blob: 9774c91b15314f700f0f2e263e0433399a849c4b [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 Rapoport4c27fe42017-02-22 15:43:25 -080080#include <linux/rmap.h>
Amir Goldstein2b4db792017-05-18 15:29:33 +030081#include <linux/uuid.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070082
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080083#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Mel Gormandd56b042015-11-06 16:28:43 -080085#include "internal.h"
86
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030087#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/* Pretend that each entry is of this size in directory's i_size */
91#define BOGO_DIRENT_SIZE 20
92
Hugh Dickins69f07ec2011-08-03 16:21:26 -070093/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94#define SHORT_SYMLINK_LEN 128
95
Hugh Dickins1aac1402012-05-29 15:06:42 -070096/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070097 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_mutex making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700100 */
101struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -0700102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -0700103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
107};
108
Al Viro0b5071d2019-09-08 18:49:18 -0400109struct shmem_options {
110 unsigned long long blocks;
111 unsigned long long inodes;
112 struct mempolicy *mpol;
113 kuid_t uid;
114 kgid_t gid;
115 umode_t mode;
Chris Downea3271f2020-08-06 23:20:25 -0700116 bool full_inums;
Al Viro0b5071d2019-09-08 18:49:18 -0400117 int huge;
118 int seen;
119#define SHMEM_SEEN_BLOCKS 1
120#define SHMEM_SEEN_INODES 2
121#define SHMEM_SEEN_HUGE 4
Chris Downea3271f2020-08-06 23:20:25 -0700122#define SHMEM_SEEN_INUMS 8
Al Viro0b5071d2019-09-08 18:49:18 -0400123};
124
Andrew Mortonb76db732008-02-08 04:21:49 -0800125#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800126static unsigned long shmem_default_max_blocks(void)
127{
Arun KSca79b0c2018-12-28 00:34:29 -0800128 return totalram_pages() / 2;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800129}
130
131static unsigned long shmem_default_max_inodes(void)
132{
Arun KSca79b0c2018-12-28 00:34:29 -0800133 unsigned long nr_pages = totalram_pages();
134
135 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800136}
Andrew Mortonb76db732008-02-08 04:21:49 -0800137#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800138
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700139static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
140static int shmem_replace_page(struct page **pagep, gfp_t gfp,
141 struct shmem_inode_info *info, pgoff_t index);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -0800142static int shmem_swapin_page(struct inode *inode, pgoff_t index,
143 struct page **pagep, enum sgp_type sgp,
144 gfp_t gfp, struct vm_area_struct *vma,
145 vm_fault_t *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700146static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700147 struct page **pagep, enum sgp_type sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800148 gfp_t gfp, struct vm_area_struct *vma,
Souptick Joarder2b740302018-08-23 17:01:36 -0700149 struct vm_fault *vmf, vm_fault_t *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700150
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700151int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700152 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700153{
154 return shmem_getpage_gfp(inode, index, pagep, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800155 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700156}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
159{
160 return sb->s_fs_info;
161}
162
163/*
164 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
165 * for shared memory and for shared anonymous (/dev/zero) mappings
166 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
167 * consistent with the pre-accounting of private mappings ...
