetz5b9c98f2018-06-07 17:05:53 -0700477static int shmem_huge __read_mostly;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700478
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700479#if defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700480static int shmem_parse_huge(const char *str)
481{
482 if (!strcmp(str, "never"))
483 return SHMEM_HUGE_NEVER;
484 if (!strcmp(str, "always"))
485 return SHMEM_HUGE_ALWAYS;
486 if (!strcmp(str, "within_size"))
487 return SHMEM_HUGE_WITHIN_SIZE;
488 if (!strcmp(str, "advise"))
489 return SHMEM_HUGE_ADVISE;
490 if (!strcmp(str, "deny"))
491 return SHMEM_HUGE_DENY;
492 if (!strcmp(str, "force"))
493 return SHMEM_HUGE_FORCE;
494 return -EINVAL;
495}
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700496#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700497
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700498#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700499static const char *shmem_format_huge(int huge)
500{
501 switch (huge) {
502 case SHMEM_HUGE_NEVER:
503 return "never";
504 case SHMEM_HUGE_ALWAYS:
505 return "always";
506 case SHMEM_HUGE_WITHIN_SIZE:
507 return "within_size";
508 case SHMEM_HUGE_ADVISE:
509 return "advise";
510 case SHMEM_HUGE_DENY:
511 return "deny";
512 case SHMEM_HUGE_FORCE:
513 return "force";
514 default:
515 VM_BUG_ON(1);
516 return "bad_val";
517 }
518}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800519#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700520
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700521static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
522 struct shrink_control *sc, unsigned long nr_to_split)
523{
524 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800525 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700526 struct inode *inode;
527 struct shmem_inode_info *info;
528 struct page *page;
529 unsigned long batch = sc ? sc->nr_to_scan : 128;
530 int removed = 0, split = 0;
531
532 if (list_empty(&sbinfo->shrinklist))
533 return SHRINK_STOP;
534
535 spin_lock(&sbinfo->shrinklist_lock);
536 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
537 info = list_entry(pos, struct shmem_inode_info, shrinklist);
538
539 /* pin the inode */
540 inode = igrab(&info->vfs_inode);
541
542 /* inode is about to be evicted */
543 if (!inode) {
544 list_del_init(&info->shrinklist);
545 removed++;
546 goto next;
547 }
548
549 /* Check if there's anything to gain */
550 if (round_up(inode->i_size, PAGE_SIZE) ==
551 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800552 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700553 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700554 goto next;
555 }
556
557 list_move(&info->shrinklist, &list);
558next:
559 if (!--batch)
560 break;
561 }
562 spin_unlock(&sbinfo->shrinklist_lock);
563
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800564 list_for_each_safe(pos, next, &to_remove) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
566 inode = &info->vfs_inode;
567 list_del_init(&info->shrinklist);
568 iput(inode);
569 }
570
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700571 list_for_each_safe(pos, next, &list) {
572 int ret;
573
574 info = list_entry(pos, struct shmem_inode_info, shrinklist);
575 inode = &info->vfs_inode;
576
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700577 if (nr_to_split && split >= nr_to_split)
578 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700579
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700580 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700581 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
582 if (!page)
583 goto drop;
584
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700585 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700586 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700587 put_page(page);
588 goto drop;
589 }
590
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700591 /*
592 * Leave the inode on the list if we failed to lock
593 * the page at this time.
594 *
595 * Waiting for the lock may lead to deadlock in the
596 * reclaim path.
597 */
598 if (!trylock_page(page)) {
599 put_page(page);
600 goto leave;
601 }
602
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700603 ret = split_huge_page(page);
604 unlock_page(page);
605 put_page(page);
606
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700607 /* If split failed leave the inode on the list */
608 if (ret)
609 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700610
611 split++;
612drop:
613 list_del_init(&info->shrinklist);
614 removed++;
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700615leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700616 iput(inode);
617 }
618
619 spin_lock(&sbinfo->shrinklist_lock);
620 list_splice_tail(&list, &sbinfo->shrinklist);
621 sbinfo->shrinklist_len -= removed;
622 spin_unlock(&sbinfo->shrinklist_lock);
623
624 return split;
625}
626
627static long shmem_unused_huge_scan(struct super_block *sb,
628 struct shrink_control *sc)
629{
630 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
631
632 if (!READ_ONCE(sbinfo->shrinklist_len))
633 return SHRINK_STOP;
634
635 return shmem_unused_huge_shrink(sbinfo, sc, 0);
636}
637
638static long shmem_unused_huge_count(struct super_block *sb,
639 struct shrink_control *sc)
640{
641 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
642 return READ_ONCE(sbinfo->shrinklist_len);
643}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700644#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700645
646#define shmem_huge SHMEM_HUGE_DENY
647
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700648static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
649 struct shrink_control *sc, unsigned long nr_to_split)
650{
651 return 0;
652}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700653#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700654
Yang Shi89fdcd22018-06-07 17:06:59 -0700655static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
656{
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700657 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Yang Shi89fdcd22018-06-07 17:06:59 -0700658 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
659 shmem_huge != SHMEM_HUGE_DENY)
660 return true;
661 return false;
662}
663
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700664/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700665 * Like add_to_page_cache_locked, but error if expected item has gone.
666 */
667static int shmem_add_to_page_cache(struct page *page,
668 struct address_space *mapping,
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700669 pgoff_t index, void *expected, gfp_t gfp,
670 struct mm_struct *charge_mm)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700671{
Matthew Wilcox552446a2017-12-01 13:25:14 -0500672 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
673 unsigned long i = 0;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -0700674 unsigned long nr = compound_nr(page);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700675 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700676
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700677 VM_BUG_ON_PAGE(PageTail(page), page);
678 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800679 VM_BUG_ON_PAGE(!PageLocked(page), page);
680 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700681 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700682
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700683 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700684 page->mapping = mapping;
685 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700686
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700687 if (!PageSwapCache(page)) {
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -0700688 error = mem_cgroup_charge(page, charge_mm, gfp);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700689 if (error) {
690 if (PageTransHuge(page)) {
691 count_vm_event(THP_FILE_FALLBACK);
692 count_vm_event(THP_FILE_FALLBACK_CHARGE);
693 }
694 goto error;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700695 }
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700696 }
697 cgroup_throttle_swaprate(page, gfp);
698
Matthew Wilcox552446a2017-12-01 13:25:14 -0500699 do {
700 void *entry;
701 xas_lock_irq(&xas);
702 entry = xas_find_conflict(&xas);
703 if (entry != expected)
704 xas_set_err(&xas, -EEXIST);
705 xas_create_range(&xas);
706 if (xas_error(&xas))
707 goto unlock;
708next:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700709 xas_store(&xas, page);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500710 if (++i < nr) {
711 xas_next(&xas);
712 goto next;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700713 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500714 if (PageTransHuge(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700715 count_vm_event(THP_FILE_ALLOC);
Johannes Weinerb8eddff2020-12-14 19:06:20 -0800716 __inc_lruvec_page_state(page, NR_SHMEM_THPS);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500717 }
718 mapping->nrpages += nr;
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700719 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
720 __mod_lruvec_page_state(page, NR_SHMEM, nr);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500721unlock:
722 xas_unlock_irq(&xas);
723 } while (xas_nomem(&xas, gfp));
724
725 if (xas_error(&xas)) {
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700726 error = xas_error(&xas);
727 goto error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700728 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500729
730 return 0;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700731error:
732 page->mapping = NULL;
733 page_ref_sub(page, nr);
734 return error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700735}
736
737/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700738 * Like delete_from_page_cache, but substitutes swap for page.
739 */
740static void shmem_delete_from_page_cache(struct page *page, void *radswap)
741{
742 struct address_space *mapping = page->mapping;
743 int error;
744
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700745 VM_BUG_ON_PAGE(PageCompound(page), page);
746
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700747 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500748 error = shmem_replace_entry(mapping, page->index, page, radswap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700749 page->mapping = NULL;
750 mapping->nrpages--;
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700751 __dec_lruvec_page_state(page, NR_FILE_PAGES);
752 __dec_lruvec_page_state(page, NR_SHMEM);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700753 xa_unlock_irq(&mapping->i_pages);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300754 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700755 BUG_ON(error);
756}
757
758/*
Matthew Wilcoxc121d3b2017-12-04 03:13:54 -0500759 * Remove swap entry from page cache, free the swap and its page cache.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700760 */
761static int shmem_free_swap(struct address_space *mapping,
762 pgoff_t index, void *radswap)
763{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700764 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700765
Matthew Wilcox55f3f7e2018-11-26 16:08:43 -0500766 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700767 if (old != radswap)
768 return -ENOENT;
769 free_swap_and_cache(radix_to_swp_entry(radswap));
770 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700771}
772
773/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800774 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800775 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800776 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800778 * as long as the inode doesn't go away and racy results are not a problem.
779 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800780unsigned long shmem_partial_swap_usage(struct address_space *mapping,
781 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800782{
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500783 XA_STATE(xas, &mapping->i_pages, start);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800784 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800785 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800786
787 rcu_read_lock();
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500788 xas_for_each(&xas, page, end - 1) {
789 if (xas_retry(&xas, page))
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700790 continue;
Matthew Wilcox3159f942017-11-03 13:30:42 -0400791 if (xa_is_value(page))
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800792 swapped++;
793
794 if (need_resched()) {
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500795 xas_pause(&xas);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800796 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800797 }
798 }
799
800 rcu_read_unlock();
801
802 return swapped << PAGE_SHIFT;
803}
804
805/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given vma is swapped out.
808 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka48131e02016-01-14 15:19:23 -0800810 * as long as the inode doesn't go away and racy results are not a problem.
811 */
812unsigned long shmem_swap_usage(struct vm_area_struct *vma)
813{
814 struct inode *inode = file_inode(vma->vm_file);
815 struct shmem_inode_info *info = SHMEM_I(inode);
816 struct address_space *mapping = inode->i_mapping;
817 unsigned long swapped;
818
819 /* Be careful as we don't hold info->lock */
820 swapped = READ_ONCE(info->swapped);
821
822 /*
823 * The easier cases are when the shmem object has nothing in swap, or
824 * the vma maps it whole. Then we can simply use the stats that we
825 * already track.
826 */
827 if (!swapped)
828 return 0;
829
830 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
831 return swapped << PAGE_SHIFT;
832
833 /* Here comes the more involved part */
834 return shmem_partial_swap_usage(mapping,
835 linear_page_index(vma, vma->vm_start),
836 linear_page_index(vma, vma->vm_end));
837}
838
839/*
Hugh Dickins24513262012-01-20 14:34:21 -0800840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
841 */
842void shmem_unlock_mapping(struct address_space *mapping)
843{
844 struct pagevec pvec;
845 pgoff_t indices[PAGEVEC_SIZE];
846 pgoff_t index = 0;
847
Mel Gorman86679822017-11-15 17:37:52 -0800848 pagevec_init(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800849 /*
850 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
851 */
852 while (!mapping_unevictable(mapping)) {
853 /*
854 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
855 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
856 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700857 pvec.nr = find_get_entries(mapping, index,
858 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800859 if (!pvec.nr)
860 break;
861 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700862 pagevec_remove_exceptionals(&pvec);
Kuo-Hsin Yang64e3d122018-11-06 13:23:24 +0000863 check_move_unevictable_pages(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800864 pagevec_release(&pvec);
865 cond_resched();
866 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700867}
868
869/*
Hugh Dickins71725ed2020-04-06 20:07:57 -0700870 * Check whether a hole-punch or truncation needs to split a huge page,
871 * returning true if no split was required, or the split has been successful.
872 *
873 * Eviction (or truncation to 0 size) should never need to split a huge page;
874 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
875 * head, and then succeeded to trylock on tail.
876 *
877 * A split can only succeed when there are no additional references on the
878 * huge page: so the split below relies upon find_get_entries() having stopped
879 * when it found a subpage of the huge page, without getting further references.
880 */
881static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
882{
883 if (!PageTransCompound(page))
884 return true;
885
886 /* Just proceed to delete a huge page wholly within the range punched */
887 if (PageHead(page) &&
888 page->index >= start && page->index + HPAGE_PMD_NR <= end)
889 return true;
890
891 /* Try to split huge page, so we can truly punch the hole or truncate */
892 return split_huge_page(page) >= 0;
893}
894
895/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -0500896 * Remove range of pages and swap entries from page cache, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700897 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700898 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700899static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
900 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700902 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300904 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
905 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
906 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
907 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700908 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700909 pgoff_t indices[PAGEVEC_SIZE];
910 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700911 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700912 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700914 if (lend == -1)
915 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700916
Mel Gorman86679822017-11-15 17:37:52 -0800917 pagevec_init(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700918 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700919 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700920 pvec.nr = find_get_entries(mapping, index,
921 min(end - index, (pgoff_t)PAGEVEC_SIZE),
922 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700923 if (!pvec.nr)
924 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700925 for (i = 0; i < pagevec_count(&pvec); i++) {
926 struct page *page = pvec.pages[i];
927
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700928 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700929 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700930 break;
931
Matthew Wilcox3159f942017-11-03 13:30:42 -0400932 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700933 if (unfalloc)
934 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700935 nr_swaps_freed += !shmem_free_swap(mapping,
936 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700937 continue;
938 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700939
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700940 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
941
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700942 if (!trylock_page(page))
943 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700944
Hugh Dickins71725ed2020-04-06 20:07:57 -0700945 if ((!unfalloc || !PageUptodate(page)) &&
946 page_mapping(page) == mapping) {
947 VM_BUG_ON_PAGE(PageWriteback(page), page);
948 if (shmem_punch_compound(page, start, end))
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700949 truncate_inode_page(mapping, page);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700950 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700951 unlock_page(page);
952 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700953 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800954 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700955 cond_resched();
956 index++;
957 }
958
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700959 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700960 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700961 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700962 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300963 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700964 if (start > end) {
965 top = partial_end;
966 partial_end = 0;
967 }
968 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700969 set_page_dirty(page);
970 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300971 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700972 }
973 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700974 if (partial_end) {
975 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700976 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700977 if (page) {
978 zero_user_segment(page, 0, partial_end);
979 set_page_dirty(page);
980 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300981 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700982 }
983 }
984 if (start >= end)
985 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700986
987 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700988 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700989 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700990
991 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700992 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700993 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700994 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700995 /* If all gone or hole-punch or unfalloc, we're done */
996 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700997 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700998 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700999 index = start;
1000 continue;
1001 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001002 for (i = 0; i < pagevec_count(&pvec); i++) {
1003 struct page *page = pvec.pages[i];
1004
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001005 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001006 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001007 break;
1008
Matthew Wilcox3159f942017-11-03 13:30:42 -04001009 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001010 if (unfalloc)
1011 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001012 if (shmem_free_swap(mapping, index, page)) {
1013 /* Swap was replaced by page: retry */
1014 index--;
1015 break;
1016 }
1017 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001018 continue;
1019 }
1020
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001021 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001022
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001023 if (!unfalloc || !PageUptodate(page)) {
Hugh Dickins71725ed2020-04-06 20:07:57 -07001024 if (page_mapping(page) != mapping) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001025 /* Page was replaced by swap: retry */
1026 unlock_page(page);
1027 index--;
1028 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001029 }
Hugh Dickins71725ed2020-04-06 20:07:57 -07001030 VM_BUG_ON_PAGE(PageWriteback(page), page);
1031 if (shmem_punch_compound(page, start, end))
1032 truncate_inode_page(mapping, page);
Hugh Dickins0783ac92020-04-20 18:14:07 -07001033 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
Hugh Dickins71725ed2020-04-06 20:07:57 -07001034 /* Wipe the page and don't get stuck */
1035 clear_highpage(page);
1036 flush_dcache_page(page);
1037 set_page_dirty(page);
1038 if (index <
1039 round_up(start, HPAGE_PMD_NR))
1040 start = index + 1;
1041 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001042 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001043 unlock_page(page);
1044 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001045 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -08001046 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001047 index++;
1048 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001049
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001050 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001051 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001053 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001054}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001056void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1057{
1058 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001059 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060}
Hugh Dickins94c1e622011-06-27 16:18:03 -07001061EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
David Howellsa528d352017-01-31 16:46:22 +00001063static int shmem_getattr(const struct path *path, struct kstat *stat,
1064 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -07001065{
David Howellsa528d352017-01-31 16:46:22 +00001066 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -07001067 struct shmem_inode_info *info = SHMEM_I(inode);
Yang Shi89fdcd22018-06-07 17:06:59 -07001068 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
Yu Zhao44a30222015-09-08 15:03:33 -07001069
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001070 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001071 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001072 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001073 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001074 }
Yu Zhao44a30222015-09-08 15:03:33 -07001075 generic_fillattr(inode, stat);
Yang Shi89fdcd22018-06-07 17:06:59 -07001076
1077 if (is_huge_enabled(sb_info))
1078 stat->blksize = HPAGE_PMD_SIZE;
1079
Yu Zhao44a30222015-09-08 15:03:33 -07001080 return 0;
1081}
1082
Hugh Dickins94c1e622011-06-27 16:18:03 -07001083static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
David Howells75c3cfa2015-03-17 22:26:12 +00001085 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -07001086 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001087 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 int error;
1089
Jan Kara31051c82016-05-26 16:55:18 +02001090 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001091 if (error)
1092 return error;
1093
Hugh Dickins94c1e622011-06-27 16:18:03 -07001094 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1095 loff_t oldsize = inode->i_size;
1096 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001097
David Herrmann40e041a2014-08-08 14:25:27 -07001098 /* protected by i_mutex */
1099 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1100 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1101 return -EPERM;
1102
Hugh Dickins94c1e622011-06-27 16:18:03 -07001103 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001104 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1105 oldsize, newsize);
1106 if (error)
1107 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001108 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001109 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001110 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001111 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001112 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001113 if (oldsize > holebegin)
1114 unmap_mapping_range(inode->i_mapping,
1115 holebegin, 0, 1);
1116 if (info->alloced)
1117 shmem_truncate_range(inode,
1118 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001119 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001120 if (oldsize > holebegin)
1121 unmap_mapping_range(inode->i_mapping,
1122 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001123
1124 /*
1125 * Part of the huge page can be beyond i_size: subject
1126 * to shrink under memory pressure.
1127 */
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001128 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001129 spin_lock(&sbinfo->shrinklist_lock);
Cong Wangd0413532017-08-10 15:24:24 -07001130 /*
1131 * _careful to defend against unlocked access to
1132 * ->shrink_list in shmem_unused_huge_shrink()
1133 */
1134 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001135 list_add_tail(&info->shrinklist,
1136 &sbinfo->shrinklist);
1137 sbinfo->shrinklist_len++;
1138 }
1139 spin_unlock(&sbinfo->shrinklist_lock);
1140 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 }
1143
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001144 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001145 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001146 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 return error;
1148}
1149
Al Viro1f895f72010-06-05 19:10:41 -04001150static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001153 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
Hui Su30e6a512020-12-14 19:06:14 -08001155 if (shmem_mapping(inode->i_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 shmem_unacct_size(info->flags, inode->i_size);
1157 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001158 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001159 if (!list_empty(&info->shrinklist)) {
1160 spin_lock(&sbinfo->shrinklist_lock);
1161 if (!list_empty(&info->shrinklist)) {
1162 list_del_init(&info->shrinklist);
1163 sbinfo->shrinklist_len--;
1164 }
1165 spin_unlock(&sbinfo->shrinklist_lock);
1166 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001167 while (!list_empty(&info->swaplist)) {
1168 /* Wait while shmem_unuse() is scanning this inode... */
1169 wait_var_event(&info->stop_eviction,
1170 !atomic_read(&info->stop_eviction));
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001171 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001172 /* ...but beware of the race if we peeked too early */
1173 if (!atomic_read(&info->stop_eviction))
1174 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001175 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
Al Viro3ed47db2016-01-22 18:08:52 -05001177 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001178
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001179 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001180 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001181 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001182 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183}
1184
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001185extern struct swap_info_struct *swap_info[];
1186
1187static int shmem_find_swap_entries(struct address_space *mapping,
1188 pgoff_t start, unsigned int nr_entries,
1189 struct page **entries, pgoff_t *indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001190 unsigned int type, bool frontswap)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001191{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001192 XA_STATE(xas, &mapping->i_pages, start);
1193 struct page *page;
Hugh Dickins87039542019-04-18 17:49:58 -07001194 swp_entry_t entry;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001195 unsigned int ret = 0;
1196
1197 if (!nr_entries)
1198 return 0;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001199
1200 rcu_read_lock();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001201 xas_for_each(&xas, page, ULONG_MAX) {
1202 if (xas_retry(&xas, page))
Mike Kravetz5b9c98f2018-06-07 17:05:53 -07001203 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001204
1205 if (!xa_is_value(page))
1206 continue;
1207
Hugh Dickins87039542019-04-18 17:49:58 -07001208 entry = radix_to_swp_entry(page);
1209 if (swp_type(entry) != type)
1210 continue;
1211 if (frontswap &&
1212 !frontswap_test(swap_info[type], swp_offset(entry)))
1213 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001214
1215 indices[ret] = xas.xa_index;
1216 entries[ret] = page;
1217
1218 if (need_resched()) {
1219 xas_pause(&xas);
1220 cond_resched_rcu();
1221 }
1222 if (++ret == nr_entries)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001223 break;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001224 }
Matthew Wilcox478922e2016-12-14 15:08:52 -08001225 rcu_read_unlock();
Matthew Wilcoxe21a2952017-11-22 08:36:00 -05001226
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001227 return ret;
1228}
1229
1230/*
1231 * Move the swapped pages for an inode to page cache. Returns the count
1232 * of pages swapped in, or the error in case of failure.
1233 */
1234static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1235 pgoff_t *indices)
1236{
1237 int i = 0;
1238 int ret = 0;
1239 int error = 0;
1240 struct address_space *mapping = inode->i_mapping;
1241
1242 for (i = 0; i < pvec.nr; i++) {
1243 struct page *page = pvec.pages[i];
1244
1245 if (!xa_is_value(page))
1246 continue;
1247 error = shmem_swapin_page(inode, indices[i],
1248 &page, SGP_CACHE,
1249 mapping_gfp_mask(mapping),
1250 NULL, NULL);
1251 if (error == 0) {
1252 unlock_page(page);
1253 put_page(page);
1254 ret++;
1255 }
1256 if (error == -ENOMEM)
1257 break;
1258 error = 0;
1259 }
1260 return error ? error : ret;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001261}
1262
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001263/*
1264 * If swap found in inode, free it and move page from swapcache to filecache.
1265 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001266static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1267 bool frontswap, unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001269 struct address_space *mapping = inode->i_mapping;
1270 pgoff_t start = 0;
1271 struct pagevec pvec;
1272 pgoff_t indices[PAGEVEC_SIZE];
1273 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1274 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001276 pagevec_init(&pvec);
1277 do {
1278 unsigned int nr_entries = PAGEVEC_SIZE;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001279
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001280 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1281 nr_entries = *fs_pages_to_unuse;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001282
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001283 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1284 pvec.pages, indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001285 type, frontswap);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001286 if (pvec.nr == 0) {
1287 ret = 0;
1288 break;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001289 }
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001290
1291 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1292 if (ret < 0)
1293 break;
1294
1295 if (frontswap_partial) {
1296 *fs_pages_to_unuse -= ret;
1297 if (*fs_pages_to_unuse == 0) {
1298 ret = FRONTSWAP_PAGES_UNUSED;
1299 break;
1300 }
1301 }
1302
1303 start = indices[pvec.nr - 1];
1304 } while (true);
1305
1306 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
1308
1309/*
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001310 * Read all the shared memory data that resides in the swap
1311 * device 'type' back into memory, so the swap device can be
1312 * unused.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001314int shmem_unuse(unsigned int type, bool frontswap,
1315 unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001317 struct shmem_inode_info *info, *next;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001318 int error = 0;
1319
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001320 if (list_empty(&shmem_swaplist))
1321 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001323 mutex_lock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001324 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1325 if (!info->swapped) {
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001326 list_del_init(&info->swaplist);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001327 continue;
1328 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001329 /*
1330 * Drop the swaplist mutex while searching the inode for swap;
1331 * but before doing so, make sure shmem_evict_inode() will not
1332 * remove placeholder inode from swaplist, nor let it be freed
1333 * (igrab() would protect from unlink, but not from unmount).
1334 */
1335 atomic_inc(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001336 mutex_unlock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001337
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001338 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001339 fs_pages_to_unuse);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001340 cond_resched();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001341
1342 mutex_lock(&shmem_swaplist_mutex);
1343 next = list_next_entry(info, swaplist);
1344 if (!info->swapped)
1345 list_del_init(&info->swaplist);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001346 if (atomic_dec_and_test(&info->stop_eviction))
1347 wake_up_var(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001348 if (error)
Hugh Dickins778dd892011-05-11 15:13:37 -07001349 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001351 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001352
Hugh Dickins778dd892011-05-11 15:13:37 -07001353 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354}
1355
1356/*
1357 * Move the page from the page cache to the swap cache.
1358 */
1359static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1360{
1361 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001364 swp_entry_t swap;
1365 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001367 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 mapping = page->mapping;
1370 index = page->index;
1371 inode = mapping->host;
1372 info = SHMEM_I(inode);
1373 if (info->flags & VM_LOCKED)
1374 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001375 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 goto redirty;
1377
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001378 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001379 * Our capabilities prevent regular writeback or sync from ever calling
1380 * shmem_writepage; but a stacking filesystem might use ->writepage of
1381 * its underlying filesystem, in which case tmpfs should write out to
1382 * swap only in response to memory pressure, and not for the writeback
1383 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001384 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001385 if (!wbc->for_reclaim) {
1386 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1387 goto redirty;
1388 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001389
1390 /*
1391 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1392 * value into swapfile.c, the only way we can correctly account for a
1393 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001394 *
1395 * That's okay for a page already fallocated earlier, but if we have
1396 * not yet completed the fallocation, then (a) we want to keep track
1397 * of this page in case we have to undo it, and (b) it may not be a
1398 * good idea to continue anyway, once we're pushing into swap. So
1399 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001400 */
1401 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001402 if (inode->i_private) {
1403 struct shmem_falloc *shmem_falloc;
1404 spin_lock(&inode->i_lock);
1405 shmem_falloc = inode->i_private;
1406 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001407 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001408 index >= shmem_falloc->start &&
1409 index < shmem_falloc->next)
1410 shmem_falloc->nr_unswapped++;
1411 else
1412 shmem_falloc = NULL;
1413 spin_unlock(&inode->i_lock);
1414 if (shmem_falloc)
1415 goto redirty;
1416 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001417 clear_highpage(page);
1418 flush_dcache_page(page);
1419 SetPageUptodate(page);
1420 }
1421
Huang Ying38d8b4e2017-07-06 15:37:18 -07001422 swap = get_swap_page(page);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001423 if (!swap.val)
1424 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001425
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001426 /*
1427 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001428 * if it's not already there. Do it now before the page is
1429 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001430 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001431 * we've incremented swapped, because shmem_unuse_inode() will
1432 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001433 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001434 mutex_lock(&shmem_swaplist_mutex);
1435 if (list_empty(&info->swaplist))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001436 list_add(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001437
Yang Shi4afab1c2019-11-30 17:58:07 -08001438 if (add_to_swap_cache(page, swap,
Joonsoo Kim3852f672020-08-11 18:30:47 -07001439 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1440 NULL) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001441 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001442 shmem_recalc_inode(inode);
1443 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001444 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001445
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001446 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001447 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1448
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001449 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001450 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001451 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 return 0;
1453 }
1454
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001455 mutex_unlock(&shmem_swaplist_mutex);
Minchan Kim75f6d6d2017-07-06 15:37:21 -07001456 put_swap_page(page, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457redirty:
1458 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001459 if (wbc->for_reclaim)
1460 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1461 unlock_page(page);
1462 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
Hugh Dickins75edd342016-05-19 17:12:44 -07001465#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001466static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001467{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001468 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001469
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001470 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001471 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001472
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001473 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001474
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001475 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001476}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001477
1478static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1479{
1480 struct mempolicy *mpol = NULL;
1481 if (sbinfo->mpol) {
1482 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1483 mpol = sbinfo->mpol;
1484 mpol_get(mpol);
1485 spin_unlock(&sbinfo->stat_lock);
1486 }
1487 return mpol;
1488}
Hugh Dickins75edd342016-05-19 17:12:44 -07001489#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1490static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1491{
1492}
1493static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1494{
1495 return NULL;
1496}
1497#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1498#ifndef CONFIG_NUMA
1499#define vm_policy vm_private_data
1500#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001501
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001502static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1503 struct shmem_inode_info *info, pgoff_t index)
1504{
1505 /* Create a pseudo vma that just contains the policy */
Kirill A. Shutemov2c4541e2018-07-26 16:37:30 -07001506 vma_init(vma, NULL);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001507 /* Bias interleave by inode number to distribute better across nodes */
1508 vma->vm_pgoff = index + info->vfs_inode.i_ino;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001509 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1510}
1511
1512static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1513{
1514 /* Drop reference taken by mpol_shared_policy_lookup() */
1515 mpol_cond_put(vma->vm_policy);
1516}
1517
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001518static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1519 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001522 struct page *page;
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001523 struct vm_fault vmf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001525 shmem_pseudo_vma_init(&pvma, info, index);
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001526 vmf.vma = &pvma;
1527 vmf.address = 0;
1528 page = swap_cluster_readahead(swap, gfp, &vmf);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001529 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001530
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001531 return page;
1532}
Mel Gorman18a2f372012-12-05 14:01:41 -08001533
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001534static struct page *shmem_alloc_hugepage(gfp_t gfp,
1535 struct shmem_inode_info *info, pgoff_t index)
1536{
1537 struct vm_area_struct pvma;
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001538 struct address_space *mapping = info->vfs_inode.i_mapping;
1539 pgoff_t hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001540 struct page *page;
1541
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001542 hindex = round_down(index, HPAGE_PMD_NR);
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001543 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1544 XA_PRESENT))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001545 return NULL;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001546
1547 shmem_pseudo_vma_init(&pvma, info, hindex);
1548 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
David Rientjes19deb762019-09-04 12:54:20 -07001549 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001550 shmem_pseudo_vma_destroy(&pvma);
1551 if (page)
1552 prep_transhuge_page(page);
David Rientjesdcdf11e2020-04-06 20:04:25 -07001553 else
1554 count_vm_event(THP_FILE_FALLBACK);
Mel Gorman18a2f372012-12-05 14:01:41 -08001555 return page;
1556}
1557
1558static struct page *shmem_alloc_page(gfp_t gfp,
1559 struct shmem_inode_info *info, pgoff_t index)
1560{
1561 struct vm_area_struct pvma;
1562 struct page *page;
1563
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001564 shmem_pseudo_vma_init(&pvma, info, index);
1565 page = alloc_page_vma(gfp, &pvma, 0);
1566 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001567
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001568 return page;
1569}
1570
1571static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
Mike Rapoport0f079692017-09-06 16:22:59 -07001572 struct inode *inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001573 pgoff_t index, bool huge)
1574{
Mike Rapoport0f079692017-09-06 16:22:59 -07001575 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001576 struct page *page;
1577 int nr;
1578 int err = -ENOSPC;
1579
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001580 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001581 huge = false;
1582 nr = huge ? HPAGE_PMD_NR : 1;
1583
Mike Rapoport0f079692017-09-06 16:22:59 -07001584 if (!shmem_inode_acct_block(inode, nr))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001585 goto failed;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001586
1587 if (huge)
1588 page = shmem_alloc_hugepage(gfp, info, index);
1589 else
1590 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001591 if (page) {
1592 __SetPageLocked(page);
1593 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001594 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001595 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001596
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001597 err = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07001598 shmem_inode_unacct_blocks(inode, nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001599failed:
1600 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001604 * When a page is moved from swapcache to shmem filecache (either by the
1605 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1606 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1607 * ignorance of the mapping it belongs to. If that mapping has special
1608 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1609 * we may need to copy to a suitable page before moving to filecache.