168 */
169static inline int shmem_acct_size(unsigned long flags, loff_t size)
170{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000171 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000172 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
175static inline void shmem_unacct_size(unsigned long flags, loff_t size)
176{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000177 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 vm_unacct_memory(VM_ACCT(size));
179}
180
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700181static inline int shmem_reacct_size(unsigned long flags,
182 loff_t oldsize, loff_t newsize)
183{
184 if (!(flags & VM_NORESERVE)) {
185 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
186 return security_vm_enough_memory_mm(current->mm,
187 VM_ACCT(newsize) - VM_ACCT(oldsize));
188 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
189 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
190 }
191 return 0;
192}
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/*
195 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700196 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
198 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
199 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700200static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700202 if (!(flags & VM_NORESERVE))
203 return 0;
204
205 return security_vm_enough_memory_mm(current->mm,
206 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static inline void shmem_unacct_blocks(unsigned long flags, long pages)
210{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000211 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300212 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Mike Rapoport0f079692017-09-06 16:22:59 -0700215static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
216{
217 struct shmem_inode_info *info = SHMEM_I(inode);
218 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
219
220 if (shmem_acct_block(info->flags, pages))
221 return false;
222
223 if (sbinfo->max_blocks) {
224 if (percpu_counter_compare(&sbinfo->used_blocks,
225 sbinfo->max_blocks - pages) > 0)
226 goto unacct;
227 percpu_counter_add(&sbinfo->used_blocks, pages);
228 }
229
230 return true;
231
232unacct:
233 shmem_unacct_blocks(info->flags, pages);
234 return false;
235}
236
237static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
238{
239 struct shmem_inode_info *info = SHMEM_I(inode);
240 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
241
242 if (sbinfo->max_blocks)
243 percpu_counter_sub(&sbinfo->used_blocks, pages);
244 shmem_unacct_blocks(info->flags, pages);
245}
246
Hugh Dickins759b9772007-03-05 00:30:28 -0800247static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700248static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800249static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800250static const struct inode_operations shmem_inode_operations;
251static const struct inode_operations shmem_dir_inode_operations;
252static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +0400253static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700254static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Mike Rapoportb0506e42017-02-22 15:43:28 -0800256bool vma_is_shmem(struct vm_area_struct *vma)
257{
258 return vma->vm_ops == &shmem_vm_ops;
259}
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800262static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Chris Downe809d5f2020-08-06 23:20:20 -0700264/*
265 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
266 * produces a novel ino for the newly allocated inode.
267 *
268 * It may also be called when making a hard link to permit the space needed by
269 * each dentry. However, in that case, no new inode number is needed since that
270 * internally draws from another pool of inode numbers (currently global
271 * get_next_ino()). This case is indicated by passing NULL as inop.
272 */
273#define SHMEM_INO_BATCH 1024
274static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800275{
276 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -0700277 ino_t ino;
278
279 if (!(sb->s_flags & SB_KERNMOUNT)) {
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800280 spin_lock(&sbinfo->stat_lock);
Byron Stanoszekbb3e96d2020-09-18 21:20:18 -0700281 if (sbinfo->max_inodes) {
282 if (!sbinfo->free_inodes) {
283 spin_unlock(&sbinfo->stat_lock);
284 return -ENOSPC;
285 }
286 sbinfo->free_inodes--;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800287 }
Chris Downe809d5f2020-08-06 23:20:20 -0700288 if (inop) {
289 ino = sbinfo->next_ino++;
290 if (unlikely(is_zero_ino(ino)))
291 ino = sbinfo->next_ino++;
Chris Downea3271f2020-08-06 23:20:25 -0700292 if (unlikely(!sbinfo->full_inums &&
293 ino > UINT_MAX)) {
Chris Downe809d5f2020-08-06 23:20:20 -0700294 /*
295 * Emulate get_next_ino uint wraparound for
296 * compatibility
297 */
Chris Downea3271f2020-08-06 23:20:25 -0700298 if (IS_ENABLED(CONFIG_64BIT))
299 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
300 __func__, MINOR(sb->s_dev));
301 sbinfo->next_ino = 1;
302 ino = sbinfo->next_ino++;
Chris Downe809d5f2020-08-06 23:20:20 -0700303 }
304 *inop = ino;
305 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800306 spin_unlock(&sbinfo->stat_lock);
Chris Downe809d5f2020-08-06 23:20:20 -0700307 } else if (inop) {
308 /*
309 * __shmem_file_setup, one of our callers, is lock-free: it
310 * doesn't hold stat_lock in shmem_reserve_inode since
311 * max_inodes is always 0, and is called from potentially
312 * unknown contexts. As such, use a per-cpu batched allocator
313 * which doesn't require the per-sb stat_lock unless we are at
314 * the batch boundary.