1610 *
1611 * In a future release, this may well be extended to respect cpuset and
1612 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1613 * but for now it is a simple matter of zone.
1614 */
1615static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1616{
1617 return page_zonenum(page) > gfp_zone(gfp);
1618}
1619
1620static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1621 struct shmem_inode_info *info, pgoff_t index)
1622{
1623 struct page *oldpage, *newpage;
1624 struct address_space *swap_mapping;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001625 swp_entry_t entry;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001626 pgoff_t swap_index;
1627 int error;
1628
1629 oldpage = *pagep;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001630 entry.val = page_private(oldpage);
1631 swap_index = swp_offset(entry);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001632 swap_mapping = page_mapping(oldpage);
1633
1634 /*
1635 * We have arrived here because our zones are constrained, so don't
1636 * limit chance of success by further cpuset and node constraints.
1637 */
1638 gfp &= ~GFP_CONSTRAINT_MASK;
1639 newpage = shmem_alloc_page(gfp, info, index);
1640 if (!newpage)
1641 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001642
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001643 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001644 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001645 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001646
Hugh Dickins9956edf2016-11-10 10:46:11 -08001647 __SetPageLocked(newpage);
1648 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001649 SetPageUptodate(newpage);
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001650 set_page_private(newpage, entry.val);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001651 SetPageSwapCache(newpage);
1652
1653 /*
1654 * Our caller will very soon move newpage out of swapcache, but it's
1655 * a nice clean interface for us to replace oldpage by newpage there.
1656 */
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001657 xa_lock_irq(&swap_mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -05001658 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001659 if (!error) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07001660 mem_cgroup_migrate(oldpage, newpage);
1661 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1662 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001663 }
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001664 xa_unlock_irq(&swap_mapping->i_pages);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001665
Hugh Dickins0142ef62012-06-07 14:21:09 -07001666 if (unlikely(error)) {
1667 /*
1668 * Is this possible? I think not, now that our callers check
1669 * both PageSwapCache and page_private after getting page lock;
1670 * but be defensive. Reverse old to newpage for clear and free.
1671 */
1672 oldpage = newpage;
1673 } else {
Johannes Weiner6058eae2020-06-03 16:02:40 -07001674 lru_cache_add(newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001675 *pagep = newpage;
1676 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001677
1678 ClearPageSwapCache(oldpage);
1679 set_page_private(oldpage, 0);
1680
1681 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001682 put_page(oldpage);
1683 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001684 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001685}
1686
1687/*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001688 * Swap in the page pointed to by *pagep.
1689 * Caller has to make sure that *pagep contains a valid swapped page.
1690 * Returns 0 and the page in pagep if success. On failure, returns the
Randy Dunlapaf44c122020-08-11 18:33:17 -07001691 * error code and NULL in *pagep.
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001692 */
1693static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1694 struct page **pagep, enum sgp_type sgp,
1695 gfp_t gfp, struct vm_area_struct *vma,
1696 vm_fault_t *fault_type)
1697{
1698 struct address_space *mapping = inode->i_mapping;
1699 struct shmem_inode_info *info = SHMEM_I(inode);
1700 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001701 struct page *page;
1702 swp_entry_t swap;
1703 int error;
1704
1705 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1706 swap = radix_to_swp_entry(*pagep);
1707 *pagep = NULL;
1708
1709 /* Look it up and read it in.. */
1710 page = lookup_swap_cache(swap, NULL, 0);
1711 if (!page) {
1712 /* Or update major stats only when swapin succeeds?? */
1713 if (fault_type) {
1714 *fault_type |= VM_FAULT_MAJOR;
1715 count_vm_event(PGMAJFAULT);
1716 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1717 }
1718 /* Here we actually start the io */
1719 page = shmem_swapin(swap, gfp, info, index);
1720 if (!page) {
1721 error = -ENOMEM;
1722 goto failed;
1723 }
1724 }
1725
1726 /* We have to do this with page locked to prevent races */
1727 lock_page(page);
1728 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1729 !shmem_confirm_swap(mapping, index, swap)) {
1730 error = -EEXIST;
1731 goto unlock;
1732 }
1733 if (!PageUptodate(page)) {
1734 error = -EIO;
1735 goto failed;
1736 }
1737 wait_on_page_writeback(page);
1738
Steven Price8a848022020-05-13 16:37:49 +01001739 /*
1740 * Some architectures may have to restore extra metadata to the
1741 * physical page after reading from swap.
1742 */
1743 arch_swap_restore(swap, page);
1744
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001745 if (shmem_should_replace_page(page, gfp)) {
1746 error = shmem_replace_page(&page, gfp, info, index);
1747 if (error)
1748 goto failed;
1749 }
1750
Johannes Weiner3fea5a42020-06-03 16:01:41 -07001751 error = shmem_add_to_page_cache(page, mapping, index,
1752 swp_to_radix_entry(swap), gfp,
1753 charge_mm);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001754 if (error)
1755 goto failed;
1756
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001757 spin_lock_irq(&info->lock);
1758 info->swapped--;
1759 shmem_recalc_inode(inode);
1760 spin_unlock_irq(&info->lock);
1761
1762 if (sgp == SGP_WRITE)
1763 mark_page_accessed(page);
1764
1765 delete_from_swap_cache(page);
1766 set_page_dirty(page);
1767 swap_free(swap);
1768
1769 *pagep = page;
1770 return 0;
1771failed:
1772 if (!shmem_confirm_swap(mapping, index, swap))
1773 error = -EEXIST;
1774unlock:
1775 if (page) {
1776 unlock_page(page);
1777 put_page(page);
1778 }
1779
1780 return error;
1781}
1782
1783/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001784 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 *
1786 * If we allocate a new one we do not mark it dirty. That's up to the
1787 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001788 * entry since a page cannot live in both the swap and page cache.
1789 *
Miles Chen28eb3c802019-09-23 15:39:34 -07001790 * vmf and fault_type are only supplied by shmem_fault:
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001791 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001793static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001794 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
Souptick Joarder2b740302018-08-23 17:01:36 -07001795 struct vm_area_struct *vma, struct vm_fault *vmf,
1796 vm_fault_t *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797{
1798 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann23f919d2016-12-12 16:42:28 -08001799 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001801 struct mm_struct *charge_mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001802 struct page *page;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001803 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001804 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001806 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001807 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001809 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001811 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1812 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813repeat:
Hugh Dickins75edd342016-05-19 17:12:44 -07001814 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001815 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001816 return -EINVAL;
1817 }
1818
1819 sbinfo = SHMEM_SB(inode->i_sb);
1820 charge_mm = vma ? vma->vm_mm : current->mm;
1821
1822 page = find_lock_entry(mapping, index);
1823 if (xa_is_value(page)) {
1824 error = shmem_swapin_page(inode, index, &page,
1825 sgp, gfp, vma, fault_type);
1826 if (error == -EEXIST)
1827 goto repeat;
1828
1829 *pagep = page;
1830 return error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001831 }
1832
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001833 if (page)
1834 hindex = page->index;
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001835 if (page && sgp == SGP_WRITE)
1836 mark_page_accessed(page);
1837
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001838 /* fallocated page? */
1839 if (page && !PageUptodate(page)) {
1840 if (sgp != SGP_READ)
1841 goto clear;
1842 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001843 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001844 page = NULL;
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001845 hindex = index;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001846 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001847 if (page || sgp == SGP_READ)
1848 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
1850 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001851 * Fast cache lookup did not find it:
1852 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001854
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001855 if (vma && userfaultfd_missing(vma)) {
1856 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1857 return 0;
1858 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001860 /* shmem_symlink() */
Hui Su30e6a512020-12-14 19:06:14 -08001861 if (!shmem_mapping(mapping))
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001862 goto alloc_nohuge;
1863 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1864 goto alloc_nohuge;
1865 if (shmem_huge == SHMEM_HUGE_FORCE)
1866 goto alloc_huge;
1867 switch (sbinfo->huge) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001868 case SHMEM_HUGE_NEVER:
1869 goto alloc_nohuge;
Kees Cook27d80fa22020-04-06 20:07:51 -07001870 case SHMEM_HUGE_WITHIN_SIZE: {
1871 loff_t i_size;
1872 pgoff_t off;
1873
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001874 off = round_up(index, HPAGE_PMD_NR);
1875 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1876 if (i_size >= HPAGE_PMD_SIZE &&
1877 i_size >> PAGE_SHIFT >= off)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001878 goto alloc_huge;
Kees Cook27d80fa22020-04-06 20:07:51 -07001879
1880 fallthrough;
1881 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001882 case SHMEM_HUGE_ADVISE:
1883 if (sgp_huge == SGP_HUGE)
1884 goto alloc_huge;
1885 /* TODO: implement fadvise() hints */
1886 goto alloc_nohuge;
1887 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001889alloc_huge:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001890 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1891 if (IS_ERR(page)) {
1892alloc_nohuge:
1893 page = shmem_alloc_and_acct_page(gfp, inode,
1894 index, false);
1895 }
1896 if (IS_ERR(page)) {
1897 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001898
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001899 error = PTR_ERR(page);
1900 page = NULL;
1901 if (error != -ENOSPC)
1902 goto unlock;
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001903 /*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001904 * Try to reclaim some space by splitting a huge page
1905 * beyond i_size on the filesystem.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001906 */
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001907 while (retry--) {
1908 int ret;
1909
1910 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1911 if (ret == SHRINK_STOP)
1912 break;
1913 if (ret)
1914 goto alloc_nohuge;
1915 }
1916 goto unlock;
1917 }
1918
1919 if (PageTransHuge(page))
1920 hindex = round_down(index, HPAGE_PMD_NR);
1921 else
1922 hindex = index;
1923
1924 if (sgp == SGP_WRITE)
1925 __SetPageReferenced(page);
1926
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001927 error = shmem_add_to_page_cache(page, mapping, hindex,
Johannes Weiner3fea5a42020-06-03 16:01:41 -07001928 NULL, gfp & GFP_RECLAIM_MASK,
1929 charge_mm);
1930 if (error)
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001931 goto unacct;
Johannes Weiner6058eae2020-06-03 16:02:40 -07001932 lru_cache_add(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001933
1934 spin_lock_irq(&info->lock);
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001935 info->alloced += compound_nr(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001936 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1937 shmem_recalc_inode(inode);
1938 spin_unlock_irq(&info->lock);
1939 alloced = true;
1940
1941 if (PageTransHuge(page) &&
1942 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1943 hindex + HPAGE_PMD_NR - 1) {
1944 /*
1945 * Part of the huge page is beyond i_size: subject
1946 * to shrink under memory pressure.
1947 */
1948 spin_lock(&sbinfo->shrinklist_lock);
1949 /*
1950 * _careful to defend against unlocked access to
1951 * ->shrink_list in shmem_unused_huge_shrink()
1952 */
1953 if (list_empty_careful(&info->shrinklist)) {
1954 list_add_tail(&info->shrinklist,
1955 &sbinfo->shrinklist);
1956 sbinfo->shrinklist_len++;
1957 }
1958 spin_unlock(&sbinfo->shrinklist_lock);
1959 }
1960
1961 /*
1962 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1963 */
1964 if (sgp == SGP_FALLOC)
1965 sgp = SGP_WRITE;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001966clear:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001967 /*
1968 * Let SGP_WRITE caller clear ends if write does not fill page;
1969 * but SGP_FALLOC on a page fallocated earlier must initialize
1970 * it now, lest undo on failure cancel our earlier guarantee.
1971 */
1972 if (sgp != SGP_WRITE && !PageUptodate(page)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001973 int i;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001974
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001975 for (i = 0; i < compound_nr(page); i++) {
1976 clear_highpage(page + i);
1977 flush_dcache_page(page + i);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001978 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001979 SetPageUptodate(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001980 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001981
Hugh Dickins54af60422011-08-03 16:21:24 -07001982 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001983 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001984 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001985 if (alloced) {
1986 ClearPageDirty(page);
1987 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001988 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001989 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001990 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001991 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001992 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001993 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001994 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001995out:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001996 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001997 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001998
Nick Piggind0217ac2007-07-19 01:47:03 -07001999 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002000 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 */
Hugh Dickins54af60422011-08-03 16:21:24 -07002002unacct:
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07002003 shmem_inode_unacct_blocks(inode, compound_nr(page));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002004
2005 if (PageTransHuge(page)) {
2006 unlock_page(page);
2007 put_page(page);
2008 goto alloc_nohuge;
2009 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07002010unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07002011 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07002012 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002013 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07002014 }
2015 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002016 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07002017 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002018 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07002019 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002020 }
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002021 if (error == -EEXIST)
Hugh Dickins54af60422011-08-03 16:21:24 -07002022 goto repeat;
2023 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024}
2025
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002026/*
2027 * This is like autoremove_wake_function, but it removes the wait queue
2028 * entry unconditionally - even if something else had already woken the
2029 * target.
2030 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02002031static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002032{
2033 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02002034 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002035 return ret;
2036}
2037
Souptick Joarder20acce62018-06-07 17:09:17 -07002038static vm_fault_t shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039{
Dave Jiang11bac802017-02-24 14:56:41 -08002040 struct vm_area_struct *vma = vmf->vma;
Al Viro496ad9a2013-01-23 17:07:38 -05002041 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002042 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002043 enum sgp_type sgp;
Souptick Joarder20acce62018-06-07 17:09:17 -07002044 int err;
2045 vm_fault_t ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002047 /*
2048 * Trinity finds that probing a hole which tmpfs is punching can
2049 * prevent the hole-punch from ever completing: which in turn
2050 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07002051 * faulting pages into the hole while it's being punched. Although
2052 * shmem_undo_range() does remove the additions, it may be unable to
2053 * keep up, as each new page needs its own unmap_mapping_range() call,
2054 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2055 *
2056 * It does not matter if we sometimes reach this check just before the
2057 * hole-punch begins, so that one fault then races with the punch:
2058 * we just need to make racing faults a rare case.
2059 *
2060 * The implementation below would be much simpler if we just used a
2061 * standard mutex or completion: but we cannot take i_mutex in fault,
2062 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002063 */
2064 if (unlikely(inode->i_private)) {
2065 struct shmem_falloc *shmem_falloc;
2066
2067 spin_lock(&inode->i_lock);
2068 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002069 if (shmem_falloc &&
2070 shmem_falloc->waitq &&
2071 vmf->pgoff >= shmem_falloc->start &&
2072 vmf->pgoff < shmem_falloc->next) {
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002073 struct file *fpin;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002074 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002075 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002076
2077 ret = VM_FAULT_NOPAGE;
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002078 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2079 if (fpin)
Hugh Dickins8e205f72014-07-23 14:00:10 -07002080 ret = VM_FAULT_RETRY;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002081
2082 shmem_falloc_waitq = shmem_falloc->waitq;
2083 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2084 TASK_UNINTERRUPTIBLE);
2085 spin_unlock(&inode->i_lock);
2086 schedule();
2087
2088 /*
2089 * shmem_falloc_waitq points into the shmem_fallocate()
2090 * stack of the hole-punching task: shmem_falloc_waitq
2091 * is usually invalid by the time we reach here, but
2092 * finish_wait() does not dereference it in that case;
2093 * though i_lock needed lest racing with wake_up_all().
2094 */
2095 spin_lock(&inode->i_lock);
2096 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2097 spin_unlock(&inode->i_lock);
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002098
2099 if (fpin)
2100 fput(fpin);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002101 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002102 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07002103 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002104 }
2105
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002106 sgp = SGP_CACHE;
Michal Hocko18600332017-07-10 15:48:02 -07002107
2108 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2109 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002110 sgp = SGP_NOHUGE;
Michal Hocko18600332017-07-10 15:48:02 -07002111 else if (vma->vm_flags & VM_HUGEPAGE)
2112 sgp = SGP_HUGE;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002113
Souptick Joarder20acce62018-06-07 17:09:17 -07002114 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08002115 gfp, vma, vmf, &ret);
Souptick Joarder20acce62018-06-07 17:09:17 -07002116 if (err)
2117 return vmf_error(err);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002118 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119}
2120
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002121unsigned long shmem_get_unmapped_area(struct file *file,
2122 unsigned long uaddr, unsigned long len,
2123 unsigned long pgoff, unsigned long flags)
2124{
2125 unsigned long (*get_area)(struct file *,
2126 unsigned long, unsigned long, unsigned long, unsigned long);
2127 unsigned long addr;
2128 unsigned long offset;
2129 unsigned long inflated_len;
2130 unsigned long inflated_addr;
2131 unsigned long inflated_offset;
2132
2133 if (len > TASK_SIZE)
2134 return -ENOMEM;
2135
2136 get_area = current->mm->get_unmapped_area;
2137 addr = get_area(file, uaddr, len, pgoff, flags);
2138
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002139 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002140 return addr;
2141 if (IS_ERR_VALUE(addr))
2142 return addr;
2143 if (addr & ~PAGE_MASK)
2144 return addr;
2145 if (addr > TASK_SIZE - len)
2146 return addr;
2147
2148 if (shmem_huge == SHMEM_HUGE_DENY)
2149 return addr;
2150 if (len < HPAGE_PMD_SIZE)
2151 return addr;
2152 if (flags & MAP_FIXED)
2153 return addr;
2154 /*
2155 * Our priority is to support MAP_SHARED mapped hugely;
2156 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002157 * But if caller specified an address hint and we allocated area there
2158 * successfully, respect that as before.
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002159 */
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002160 if (uaddr == addr)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002161 return addr;
2162
2163 if (shmem_huge != SHMEM_HUGE_FORCE) {
2164 struct super_block *sb;
2165
2166 if (file) {
2167 VM_BUG_ON(file->f_op != &shmem_file_operations);
2168 sb = file_inode(file)->i_sb;
2169 } else {
2170 /*
2171 * Called directly from mm/mmap.c, or drivers/char/mem.c
2172 * for "/dev/zero", to create a shared anonymous object.
2173 */
2174 if (IS_ERR(shm_mnt))
2175 return addr;
2176 sb = shm_mnt->mnt_sb;
2177 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002178 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002179 return addr;
2180 }
2181
2182 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2183 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2184 return addr;
2185 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2186 return addr;
2187
2188 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2189 if (inflated_len > TASK_SIZE)
2190 return addr;
2191 if (inflated_len < len)
2192 return addr;
2193
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002194 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002195 if (IS_ERR_VALUE(inflated_addr))
2196 return addr;
2197 if (inflated_addr & ~PAGE_MASK)
2198 return addr;
2199
2200 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2201 inflated_addr += offset - inflated_offset;
2202 if (inflated_offset > offset)
2203 inflated_addr += HPAGE_PMD_SIZE;
2204
2205 if (inflated_addr > TASK_SIZE - len)
2206 return addr;
2207 return inflated_addr;
2208}
2209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002211static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
Al Viro496ad9a2013-01-23 17:07:38 -05002213 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002214 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215}
2216
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002217static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2218 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
Al Viro496ad9a2013-01-23 17:07:38 -05002220 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002221 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002223 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2224 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225}
2226#endif
2227
2228int shmem_lock(struct file *file, int lock, struct user_struct *user)
2229{
Al Viro496ad9a2013-01-23 17:07:38 -05002230 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 struct shmem_inode_info *info = SHMEM_I(inode);
2232 int retval = -ENOMEM;
2233
Hugh Dickinsea0dfeb2020-04-20 18:14:14 -07002234 /*
2235 * What serializes the accesses to info->flags?
2236 * ipc_lock_object() when called from shmctl_do_lock(),
2237 * no serialization needed when called from shm_destroy().
2238 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 if (lock && !(info->flags & VM_LOCKED)) {
2240 if (!user_shm_lock(inode->i_size, user))
2241 goto out_nomem;
2242 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002243 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 }
2245 if (!lock && (info->flags & VM_LOCKED) && user) {
2246 user_shm_unlock(inode->i_size, user);
2247 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002248 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 }
2250 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252out_nomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return retval;
2254}
2255
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002256static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257{
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002258 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2259
2260 if (info->seals & F_SEAL_FUTURE_WRITE) {
2261 /*
2262 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2263 * "future write" seal active.
2264 */
2265 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2266 return -EPERM;
2267
2268 /*
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002269 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2270 * MAP_SHARED and read-only, take care to not allow mprotect to
2271 * revert protections on such mappings. Do this only for shared
2272 * mappings. For private mappings, don't need to mask
2273 * VM_MAYWRITE as we still want them to be COW-writable.
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002274 */
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002275 if (vma->vm_flags & VM_SHARED)
2276 vma->vm_flags &= ~(VM_MAYWRITE);
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002277 }
2278
Catalin Marinas51b0bff2019-11-29 12:45:08 +00002279 /* arm64 - allow memory tagging on RAM-based files */
2280 vma->vm_flags |= VM_MTE_ALLOWED;
2281
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 file_accessed(file);
2283 vma->vm_ops = &shmem_vm_ops;
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002284 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002285 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2286 (vma->vm_end & HPAGE_PMD_MASK)) {
2287 khugepaged_enter(vma, vma->vm_flags);
2288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 return 0;
2290}
2291
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002292static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002293 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294{
2295 struct inode *inode;
2296 struct shmem_inode_info *info;
2297 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -07002298 ino_t ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299
Chris Downe809d5f2020-08-06 23:20:20 -07002300 if (shmem_reserve_inode(sb, &ino))
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002301 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
2303 inode = new_inode(sb);
2304 if (inode) {
Chris Downe809d5f2020-08-06 23:20:20 -07002305 inode->i_ino = ino;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002306 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002308 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
Arnd Bergmann46c9a942018-08-17 15:45:09 -07002309 inode->i_generation = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 info = SHMEM_I(inode);
2311 memset(info, 0, (char *)inode - (char *)info);
2312 spin_lock_init(&info->lock);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07002313 atomic_set(&info->stop_eviction, 0);
David Herrmann40e041a2014-08-08 14:25:27 -07002314 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002315 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002316 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002318 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002319 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320
2321 switch (mode & S_IFMT) {
2322 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002323 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 init_special_inode(inode, mode, dev);
2325 break;
2326 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002327 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 inode->i_op = &shmem_inode_operations;
2329 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002330 mpol_shared_policy_init(&info->policy,
2331 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 break;
2333 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002334 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 /* Some things misbehave if size == 0 on a directory */
2336 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2337 inode->i_op = &shmem_dir_inode_operations;
2338 inode->i_fop = &simple_dir_operations;
2339 break;
2340 case S_IFLNK:
2341 /*
2342 * Must not load anything in the rbtree,
2343 * mpol_free_shared_policy will not be called.
2344 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002345 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 break;
2347 }
Joel Fernandes (Google)b45d71f2018-09-20 12:22:39 -07002348
2349 lockdep_annotate_inode_mutex_key(inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002350 } else
2351 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 return inode;
2353}
2354
Mike Rapoport8d103962017-09-06 16:23:02 -07002355static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2356 pmd_t *dst_pmd,
2357 struct vm_area_struct *dst_vma,
2358 unsigned long dst_addr,
2359 unsigned long src_addr,
2360 bool zeropage,
2361 struct page **pagep)
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002362{
2363 struct inode *inode = file_inode(dst_vma->vm_file);
2364 struct shmem_inode_info *info = SHMEM_I(inode);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002365 struct address_space *mapping = inode->i_mapping;
2366 gfp_t gfp = mapping_gfp_mask(mapping);
2367 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002368 spinlock_t *ptl;
2369 void *page_kaddr;
2370 struct page *page;
2371 pte_t _dst_pte, *dst_pte;
2372 int ret;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002373 pgoff_t offset, max_off;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002374
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002375 ret = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07002376 if (!shmem_inode_acct_block(inode, 1))
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002377 goto out;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002378
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002379 if (!*pagep) {
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002380 page = shmem_alloc_page(gfp, info, pgoff);
2381 if (!page)
Mike Rapoport0f079692017-09-06 16:22:59 -07002382 goto out_unacct_blocks;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002383
Mike Rapoport8d103962017-09-06 16:23:02 -07002384 if (!zeropage) { /* mcopy_atomic */
2385 page_kaddr = kmap_atomic(page);
2386 ret = copy_from_user(page_kaddr,
2387 (const void __user *)src_addr,
2388 PAGE_SIZE);
2389 kunmap_atomic(page_kaddr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002390
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002391 /* fallback to copy_from_user outside mmap_lock */
Mike Rapoport8d103962017-09-06 16:23:02 -07002392 if (unlikely(ret)) {
2393 *pagep = page;
2394 shmem_inode_unacct_blocks(inode, 1);
2395 /* don't free the page */
Andrea Arcangeli9e368252018-11-30 14:09:25 -08002396 return -ENOENT;
Mike Rapoport8d103962017-09-06 16:23:02 -07002397 }
2398 } else { /* mfill_zeropage_atomic */
2399 clear_highpage(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002400 }
2401 } else {
2402 page = *pagep;
2403 *pagep = NULL;
2404 }
2405
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002406 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2407 __SetPageLocked(page);
2408 __SetPageSwapBacked(page);
Andrea Arcangelia425d352017-02-22 15:43:52 -08002409 __SetPageUptodate(page);
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002410
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002411 ret = -EFAULT;
2412 offset = linear_page_index(dst_vma, dst_addr);
2413 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2414 if (unlikely(offset >= max_off))
2415 goto out_release;
2416
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002417 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2418 gfp & GFP_RECLAIM_MASK, dst_mm);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002419 if (ret)
2420 goto out_release;
2421
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002422 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2423 if (dst_vma->vm_flags & VM_WRITE)
2424 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002425 else {
2426 /*
2427 * We don't set the pte dirty if the vma has no
2428 * VM_WRITE permission, so mark the page dirty or it
2429 * could be freed from under us. We could do it
2430 * unconditionally before unlock_page(), but doing it
2431 * only if VM_WRITE is not set is faster.