Chris Downea3271f2020-08-06 23:20:25 -0700315 *
316 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
317 * shmem mounts are not exposed to userspace, so we don't need
318 * to worry about things like glibc compatibility.
Chris Downe809d5f2020-08-06 23:20:20 -0700319 */
320 ino_t *next_ino;
321 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
322 ino = *next_ino;
323 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
324 spin_lock(&sbinfo->stat_lock);
325 ino = sbinfo->next_ino;
326 sbinfo->next_ino += SHMEM_INO_BATCH;
327 spin_unlock(&sbinfo->stat_lock);
328 if (unlikely(is_zero_ino(ino)))
329 ino++;
330 }
331 *inop = ino;
332 *next_ino = ++ino;
333 put_cpu();
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800334 }
Chris Downe809d5f2020-08-06 23:20:20 -0700335
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800336 return 0;
337}
338
339static void shmem_free_inode(struct super_block *sb)
340{
341 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
342 if (sbinfo->max_inodes) {
343 spin_lock(&sbinfo->stat_lock);
344 sbinfo->free_inodes++;
345 spin_unlock(&sbinfo->stat_lock);
346 }
347}
348
Randy Dunlap46711812008-03-19 17:00:41 -0700349/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700350 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 * @inode: inode to recalc
352 *
353 * We have to calculate the free blocks since the mm can drop
354 * undirtied hole pages behind our back.
355 *
356 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
357 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
358 *
359 * It has to be called with the spinlock held.
360 */
361static void shmem_recalc_inode(struct inode *inode)
362{
363 struct shmem_inode_info *info = SHMEM_I(inode);
364 long freed;
365
366 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
367 if (freed > 0) {
368 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700369 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Mike Rapoport0f079692017-09-06 16:22:59 -0700370 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
372}
373
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700374bool shmem_charge(struct inode *inode, long pages)
375{
376 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700377 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700378
Mike Rapoport0f079692017-09-06 16:22:59 -0700379 if (!shmem_inode_acct_block(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700380 return false;
Mike Rapoportb1cc94a2017-09-06 16:22:56 -0700381
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800382 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
383 inode->i_mapping->nrpages += pages;
384
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700385 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700386 info->alloced += pages;
387 inode->i_blocks += pages * BLOCKS_PER_PAGE;
388 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700389 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700390
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700391 return true;
392}
393
394void shmem_uncharge(struct inode *inode, long pages)
395{
396 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700397 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700398
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800399 /* nrpages adjustment done by __delete_from_page_cache() or caller */
400
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700401 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700402 info->alloced -= pages;
403 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
404 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700405 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700406
Mike Rapoport0f079692017-09-06 16:22:59 -0700407 shmem_inode_unacct_blocks(inode, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700408}
409
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700410/*
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500411 * Replace item expected in xarray by a new item, while holding xa_lock.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700412 */
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500413static int shmem_replace_entry(struct address_space *mapping,
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700414 pgoff_t index, void *expected, void *replacement)
415{
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500416 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700417 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700418
419 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700420 VM_BUG_ON(!replacement);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500421 item = xas_load(&xas);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700422 if (item != expected)
423 return -ENOENT;
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500424 xas_store(&xas, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700425 return 0;
426}
427
428/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700429 * Sometimes, before we decide whether to proceed or to fail, we must check
430 * that an entry was not already brought back from swap by a racing thread.
431 *
432 * Checking page is not enough: by the time a SwapCache page is locked, it
433 * might be reused, and again be SwapCache, using the same swap as before.