2432 */
2433 set_page_dirty(page);
2434 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002435
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002436 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002437
2438 ret = -EFAULT;
2439 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2440 if (unlikely(offset >= max_off))
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002441 goto out_release_unlock;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002442
2443 ret = -EEXIST;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002444 if (!pte_none(*dst_pte))
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002445 goto out_release_unlock;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002446
Johannes Weiner6058eae2020-06-03 16:02:40 -07002447 lru_cache_add(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002448
Yang Shi94b7cc02020-04-20 18:14:17 -07002449 spin_lock_irq(&info->lock);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002450 info->alloced++;
2451 inode->i_blocks += BLOCKS_PER_PAGE;
2452 shmem_recalc_inode(inode);
Yang Shi94b7cc02020-04-20 18:14:17 -07002453 spin_unlock_irq(&info->lock);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002454
2455 inc_mm_counter(dst_mm, mm_counter_file(page));
2456 page_add_file_rmap(page, false);
2457 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2458
2459 /* No need to invalidate - it was non-present before */
2460 update_mmu_cache(dst_vma, dst_addr, dst_pte);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002461 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002462 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002463 ret = 0;
2464out:
2465 return ret;
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002466out_release_unlock:
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002467 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002468 ClearPageDirty(page);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002469 delete_from_page_cache(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002470out_release:
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002471 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002472 put_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002473out_unacct_blocks:
Mike Rapoport0f079692017-09-06 16:22:59 -07002474 shmem_inode_unacct_blocks(inode, 1);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002475 goto out;
2476}
2477
Mike Rapoport8d103962017-09-06 16:23:02 -07002478int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2479 pmd_t *dst_pmd,
2480 struct vm_area_struct *dst_vma,
2481 unsigned long dst_addr,
2482 unsigned long src_addr,
2483 struct page **pagep)
2484{
2485 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2486 dst_addr, src_addr, false, pagep);
2487}
2488
2489int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2490 pmd_t *dst_pmd,
2491 struct vm_area_struct *dst_vma,
2492 unsigned long dst_addr)
2493{
2494 struct page *page = NULL;
2495
2496 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2497 dst_addr, 0, true, &page);
2498}
2499
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002501static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002502static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002504#ifdef CONFIG_TMPFS_XATTR
2505static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2506#else
2507#define shmem_initxattrs NULL
2508#endif
2509
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002511shmem_write_begin(struct file *file, struct address_space *mapping,
2512 loff_t pos, unsigned len, unsigned flags,
2513 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
Nick Piggin800d15a2007-10-16 01:25:03 -07002515 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002516 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002517 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002518
2519 /* i_mutex is held by caller */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002520 if (unlikely(info->seals & (F_SEAL_GROW |
2521 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2522 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
David Herrmann40e041a2014-08-08 14:25:27 -07002523 return -EPERM;
2524 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2525 return -EPERM;
2526 }
2527
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002528 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002529}
2530
2531static int
2532shmem_write_end(struct file *file, struct address_space *mapping,
2533 loff_t pos, unsigned len, unsigned copied,
2534 struct page *page, void *fsdata)
2535{
2536 struct inode *inode = mapping->host;
2537
Hugh Dickinsd3602442008-02-04 22:28:44 -08002538 if (pos + copied > inode->i_size)
2539 i_size_write(inode, pos + copied);
2540
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002541 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002542 struct page *head = compound_head(page);
2543 if (PageTransCompound(page)) {
2544 int i;
2545
2546 for (i = 0; i < HPAGE_PMD_NR; i++) {
2547 if (head + i == page)
2548 continue;
2549 clear_highpage(head + i);
2550 flush_dcache_page(head + i);
2551 }
2552 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002553 if (copied < PAGE_SIZE) {
2554 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002555 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002556 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002557 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002558 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002559 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002560 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002561 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002562 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002563
Nick Piggin800d15a2007-10-16 01:25:03 -07002564 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565}
2566
Al Viro2ba5bbe2014-04-02 20:00:02 -04002567static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568{
Al Viro6e58e792014-02-03 17:07:03 -05002569 struct file *file = iocb->ki_filp;
2570 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002572 pgoff_t index;
2573 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002574 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002575 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002576 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002577 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002578
2579 /*
2580 * Might this read be for a stacking filesystem? Then when reading
2581 * holes of a sparse file, we actually need to allocate those pages,
2582 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2583 */
Al Viro777eda22014-12-17 04:46:46 -05002584 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002585 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002587 index = *ppos >> PAGE_SHIFT;
2588 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
2590 for (;;) {
2591 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002592 pgoff_t end_index;
2593 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 loff_t i_size = i_size_read(inode);
2595
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002596 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 if (index > end_index)
2598 break;
2599 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002600 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 if (nr <= offset)
2602 break;
2603 }
2604
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002605 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002606 if (error) {
2607 if (error == -EINVAL)
2608 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 break;
2610 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002611 if (page) {
2612 if (sgp == SGP_CACHE)
2613 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002614 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 /*
2618 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002619 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002621 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002623 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002625 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 if (nr <= offset) {
2627 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002628 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 break;
2630 }
2631 }
2632 nr -= offset;
2633
2634 if (page) {
2635 /*
2636 * If users can be writing to this page using arbitrary
2637 * virtual addresses, take care about potential aliasing
2638 * before reading the page on the kernel side.
2639 */
2640 if (mapping_writably_mapped(mapping))
2641 flush_dcache_page(page);
2642 /*
2643 * Mark the page accessed if we read the beginning.
2644 */
2645 if (!offset)
2646 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002647 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002649 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651
2652 /*
2653 * Ok, we have the page, and it's up-to-date, so
2654 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002656 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002657 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002659 index += offset >> PAGE_SHIFT;
2660 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002662 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002663 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 break;
Al Viro6e58e792014-02-03 17:07:03 -05002665 if (ret < nr) {
2666 error = -EFAULT;
2667 break;
2668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 cond_resched();
2670 }
2671
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002672 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002673 file_accessed(file);
2674 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675}
2676
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002677/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002678 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002679 */
2680static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002681 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002682{
2683 struct page *page;
2684 struct pagevec pvec;
2685 pgoff_t indices[PAGEVEC_SIZE];
2686 bool done = false;
2687 int i;
2688
Mel Gorman86679822017-11-15 17:37:52 -08002689 pagevec_init(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002690 pvec.nr = 1; /* start small: we may be there already */
2691 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002692 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002693 pvec.nr, pvec.pages, indices);
2694 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002695 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002696 index = end;
2697 break;
2698 }
2699 for (i = 0; i < pvec.nr; i++, index++) {
2700 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002701 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002702 done = true;
2703 break;
2704 }
2705 index = indices[i];
2706 }
2707 page = pvec.pages[i];
Matthew Wilcox3159f942017-11-03 13:30:42 -04002708 if (page && !xa_is_value(page)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002709 if (!PageUptodate(page))
2710 page = NULL;
2711 }
2712 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002713 (page && whence == SEEK_DATA) ||
2714 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002715 done = true;
2716 break;
2717 }
2718 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002719 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002720 pagevec_release(&pvec);
2721 pvec.nr = PAGEVEC_SIZE;
2722 cond_resched();
2723 }
2724 return index;
2725}
2726
Andrew Morton965c8e52012-12-17 15:59:39 -08002727static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002728{
2729 struct address_space *mapping = file->f_mapping;
2730 struct inode *inode = mapping->host;
2731 pgoff_t start, end;
2732 loff_t new_offset;
2733
Andrew Morton965c8e52012-12-17 15:59:39 -08002734 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2735 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002736 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002737 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002738 /* We're holding i_mutex so we can access i_size directly */
2739
Yufen Yu1a413642018-11-16 15:08:39 -08002740 if (offset < 0 || offset >= inode->i_size)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002741 offset = -ENXIO;
2742 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002743 start = offset >> PAGE_SHIFT;
2744 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002745 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002746 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002747 if (new_offset > offset) {
2748 if (new_offset < inode->i_size)
2749 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002750 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002751 offset = -ENXIO;
2752 else
2753 offset = inode->i_size;
2754 }
2755 }
2756
Hugh Dickins387aae62013-08-04 11:30:25 -07002757 if (offset >= 0)
2758 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002759 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002760 return offset;
2761}
2762
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002763static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2764 loff_t len)
2765{
Al Viro496ad9a2013-01-23 17:07:38 -05002766 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002767 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002768 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002769 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002770 pgoff_t start, index, end;
2771 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002772
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002773 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2774 return -EOPNOTSUPP;
2775
Al Viro59551022016-01-22 15:40:57 -05002776 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002777
2778 if (mode & FALLOC_FL_PUNCH_HOLE) {
2779 struct address_space *mapping = file->f_mapping;
2780 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2781 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002782 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002783
David Herrmann40e041a2014-08-08 14:25:27 -07002784 /* protected by i_mutex */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002785 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
David Herrmann40e041a2014-08-08 14:25:27 -07002786 error = -EPERM;
2787 goto out;
2788 }
2789
Hugh Dickins8e205f72014-07-23 14:00:10 -07002790 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junaa71ecd2019-11-30 17:58:11 -08002791 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002792 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2793 spin_lock(&inode->i_lock);
2794 inode->i_private = &shmem_falloc;
2795 spin_unlock(&inode->i_lock);
2796
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002797 if ((u64)unmap_end > (u64)unmap_start)
2798 unmap_mapping_range(mapping, unmap_start,
2799 1 + unmap_end - unmap_start, 0);
2800 shmem_truncate_range(inode, offset, offset + len - 1);
2801 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002802
2803 spin_lock(&inode->i_lock);
2804 inode->i_private = NULL;
2805 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02002806 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002807 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002808 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002809 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002810 }
2811
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002812 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2813 error = inode_newsize_ok(inode, offset + len);
2814 if (error)
2815 goto out;
2816
David Herrmann40e041a2014-08-08 14:25:27 -07002817 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2818 error = -EPERM;
2819 goto out;
2820 }
2821
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002822 start = offset >> PAGE_SHIFT;
2823 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002824 /* Try to avoid a swapstorm if len is impossible to satisfy */
2825 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2826 error = -ENOSPC;
2827 goto out;
2828 }
2829
Hugh Dickins8e205f72014-07-23 14:00:10 -07002830 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002831 shmem_falloc.start = start;
2832 shmem_falloc.next = start;
2833 shmem_falloc.nr_falloced = 0;
2834 shmem_falloc.nr_unswapped = 0;
2835 spin_lock(&inode->i_lock);
2836 inode->i_private = &shmem_falloc;
2837 spin_unlock(&inode->i_lock);
2838
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002839 for (index = start; index < end; index++) {
2840 struct page *page;
2841
2842 /*
2843 * Good, the fallocate(2) manpage permits EINTR: we may have
2844 * been interrupted because we are using up too much memory.
2845 */
2846 if (signal_pending(current))
2847 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002848 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2849 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002850 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002851 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002852 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002853 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002854 if (index > start) {
2855 shmem_undo_range(inode,
2856 (loff_t)start << PAGE_SHIFT,
2857 ((loff_t)index << PAGE_SHIFT) - 1, true);
2858 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002859 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002860 }
2861
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002862 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002863 * Inform shmem_writepage() how far we have reached.
2864 * No need for lock or barrier: we have the page lock.
2865 */
2866 shmem_falloc.next++;
2867 if (!PageUptodate(page))
2868 shmem_falloc.nr_falloced++;
2869
2870 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002871 * If !PageUptodate, leave it that way so that freeable pages
2872 * can be recognized if we need to rollback on error later.
2873 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002874 * than free the pages we are allocating (and SGP_CACHE pages
2875 * might still be clean: we now need to mark those dirty too).
2876 */
2877 set_page_dirty(page);
2878 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002879 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002880 cond_resched();
2881 }
2882
2883 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2884 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002885 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002886undone:
2887 spin_lock(&inode->i_lock);
2888 inode->i_private = NULL;
2889 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002890out:
Al Viro59551022016-01-22 15:40:57 -05002891 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002892 return error;
2893}
2894
David Howells726c3342006-06-23 02:02:58 -07002895static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896{
David Howells726c3342006-06-23 02:02:58 -07002897 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898
2899 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002900 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002902 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002904 buf->f_bavail =
2905 buf->f_bfree = sbinfo->max_blocks -
2906 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002907 }
2908 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 buf->f_files = sbinfo->max_inodes;
2910 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 }
2912 /* else leave those fields 0 like simple_statfs */
2913 return 0;
2914}
2915
2916/*
2917 * File creation. Allocate an inode, and we're done..
2918 */
2919static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002920shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002922 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 int error = -ENOSPC;
2924
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002925 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002927 error = simple_acl_create(dir, inode);
2928 if (error)
2929 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002930 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002931 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002932 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002933 if (error && error != -EOPNOTSUPP)
2934 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002935
Al Viro718deb62009-12-16 19:35:36 -05002936 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002938 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 d_instantiate(dentry, inode);
2940 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 }
2942 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002943out_iput:
2944 iput(inode);
2945 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946}
2947
Al Viro60545d02013-06-07 01:20:27 -04002948static int
2949shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2950{
2951 struct inode *inode;
2952 int error = -ENOSPC;
2953
2954 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2955 if (inode) {
2956 error = security_inode_init_security(inode, dir,
2957 NULL,
2958 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002959 if (error && error != -EOPNOTSUPP)
2960 goto out_iput;
2961 error = simple_acl_create(dir, inode);
2962 if (error)
2963 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002964 d_tmpfile(dentry, inode);
2965 }
2966 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002967out_iput:
2968 iput(inode);
2969 return error;
Al Viro60545d02013-06-07 01:20:27 -04002970}
2971
Al Viro18bb1db2011-07-26 01:41:39 -04002972static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973{
2974 int error;
2975
2976 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2977 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002978 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 return 0;
2980}
2981
Al Viro4acdaf22011-07-26 01:42:34 -04002982static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002983 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984{
2985 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2986}
2987
2988/*
2989 * Link a file..
2990 */
2991static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2992{
David Howells75c3cfa2015-03-17 22:26:12 +00002993 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong29b00e62019-02-22 22:35:32 -08002994 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
2996 /*
2997 * No ordinary (disk based) filesystem counts links as inodes;
2998 * but each new link needs a new dentry, pinning lowmem, and
2999 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wong1062af92019-02-21 08:48:09 -08003000 * But if an O_TMPFILE file is linked into the tmpfs, the
3001 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 */
Darrick J. Wong1062af92019-02-21 08:48:09 -08003003 if (inode->i_nlink) {
Chris Downe809d5f2020-08-06 23:20:20 -07003004 ret = shmem_reserve_inode(inode->i_sb, NULL);
Darrick J. Wong1062af92019-02-21 08:48:09 -08003005 if (ret)
3006 goto out;
3007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
3009 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003010 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07003011 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04003012 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 dget(dentry); /* Extra pinning count for the created dentry */
3014 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003015out:
3016 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017}
3018
3019static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3020{
David Howells75c3cfa2015-03-17 22:26:12 +00003021 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003023 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3024 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
3026 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003027 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003028 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 dput(dentry); /* Undo the count from "create" - this does all the work */
3030 return 0;
3031}
3032
3033static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3034{
3035 if (!simple_empty(dentry))
3036 return -ENOTEMPTY;
3037
David Howells75c3cfa2015-03-17 22:26:12 +00003038 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003039 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 return shmem_unlink(dir, dentry);
3041}
3042
Miklos Szeredi37456772014-07-23 15:15:34 +02003043static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3044{
David Howellse36cb0b2015-01-29 12:02:35 +00003045 bool old_is_dir = d_is_dir(old_dentry);
3046 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003047
3048 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3049 if (old_is_dir) {
3050 drop_nlink(old_dir);
3051 inc_nlink(new_dir);
3052 } else {
3053 drop_nlink(new_dir);
3054 inc_nlink(old_dir);
3055 }
3056 }
3057 old_dir->i_ctime = old_dir->i_mtime =
3058 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003059 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003060 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02003061
3062 return 0;
3063}
3064
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003065static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3066{
3067 struct dentry *whiteout;
3068 int error;
3069
3070 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3071 if (!whiteout)
3072 return -ENOMEM;
3073
3074 error = shmem_mknod(old_dir, whiteout,
3075 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3076 dput(whiteout);
3077 if (error)
3078 return error;
3079
3080 /*
3081 * Cheat and hash the whiteout while the old dentry is still in
3082 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3083 *
3084 * d_lookup() will consistently find one of them at this point,
3085 * not sure which one, but that isn't even important.
3086 */
3087 d_rehash(whiteout);
3088 return 0;
3089}
3090
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091/*
3092 * The VFS layer already does all the dentry stuff for rename,
3093 * we just have to decrement the usage count for the target if
3094 * it exists so that the VFS layer correctly free's it when it
3095 * gets overwritten.
3096 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003097static 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 -07003098{
David Howells75c3cfa2015-03-17 22:26:12 +00003099 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 int they_are_dirs = S_ISDIR(inode->i_mode);
3101
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003102 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003103 return -EINVAL;
3104
Miklos Szeredi37456772014-07-23 15:15:34 +02003105 if (flags & RENAME_EXCHANGE)
3106 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3107
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 if (!simple_empty(new_dentry))
3109 return -ENOTEMPTY;
3110
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003111 if (flags & RENAME_WHITEOUT) {
3112 int error;
3113
3114 error = shmem_whiteout(old_dir, old_dentry);
3115 if (error)
3116 return error;
3117 }
3118
David Howells75c3cfa2015-03-17 22:26:12 +00003119 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003121 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003122 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003123 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003126 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003127 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 }
3129
3130 old_dir->i_size -= BOGO_DIRENT_SIZE;
3131 new_dir->i_size += BOGO_DIRENT_SIZE;
3132 old_dir->i_ctime = old_dir->i_mtime =
3133 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003134 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 return 0;
3136}
3137
3138static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3139{
3140 int error;
3141 int len;
3142 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003143 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144
3145 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003146 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 return -ENAMETOOLONG;
3148
Joe Perches0825a6f2018-06-14 15:27:58 -07003149 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3150 VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 if (!inode)
3152 return -ENOSPC;
3153
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003154 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003155 shmem_initxattrs, NULL);
Mateusz Nosek343c3d72020-04-06 20:07:54 -07003156 if (error && error != -EOPNOTSUPP) {
3157 iput(inode);
3158 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003159 }
3160
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003162 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003163 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3164 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003165 iput(inode);
3166 return -ENOMEM;
3167 }
3168 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003170 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003171 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 if (error) {
3173 iput(inode);
3174 return error;
3175 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003176 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003178 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003179 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003181 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003182 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003185 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 d_instantiate(dentry, inode);
3187 dget(dentry);
3188 return 0;
3189}
3190
Al Virofceef392015-12-29 15:58:39 -05003191static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192{
Al Virofceef392015-12-29 15:58:39 -05003193 mark_page_accessed(arg);
3194 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195}
3196
Al Viro6b255392015-11-17 10:20:54 -05003197static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003198 struct inode *inode,
3199 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003202 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003203 if (!dentry) {
3204 page = find_get_page(inode->i_mapping, 0);
3205 if (!page)
3206 return ERR_PTR(-ECHILD);
3207 if (!PageUptodate(page)) {
3208 put_page(page);
3209 return ERR_PTR(-ECHILD);
3210 }
3211 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003212 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003213 if (error)
3214 return ERR_PTR(error);
3215 unlock_page(page);
3216 }
Al Virofceef392015-12-29 15:58:39 -05003217 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003218 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219}
3220
Eric Parisb09e0fa2011-05-24 17:12:39 -07003221#ifdef CONFIG_TMPFS_XATTR
3222/*
3223 * Superblocks without xattr inode operations may get some security.* xattr
3224 * support from the LSM "for free". As soon as we have any other xattrs
3225 * like ACLs, we also need to implement the security.* handlers at
3226 * filesystem level, though.
3227 */
3228
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003229/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003230 * Callback for security_inode_init_security() for acquiring xattrs.
3231 */
3232static int shmem_initxattrs(struct inode *inode,
3233 const struct xattr *xattr_array,
3234 void *fs_info)
3235{
3236 struct shmem_inode_info *info = SHMEM_I(inode);
3237 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003238 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003239 size_t len;
3240
3241 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003242 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003243 if (!new_xattr)
3244 return -ENOMEM;
3245
3246 len = strlen(xattr->name) + 1;
3247 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3248 GFP_KERNEL);
3249 if (!new_xattr->name) {
Chengguang Xu3bef7352020-07-23 21:15:14 -07003250 kvfree(new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003251 return -ENOMEM;
3252 }
3253
3254 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3255 XATTR_SECURITY_PREFIX_LEN);
3256 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3257 xattr->name, len);
3258
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003259 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003260 }
3261
3262 return 0;
3263}
3264
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003265static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003266 struct dentry *unused, struct inode *inode,
3267 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003268{
Al Virob2968212016-04-10 20:48:24 -04003269 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003270
3271 name = xattr_full_name(handler, name);
3272 return simple_xattr_get(&info->xattrs, name, buffer, size);
3273}
3274
3275static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003276 struct dentry *unused, struct inode *inode,
3277 const char *name, const void *value,
3278 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003279{
Al Viro59301222016-05-27 10:19:30 -04003280 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003281
3282 name = xattr_full_name(handler, name);
Daniel Xua46a2292020-03-12 13:03:15 -07003283 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003284}
3285
3286static const struct xattr_handler shmem_security_xattr_handler = {
3287 .prefix = XATTR_SECURITY_PREFIX,
3288 .get = shmem_xattr_handler_get,
3289 .set = shmem_xattr_handler_set,
3290};
3291
3292static const struct xattr_handler shmem_trusted_xattr_handler = {
3293 .prefix = XATTR_TRUSTED_PREFIX,
3294 .get = shmem_xattr_handler_get,
3295 .set = shmem_xattr_handler_set,
3296};
3297
Eric Parisb09e0fa2011-05-24 17:12:39 -07003298static const struct xattr_handler *shmem_xattr_handlers[] = {
3299#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003300 &posix_acl_access_xattr_handler,
3301 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003302#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003303 &shmem_security_xattr_handler,
3304 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003305 NULL
3306};
3307
Eric Parisb09e0fa2011-05-24 17:12:39 -07003308static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3309{
David Howells75c3cfa2015-03-17 22:26:12 +00003310 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003311 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003312}
3313#endif /* CONFIG_TMPFS_XATTR */
3314
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003315static const struct inode_operations shmem_short_symlink_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003316 .get_link = simple_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,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003319#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320};
3321
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003322static const struct inode_operations shmem_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003323 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003324#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003325 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003326#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003327};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003328
David M. Grimes91828a42006-10-17 00:09:45 -07003329static struct dentry *shmem_get_parent(struct dentry *child)
3330{
3331 return ERR_PTR(-ESTALE);
3332}
3333
3334static int shmem_match(struct inode *ino, void *vfh)
3335{
3336 __u32 *fh = vfh;
3337 __u64 inum = fh[2];
3338 inum = (inum << 32) | fh[1];
3339 return ino->i_ino == inum && fh[0] == ino->i_generation;
3340}
3341
Amir Goldstein12ba7802018-06-07 17:07:15 -07003342/* Find any alias of inode, but prefer a hashed alias */
3343static struct dentry *shmem_find_alias(struct inode *inode)
3344{
3345 struct dentry *alias = d_find_alias(inode);
3346
3347 return alias ?: d_find_any_alias(inode);
3348}
3349
3350
Christoph Hellwig480b1162007-10-21 16:42:13 -07003351static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3352 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003353{
David M. Grimes91828a42006-10-17 00:09:45 -07003354 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003355 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003356 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003357
Christoph Hellwig480b1162007-10-21 16:42:13 -07003358 if (fh_len < 3)
3359 return NULL;
3360
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003361 inum = fid->raw[2];
3362 inum = (inum << 32) | fid->raw[1];
3363
Christoph Hellwig480b1162007-10-21 16:42:13 -07003364 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3365 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003366 if (inode) {
Amir Goldstein12ba7802018-06-07 17:07:15 -07003367 dentry = shmem_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003368 iput(inode);
3369 }
3370
Christoph Hellwig480b1162007-10-21 16:42:13 -07003371 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003372}
3373
Al Virob0b03822012-04-02 14:34:06 -04003374static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3375 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003376{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303377 if (*len < 3) {
3378 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003379 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303380 }
David M. Grimes91828a42006-10-17 00:09:45 -07003381
Al Viro1d3382cb2010-10-23 15:19:20 -04003382 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003383 /* Unfortunately insert_inode_hash is not idempotent,
3384 * so as we hash inodes here rather than at creation
3385 * time, we need a lock to ensure we only try
3386 * to do it once
3387 */
3388 static DEFINE_SPINLOCK(lock);
3389 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003390 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003391 __insert_inode_hash(inode,
3392 inode->i_ino + inode->i_generation);
3393 spin_unlock(&lock);
3394 }
3395
3396 fh[0] = inode->i_generation;
3397 fh[1] = inode->i_ino;
3398 fh[2] = ((__u64)inode->i_ino) >> 32;
3399
3400 *len = 3;
3401 return 1;
3402}
3403
Christoph Hellwig39655162007-10-21 16:42:17 -07003404static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003405 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003406 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003407 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003408};
3409
Al Viro626c3922019-09-08 20:28:06 -04003410enum shmem_param {
3411 Opt_gid,
3412 Opt_huge,
3413 Opt_mode,
3414 Opt_mpol,
3415 Opt_nr_blocks,
3416 Opt_nr_inodes,
3417 Opt_size,
3418 Opt_uid,
Chris Downea3271f2020-08-06 23:20:25 -07003419 Opt_inode32,
3420 Opt_inode64,
Al Viro626c3922019-09-08 20:28:06 -04003421};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422
Al Viro5eede622019-12-16 13:33:32 -05003423static const struct constant_table shmem_param_enums_huge[] = {
Al Viro2710c957a2019-09-06 22:12:08 -04003424 {"never", SHMEM_HUGE_NEVER },
3425 {"always", SHMEM_HUGE_ALWAYS },
3426 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3427 {"advise", SHMEM_HUGE_ADVISE },
Al Viro2710c957a2019-09-06 22:12:08 -04003428 {}
3429};
3430
Al Virod7167b12019-09-07 07:23:15 -04003431const struct fs_parameter_spec shmem_fs_parameters[] = {
Al Viro626c3922019-09-08 20:28:06 -04003432 fsparam_u32 ("gid", Opt_gid),
Al Viro2710c957a2019-09-06 22:12:08 -04003433 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
Al Viro626c3922019-09-08 20:28:06 -04003434 fsparam_u32oct("mode", Opt_mode),
3435 fsparam_string("mpol", Opt_mpol),
3436 fsparam_string("nr_blocks", Opt_nr_blocks),
3437 fsparam_string("nr_inodes", Opt_nr_inodes),
3438 fsparam_string("size", Opt_size),
3439 fsparam_u32 ("uid", Opt_uid),
Chris Downea3271f2020-08-06 23:20:25 -07003440 fsparam_flag ("inode32", Opt_inode32),
3441 fsparam_flag ("inode64", Opt_inode64),
Al Viro626c3922019-09-08 20:28:06 -04003442 {}
3443};
3444
David Howellsf3235622019-03-25 16:38:31 +00003445static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
Al Viro626c3922019-09-08 20:28:06 -04003446{
David Howellsf3235622019-03-25 16:38:31 +00003447 struct shmem_options *ctx = fc->fs_private;
Al Viro626c3922019-09-08 20:28:06 -04003448 struct fs_parse_result result;
3449 unsigned long long size;
3450 char *rest;
3451 int opt;
3452
Al Virod7167b12019-09-07 07:23:15 -04003453 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
David Howellsf3235622019-03-25 16:38:31 +00003454 if (opt < 0)
Al Viro626c3922019-09-08 20:28:06 -04003455 return opt;
Al Viro626c3922019-09-08 20:28:06 -04003456
3457 switch (opt) {
3458 case Opt_size:
3459 size = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003460 if (*rest == '%') {
3461 size <<= PAGE_SHIFT;
3462 size *= totalram_pages();
3463 do_div(size, 100);
3464 rest++;
3465 }
3466 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003467 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003468 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3469 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003470 break;
3471 case Opt_nr_blocks:
3472 ctx->blocks = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003473 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003474 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003475 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003476 break;
3477 case Opt_nr_inodes:
3478 ctx->inodes = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003479 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003480 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003481 ctx->seen |= SHMEM_SEEN_INODES;
Al Viro626c3922019-09-08 20:28:06 -04003482 break;
3483 case Opt_mode:
3484 ctx->mode = result.uint_32 & 07777;
3485 break;
3486 case Opt_uid:
3487 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003488 if (!uid_valid(ctx->uid))
Al Viro626c3922019-09-08 20:28:06 -04003489 goto bad_value;
3490 break;
3491 case Opt_gid:
3492 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003493 if (!gid_valid(ctx->gid))
Al Viro626c3922019-09-08 20:28:06 -04003494 goto bad_value;
3495 break;
3496 case Opt_huge:
3497 ctx->huge = result.uint_32;
3498 if (ctx->huge != SHMEM_HUGE_NEVER &&
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003499 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Al Viro626c3922019-09-08 20:28:06 -04003500 has_transparent_hugepage()))
3501 goto unsupported_parameter;
Al Viroe04dc422019-09-08 19:20:12 -04003502 ctx->seen |= SHMEM_SEEN_HUGE;
Al Viro626c3922019-09-08 20:28:06 -04003503 break;
3504 case Opt_mpol:
3505 if (IS_ENABLED(CONFIG_NUMA)) {
3506 mpol_put(ctx->mpol);
3507 ctx->mpol = NULL;
3508 if (mpol_parse_str(param->string, &ctx->mpol))
3509 goto bad_value;
3510 break;
3511 }
3512 goto unsupported_parameter;
Chris Downea3271f2020-08-06 23:20:25 -07003513 case Opt_inode32:
3514 ctx->full_inums = false;
3515 ctx->seen |= SHMEM_SEEN_INUMS;
3516 break;
3517 case Opt_inode64:
3518 if (sizeof(ino_t) < 8) {
3519 return invalfc(fc,
3520 "Cannot use inode64 with <64bit inums in kernel\n");
3521 }
3522 ctx->full_inums = true;
3523 ctx->seen |= SHMEM_SEEN_INUMS;
3524 break;
Al Viroe04dc422019-09-08 19:20:12 -04003525 }
3526 return 0;
3527
Al Viro626c3922019-09-08 20:28:06 -04003528unsupported_parameter:
Al Virof35aa2b2019-12-21 21:35:55 -05003529 return invalfc(fc, "Unsupported parameter '%s'", param->key);
Al Viro626c3922019-09-08 20:28:06 -04003530bad_value:
Al Virof35aa2b2019-12-21 21:35:55 -05003531 return invalfc(fc, "Bad value for '%s'", param->key);
Al Viroe04dc422019-09-08 19:20:12 -04003532}
3533
David Howellsf3235622019-03-25 16:38:31 +00003534static int shmem_parse_options(struct fs_context *fc, void *data)
Al Viroe04dc422019-09-08 19:20:12 -04003535{
David Howellsf3235622019-03-25 16:38:31 +00003536 char *options = data;
3537
Al Viro33f37c62019-10-09 22:48:01 -04003538 if (options) {
3539 int err = security_sb_eat_lsm_opts(options, &fc->security);
3540 if (err)
3541 return err;
3542 }
3543
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003544 while (options != NULL) {
Al Viro626c3922019-09-08 20:28:06 -04003545 char *this_char = options;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003546 for (;;) {
3547 /*
3548 * NUL-terminate this option: unfortunately,
3549 * mount options form a comma-separated list,
3550 * but mpol's nodelist may also contain commas.
3551 */
3552 options = strchr(options, ',');
3553 if (options == NULL)
3554 break;
3555 options++;
3556 if (!isdigit(*options)) {
3557 options[-1] = '\0';
3558 break;
3559 }
3560 }
Al Viro626c3922019-09-08 20:28:06 -04003561 if (*this_char) {
3562 char *value = strchr(this_char,'=');
David Howellsf3235622019-03-25 16:38:31 +00003563 size_t len = 0;
Al Viro626c3922019-09-08 20:28:06 -04003564 int err;
3565
3566 if (value) {
3567 *value++ = '\0';
David Howellsf3235622019-03-25 16:38:31 +00003568 len = strlen(value);
Al Viro626c3922019-09-08 20:28:06 -04003569 }
David Howellsf3235622019-03-25 16:38:31 +00003570 err = vfs_parse_fs_string(fc, this_char, value, len);
3571 if (err < 0)
3572 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 }
3575 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576}
3577
David Howellsf3235622019-03-25 16:38:31 +00003578/*
3579 * Reconfigure a shmem filesystem.
3580 *
3581 * Note that we disallow change from limited->unlimited blocks/inodes while any
3582 * are in use; but we must separately disallow unlimited->limited, because in
3583 * that case we have no record of how much is already in use.
3584 */
3585static int shmem_reconfigure(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586{
David Howellsf3235622019-03-25 16:38:31 +00003587 struct shmem_options *ctx = fc->fs_private;
3588 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003589 unsigned long inodes;
David Howellsf3235622019-03-25 16:38:31 +00003590 const char *err;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003591
3592 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003593 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
David Howellsf3235622019-03-25 16:38:31 +00003594 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3595 if (!sbinfo->max_blocks) {
3596 err = "Cannot retroactively limit size";
Al Viro0b5071d2019-09-08 18:49:18 -04003597 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003598 }
Al Viro0b5071d2019-09-08 18:49:18 -04003599 if (percpu_counter_compare(&sbinfo->used_blocks,
David Howellsf3235622019-03-25 16:38:31 +00003600 ctx->blocks) > 0) {
3601 err = "Too small a size for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003602 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003603 }
Al Viro0b5071d2019-09-08 18:49:18 -04003604 }
David Howellsf3235622019-03-25 16:38:31 +00003605 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3606 if (!sbinfo->max_inodes) {
3607 err = "Cannot retroactively limit inodes";
Al Viro0b5071d2019-09-08 18:49:18 -04003608 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003609 }
3610 if (ctx->inodes < inodes) {
3611 err = "Too few inodes for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003612 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003613 }
Al Viro0b5071d2019-09-08 18:49:18 -04003614 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003615
Chris Downea3271f2020-08-06 23:20:25 -07003616 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3617 sbinfo->next_ino > UINT_MAX) {
3618 err = "Current inum too high to switch to 32-bit inums";
3619 goto out;
3620 }
3621
David Howellsf3235622019-03-25 16:38:31 +00003622 if (ctx->seen & SHMEM_SEEN_HUGE)
3623 sbinfo->huge = ctx->huge;
Chris Downea3271f2020-08-06 23:20:25 -07003624 if (ctx->seen & SHMEM_SEEN_INUMS)
3625 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00003626 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3627 sbinfo->max_blocks = ctx->blocks;
3628 if (ctx->seen & SHMEM_SEEN_INODES) {
3629 sbinfo->max_inodes = ctx->inodes;
3630 sbinfo->free_inodes = ctx->inodes - inodes;
Al Viro0b5071d2019-09-08 18:49:18 -04003631 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003632
Greg Thelen5f001102013-02-22 16:36:01 -08003633 /*
3634 * Preserve previous mempolicy unless mpol remount option was specified.
3635 */
David Howellsf3235622019-03-25 16:38:31 +00003636 if (ctx->mpol) {
Greg Thelen5f001102013-02-22 16:36:01 -08003637 mpol_put(sbinfo->mpol);
David Howellsf3235622019-03-25 16:38:31 +00003638 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3639 ctx->mpol = NULL;
Greg Thelen5f001102013-02-22 16:36:01 -08003640 }
David Howellsf3235622019-03-25 16:38:31 +00003641 spin_unlock(&sbinfo->stat_lock);
3642 return 0;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003643out:
3644 spin_unlock(&sbinfo->stat_lock);
Al Virof35aa2b2019-12-21 21:35:55 -05003645 return invalfc(fc, "%s", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003647
Al Viro34c80b12011-12-08 21:32:45 -05003648static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003649{
Al Viro34c80b12011-12-08 21:32:45 -05003650 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003651
3652 if (sbinfo->max_blocks != shmem_default_max_blocks())
3653 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003654 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003655 if (sbinfo->max_inodes != shmem_default_max_inodes())
3656 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
Joe Perches0825a6f2018-06-14 15:27:58 -07003657 if (sbinfo->mode != (0777 | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003658 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003659 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3660 seq_printf(seq, ",uid=%u",
3661 from_kuid_munged(&init_user_ns, sbinfo->uid));
3662 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3663 seq_printf(seq, ",gid=%u",
3664 from_kgid_munged(&init_user_ns, sbinfo->gid));
Chris Downea3271f2020-08-06 23:20:25 -07003665
3666 /*
3667 * Showing inode{64,32} might be useful even if it's the system default,
3668 * since then people don't have to resort to checking both here and
3669 * /proc/config.gz to confirm 64-bit inums were successfully applied
3670 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3671 *
3672 * We hide it when inode64 isn't the default and we are using 32-bit
3673 * inodes, since that probably just means the feature isn't even under
3674 * consideration.
3675 *
3676 * As such:
3677 *
3678 * +-----------------+-----------------+
3679 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3680 * +------------------+-----------------+-----------------+
3681 * | full_inums=true | show | show |
3682 * | full_inums=false | show | hide |
3683 * +------------------+-----------------+-----------------+
3684 *
3685 */
3686 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3687 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003688#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003689 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3690 if (sbinfo->huge)
3691 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3692#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003693 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003694 return 0;
3695}
David Herrmann9183df22014-08-08 14:25:29 -07003696
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003697#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698
3699static void shmem_put_super(struct super_block *sb)
3700{
Hugh Dickins602586a2010-08-17 15:23:56 -07003701 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3702
Chris Downe809d5f2020-08-06 23:20:20 -07003703 free_percpu(sbinfo->ino_batch);
Hugh Dickins602586a2010-08-17 15:23:56 -07003704 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003705 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003706 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 sb->s_fs_info = NULL;
3708}
3709
David Howellsf3235622019-03-25 16:38:31 +00003710static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711{
David Howellsf3235622019-03-25 16:38:31 +00003712 struct shmem_options *ctx = fc->fs_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003714 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003715 int err = -ENOMEM;
3716
3717 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003718 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003719 L1_CACHE_BYTES), GFP_KERNEL);
3720 if (!sbinfo)
3721 return -ENOMEM;
3722
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003723 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003725#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 /*
3727 * Per default we only allow half of the physical ram per
3728 * tmpfs instance, limiting inodes to one per page of lowmem;
3729 * but the internal instance is left unlimited.
3730 */
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003731 if (!(sb->s_flags & SB_KERNMOUNT)) {
David Howellsf3235622019-03-25 16:38:31 +00003732 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3733 ctx->blocks = shmem_default_max_blocks();
3734 if (!(ctx->seen & SHMEM_SEEN_INODES))
3735 ctx->inodes = shmem_default_max_inodes();
Chris Downea3271f2020-08-06 23:20:25 -07003736 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3737 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
Al Viroca4e0512013-08-31 12:57:10 -04003738 } else {
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003739 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 }
David M. Grimes91828a42006-10-17 00:09:45 -07003741 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003742 sb->s_flags |= SB_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743#else
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003744 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745#endif
David Howellsf3235622019-03-25 16:38:31 +00003746 sbinfo->max_blocks = ctx->blocks;
3747 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
Chris Downe809d5f2020-08-06 23:20:20 -07003748 if (sb->s_flags & SB_KERNMOUNT) {
3749 sbinfo->ino_batch = alloc_percpu(ino_t);
3750 if (!sbinfo->ino_batch)
3751 goto failed;
3752 }
David Howellsf3235622019-03-25 16:38:31 +00003753 sbinfo->uid = ctx->uid;
3754 sbinfo->gid = ctx->gid;
Chris Downea3271f2020-08-06 23:20:25 -07003755 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00003756 sbinfo->mode = ctx->mode;
3757 sbinfo->huge = ctx->huge;
3758 sbinfo->mpol = ctx->mpol;
3759 ctx->mpol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003761 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003762 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003763 goto failed;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003764 spin_lock_init(&sbinfo->shrinklist_lock);
3765 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003766
Hugh Dickins285b2c42011-08-03 16:21:20 -07003767 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003768 sb->s_blocksize = PAGE_SIZE;
3769 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 sb->s_magic = TMPFS_MAGIC;
3771 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003772 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003773#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003774 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003775#endif
3776#ifdef CONFIG_TMPFS_POSIX_ACL
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003777 sb->s_flags |= SB_POSIXACL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003778#endif
Amir Goldstein2b4db792017-05-18 15:29:33 +03003779 uuid_gen(&sb->s_uuid);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003780
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003781 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782 if (!inode)
3783 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003784 inode->i_uid = sbinfo->uid;
3785 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003786 sb->s_root = d_make_root(inode);
3787 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003788 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789 return 0;
3790
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791failed:
3792 shmem_put_super(sb);
3793 return err;
3794}
3795
David Howellsf3235622019-03-25 16:38:31 +00003796static int shmem_get_tree(struct fs_context *fc)
3797{
3798 return get_tree_nodev(fc, shmem_fill_super);
3799}
3800
3801static void shmem_free_fc(struct fs_context *fc)
3802{
3803 struct shmem_options *ctx = fc->fs_private;
3804
3805 if (ctx) {
3806 mpol_put(ctx->mpol);
3807 kfree(ctx);
3808 }
3809}
3810
3811static const struct fs_context_operations shmem_fs_context_ops = {
3812 .free = shmem_free_fc,
3813 .get_tree = shmem_get_tree,
3814#ifdef CONFIG_TMPFS
3815 .parse_monolithic = shmem_parse_options,
3816 .parse_param = shmem_parse_one,
3817 .reconfigure = shmem_reconfigure,
3818#endif
3819};
3820
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003821static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822
3823static struct inode *shmem_alloc_inode(struct super_block *sb)
3824{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003825 struct shmem_inode_info *info;
3826 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3827 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003829 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830}
3831
Al Viro74b1da52019-04-15 23:19:05 -04003832static void shmem_free_in_core_inode(struct inode *inode)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003833{
Al Viro84e710d2016-04-15 00:58:55 -04003834 if (S_ISLNK(inode->i_mode))
3835 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003836 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3837}
3838
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839static void shmem_destroy_inode(struct inode *inode)
3840{
Al Viro09208d12011-07-26 03:15:03 -04003841 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843}
3844
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003845static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003847 struct shmem_inode_info *info = foo;
3848 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849}
3850
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003851static void shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852{
3853 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3854 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003855 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856}
3857
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003858static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003860 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861}
3862
Hui Su30e6a512020-12-14 19:06:14 -08003863const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003865 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003867 .write_begin = shmem_write_begin,
3868 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003870#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003871 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003872#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003873 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874};
Hui Su30e6a512020-12-14 19:06:14 -08003875EXPORT_SYMBOL(shmem_aops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876
Helge Deller15ad7cd2006-12-06 20:40:36 -08003877static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003879 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003881 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003882 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003883 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003884 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003885 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003886 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003887 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888#endif
3889};
3890
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003891static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003892 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003893 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003894#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003895 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003896 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003897#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898};
3899
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003900static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901#ifdef CONFIG_TMPFS
3902 .create = shmem_create,
3903 .lookup = simple_lookup,
3904 .link = shmem_link,
3905 .unlink = shmem_unlink,
3906 .symlink = shmem_symlink,
3907 .mkdir = shmem_mkdir,
3908 .rmdir = shmem_rmdir,
3909 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003910 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003911 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003913#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003914 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003915#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003916#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003917 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003918 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003919#endif
3920};
3921
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003922static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003923#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003924 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003925#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003926#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003927 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003928 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003929#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930};
3931
Hugh Dickins759b9772007-03-05 00:30:28 -08003932static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 .alloc_inode = shmem_alloc_inode,
Al Viro74b1da52019-04-15 23:19:05 -04003934 .free_inode = shmem_free_in_core_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935 .destroy_inode = shmem_destroy_inode,
3936#ifdef CONFIG_TMPFS
3937 .statfs = shmem_statfs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003938 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939#endif
Al Viro1f895f72010-06-05 19:10:41 -04003940 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 .drop_inode = generic_delete_inode,
3942 .put_super = shmem_put_super,
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003943#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003944 .nr_cached_objects = shmem_unused_huge_count,
3945 .free_cached_objects = shmem_unused_huge_scan,
3946#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947};
3948
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003949static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003950 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003951 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952#ifdef CONFIG_NUMA
3953 .set_policy = shmem_set_policy,
3954 .get_policy = shmem_get_policy,
3955#endif
3956};
3957
David Howellsf3235622019-03-25 16:38:31 +00003958int shmem_init_fs_context(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959{
David Howellsf3235622019-03-25 16:38:31 +00003960 struct shmem_options *ctx;
3961
3962 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3963 if (!ctx)
3964 return -ENOMEM;
3965
3966 ctx->mode = 0777 | S_ISVTX;
3967 ctx->uid = current_fsuid();
3968 ctx->gid = current_fsgid();
3969
3970 fc->fs_private = ctx;
3971 fc->ops = &shmem_fs_context_ops;
3972 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973}
3974
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003975static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 .owner = THIS_MODULE,
3977 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00003978 .init_fs_context = shmem_init_fs_context,
3979#ifdef CONFIG_TMPFS
Al Virod7167b12019-09-07 07:23:15 -04003980 .parameters = shmem_fs_parameters,
David Howellsf3235622019-03-25 16:38:31 +00003981#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 .kill_sb = kill_litter_super,
Matthew Wilcox (Oracle)01c70262020-10-15 20:06:00 -07003983 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003986int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987{
3988 int error;
3989
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003990 shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003992 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003994 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 goto out2;
3996 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003997
Al Viroca4e0512013-08-31 12:57:10 -04003998 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 if (IS_ERR(shm_mnt)) {
4000 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07004001 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 goto out1;
4003 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004004
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004005#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004006 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004007 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4008 else
4009 shmem_huge = 0; /* just in case it was patched */
4010#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011 return 0;
4012
4013out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004014 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004016 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017 shm_mnt = ERR_PTR(error);
4018 return error;
4019}
Matt Mackall853ac432009-01-06 14:40:20 -08004020
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004021#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004022static ssize_t shmem_enabled_show(struct kobject *kobj,
Joe Perches79d4d382020-12-14 19:14:53 -08004023 struct kobj_attribute *attr, char *buf)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004024{
Colin Ian King26083eb62019-11-30 17:58:04 -08004025 static const int values[] = {
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004026 SHMEM_HUGE_ALWAYS,
4027 SHMEM_HUGE_WITHIN_SIZE,
4028 SHMEM_HUGE_ADVISE,
4029 SHMEM_HUGE_NEVER,
4030 SHMEM_HUGE_DENY,
4031 SHMEM_HUGE_FORCE,
4032 };
Joe Perches79d4d382020-12-14 19:14:53 -08004033 int len = 0;
4034 int i;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004035
Joe Perches79d4d382020-12-14 19:14:53 -08004036 for (i = 0; i < ARRAY_SIZE(values); i++) {
4037 len += sysfs_emit_at(buf, len,
4038 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4039 i ? " " : "",
4040 shmem_format_huge(values[i]));
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004041 }
Joe Perches79d4d382020-12-14 19:14:53 -08004042
4043 len += sysfs_emit_at(buf, len, "\n");
4044
4045 return len;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004046}
4047
4048static ssize_t shmem_enabled_store(struct kobject *kobj,
4049 struct kobj_attribute *attr, const char *buf, size_t count)
4050{
4051 char tmp[16];
4052 int huge;
4053
4054 if (count + 1 > sizeof(tmp))
4055 return -EINVAL;
4056 memcpy(tmp, buf, count);
4057 tmp[count] = '\0';
4058 if (count && tmp[count - 1] == '\n')
4059 tmp[count - 1] = '\0';
4060
4061 huge = shmem_parse_huge(tmp);
4062 if (huge == -EINVAL)
4063 return -EINVAL;
4064 if (!has_transparent_hugepage() &&
4065 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4066 return -EINVAL;
4067
4068 shmem_huge = huge;
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004069 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004070 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4071 return count;
4072}
4073
4074struct kobj_attribute shmem_enabled_attr =
4075 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004076#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004077
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004078#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004079bool shmem_huge_enabled(struct vm_area_struct *vma)
4080{
4081 struct inode *inode = file_inode(vma->vm_file);
4082 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4083 loff_t i_size;
4084 pgoff_t off;
4085
Yang Shic0630662019-07-18 15:57:27 -07004086 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4087 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4088 return false;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004089 if (shmem_huge == SHMEM_HUGE_FORCE)
4090 return true;
4091 if (shmem_huge == SHMEM_HUGE_DENY)
4092 return false;
4093 switch (sbinfo->huge) {
4094 case SHMEM_HUGE_NEVER:
4095 return false;
4096 case SHMEM_HUGE_ALWAYS:
4097 return true;
4098 case SHMEM_HUGE_WITHIN_SIZE:
4099 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4100 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4101 if (i_size >= HPAGE_PMD_SIZE &&
4102 i_size >> PAGE_SHIFT >= off)
4103 return true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07004104 fallthrough;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004105 case SHMEM_HUGE_ADVISE:
4106 /* TODO: implement fadvise() hints */
4107 return (vma->vm_flags & VM_HUGEPAGE);
4108 default:
4109 VM_BUG_ON(1);
4110 return false;
4111 }
4112}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004113#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004114
Matt Mackall853ac432009-01-06 14:40:20 -08004115#else /* !CONFIG_SHMEM */
4116
4117/*
4118 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4119 *
4120 * This is intended for small system where the benefits of the full
4121 * shmem code (swap-backed and resource-limited) are outweighed by
4122 * their complexity. On systems without swap this code should be
4123 * effectively equivalent, but much lighter weight.
4124 */
4125
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004126static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004127 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00004128 .init_fs_context = ramfs_init_fs_context,
Al Virod7167b12019-09-07 07:23:15 -04004129 .parameters = ramfs_fs_parameters,
Matt Mackall853ac432009-01-06 14:40:20 -08004130 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004131 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004132};
4133
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004134int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004135{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004136 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004137
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004138 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004139 BUG_ON(IS_ERR(shm_mnt));
4140
4141 return 0;
4142}
4143
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08004144int shmem_unuse(unsigned int type, bool frontswap,
4145 unsigned long *fs_pages_to_unuse)
Matt Mackall853ac432009-01-06 14:40:20 -08004146{
4147 return 0;
4148}
4149
Hugh Dickins3f96b792009-09-21 17:03:37 -07004150int shmem_lock(struct file *file, int lock, struct user_struct *user)
4151{
4152 return 0;
4153}
4154
Hugh Dickins24513262012-01-20 14:34:21 -08004155void shmem_unlock_mapping(struct address_space *mapping)
4156{
4157}
4158
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004159#ifdef CONFIG_MMU
4160unsigned long shmem_get_unmapped_area(struct file *file,
4161 unsigned long addr, unsigned long len,
4162 unsigned long pgoff, unsigned long flags)
4163{
4164 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4165}
4166#endif
4167
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004168void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004169{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004170 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004171}
4172EXPORT_SYMBOL_GPL(shmem_truncate_range);
4173
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004174#define shmem_vm_ops generic_file_vm_ops
4175#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004176#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004177#define shmem_acct_size(flags, size) 0
4178#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004179
4180#endif /* CONFIG_SHMEM */
4181
4182/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183
Matthew Auld703321b2017-10-06 23:18:13 +01004184static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
Eric Parisc7277092013-12-02 11:24:19 +00004185 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187 struct inode *inode;
Al Viro93dec2d2018-07-08 23:02:03 -04004188 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189
Matthew Auld703321b2017-10-06 23:18:13 +01004190 if (IS_ERR(mnt))
4191 return ERR_CAST(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004192
Hugh Dickins285b2c42011-08-03 16:21:20 -07004193 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194 return ERR_PTR(-EINVAL);
4195
4196 if (shmem_acct_size(flags, size))
4197 return ERR_PTR(-ENOMEM);
4198
Al Viro93dec2d2018-07-08 23:02:03 -04004199 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4200 flags);
Al Virodac2d1f2018-06-09 09:27:41 -04004201 if (unlikely(!inode)) {
4202 shmem_unacct_size(flags, size);
4203 return ERR_PTR(-ENOSPC);
4204 }
Eric Parisc7277092013-12-02 11:24:19 +00004205 inode->i_flags |= i_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004207 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004208 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
Al Viro93dec2d2018-07-08 23:02:03 -04004209 if (!IS_ERR(res))
4210 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4211 &shmem_file_operations);
Al Viro26567cd2013-03-01 20:22:53 -05004212 if (IS_ERR(res))
Al Viro93dec2d2018-07-08 23:02:03 -04004213 iput(inode);
Al Viro6b4d0b22013-02-14 21:37:26 -05004214 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215}
Eric Parisc7277092013-12-02 11:24:19 +00004216
4217/**
4218 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4219 * kernel internal. There will be NO LSM permission checks against the
4220 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004221 * higher layer. The users are the big_key and shm implementations. LSM
4222 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004223 * @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_kernel_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, S_PRIVATE);
Eric Parisc7277092013-12-02 11:24:19 +00004230}
4231
4232/**
4233 * shmem_file_setup - get an unlinked file living in tmpfs
4234 * @name: name for dentry (to be seen in /proc/<pid>/maps
4235 * @size: size to be set for the file
4236 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4237 */
4238struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4239{
Matthew Auld703321b2017-10-06 23:18:13 +01004240 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
Eric Parisc7277092013-12-02 11:24:19 +00004241}
Keith Packard395e0dd2008-06-20 00:08:06 -07004242EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243
Randy Dunlap46711812008-03-19 17:00:41 -07004244/**
Matthew Auld703321b2017-10-06 23:18:13 +01004245 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4246 * @mnt: the tmpfs mount where the file will be created
4247 * @name: name for dentry (to be seen in /proc/<pid>/maps
4248 * @size: size to be set for the file
4249 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4250 */
4251struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4252 loff_t size, unsigned long flags)
4253{
4254 return __shmem_file_setup(mnt, name, size, flags, 0);
4255}
4256EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4257
4258/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004259 * shmem_zero_setup - setup a shared anonymous mapping
Peter Collingbourne45e55302020-08-06 23:23:37 -07004260 * @vma: the vma to be mmapped is prepared by do_mmap
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261 */
4262int shmem_zero_setup(struct vm_area_struct *vma)
4263{
4264 struct file *file;
4265 loff_t size = vma->vm_end - vma->vm_start;
4266
Hugh Dickins66fc1302015-06-14 09:48:09 -07004267 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07004268 * Cloning a new file under mmap_lock leads to a lock ordering conflict
Hugh Dickins66fc1302015-06-14 09:48:09 -07004269 * between XFS directory reading and selinux: since this file is only
4270 * accessible to the user through its mapping, use S_PRIVATE flag to
4271 * bypass file security, in the same way as shmem_kernel_file_setup().
4272 */
Matthew Auld703321b2017-10-06 23:18:13 +01004273 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 if (IS_ERR(file))
4275 return PTR_ERR(file);
4276
4277 if (vma->vm_file)
4278 fput(vma->vm_file);
4279 vma->vm_file = file;
4280 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004281
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004282 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004283 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4284 (vma->vm_end & HPAGE_PMD_MASK)) {
4285 khugepaged_enter(vma, vma->vm_flags);
4286 }
4287
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 return 0;
4289}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004290
4291/**
4292 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4293 * @mapping: the page's address_space
4294 * @index: the page index
4295 * @gfp: the page allocator flags to use if allocating
4296 *
4297 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4298 * with any new page allocations done using the specified allocation flags.
4299 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4300 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4301 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4302 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004303 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4304 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004305 */
4306struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4307 pgoff_t index, gfp_t gfp)
4308{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004309#ifdef CONFIG_SHMEM
4310 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004311 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004312 int error;
4313
Hui Su30e6a512020-12-14 19:06:14 -08004314 BUG_ON(!shmem_mapping(mapping));
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004315 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
Mike Rapoportcfda0522017-02-22 15:43:37 -08004316 gfp, NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004317 if (error)
4318 page = ERR_PTR(error);
4319 else
4320 unlock_page(page);
4321 return page;
4322#else
4323 /*
4324 * The tiny !SHMEM case uses ramfs without swap
4325 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004326 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004327#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004328}
4329EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);