434 */
435static bool shmem_confirm_swap(struct address_space *mapping,
436 pgoff_t index, swp_entry_t swap)
437{
Matthew Wilcoxa12831b2017-11-22 08:34:58 -0500438 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
Hugh Dickinsd1899222012-07-11 14:02:47 -0700439}
440
441/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700442 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
443 *
444 * SHMEM_HUGE_NEVER:
445 * disables huge pages for the mount;
446 * SHMEM_HUGE_ALWAYS:
447 * enables huge pages for the mount;
448 * SHMEM_HUGE_WITHIN_SIZE:
449 * only allocate huge pages if the page will be fully within i_size,
450 * also respect fadvise()/madvise() hints;
451 * SHMEM_HUGE_ADVISE:
452 * only allocate huge pages if requested with fadvise()/madvise();
453 */
454
455#define SHMEM_HUGE_NEVER 0
456#define SHMEM_HUGE_ALWAYS 1
457#define SHMEM_HUGE_WITHIN_SIZE 2
458#define SHMEM_HUGE_ADVISE 3
459
460/*
461 * Special values.
462 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
463 *
464 * SHMEM_HUGE_DENY:
465 * disables huge on shm_mnt and all mounts, for emergency use;
466 * SHMEM_HUGE_FORCE:
467 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
468 *
469 */
470#define SHMEM_HUGE_DENY (-1)
471#define SHMEM_HUGE_FORCE (-2)
472
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700473#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700474/* ifdef here to avoid bloating shmem.o when not necessary */
475
Mike Kravetz5b9c98f2018-06-07 17:05:53 -0700476static int shmem_huge __read_mostly;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700477
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700478#if defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700479static int shmem_parse_huge(const char *str)
480{
481 if (!strcmp(str, "never"))
482 return SHMEM_HUGE_NEVER;
483 if (!strcmp(str, "always"))
484 return SHMEM_HUGE_ALWAYS;
485 if (!strcmp(str, "within_size"))
486 return SHMEM_HUGE_WITHIN_SIZE;
487 if (!strcmp(str, "advise"))
488 return SHMEM_HUGE_ADVISE;
489 if (!strcmp(str, "deny"))
490 return SHMEM_HUGE_DENY;
491 if (!strcmp(str, "force"))
492 return SHMEM_HUGE_FORCE;
493 return -EINVAL;
494}
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700495#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700496
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700497#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700498static const char *shmem_format_huge(int huge)
499{
500 switch (huge) {
501 case SHMEM_HUGE_NEVER:
502 return "never";
503 case SHMEM_HUGE_ALWAYS:
504 return "always";
505 case SHMEM_HUGE_WITHIN_SIZE:
506 return "within_size";
507 case SHMEM_HUGE_ADVISE:
508 return "advise";
509 case SHMEM_HUGE_DENY:
510 return "deny";
511 case SHMEM_HUGE_FORCE:
512 return "force";
513 default:
514 VM_BUG_ON(1);
515 return "bad_val";
516 }
517}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800518#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700519
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700520static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
521 struct shrink_control *sc, unsigned long nr_to_split)
522{
523 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800524 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700525 struct inode *inode;
526 struct shmem_inode_info *info;
527 struct page *page;
528 unsigned long batch = sc ? sc->nr_to_scan : 128;
529 int removed = 0, split = 0;
530
531 if (list_empty(&sbinfo->shrinklist))
532 return SHRINK_STOP;
533
534 spin_lock(&sbinfo->shrinklist_lock);
535 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
536 info = list_entry(pos, struct shmem_inode_info, shrinklist);
537
538 /* pin the inode */
539 inode = igrab(&info->vfs_inode);
540
541 /* inode is about to be evicted */
542 if (!inode) {
543 list_del_init(&info->shrinklist);
544 removed++;
545 goto next;
546 }
547
548 /* Check if there's anything to gain */
549 if (round_up(inode->i_size, PAGE_SIZE) ==
550 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800551 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700552 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700553 goto next;
554 }
555
556 list_move(&info->shrinklist, &list);
557next:
558 if (!--batch)
559 break;
560 }
561 spin_unlock(&sbinfo->shrinklist_lock);
562
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800563 list_for_each_safe(pos, next, &to_remove) {
564 info = list_entry(pos, struct shmem_inode_info, shrinklist);
565 inode = &info->vfs_inode;
566 list_del_init(&info->shrinklist);
567 iput(inode);
568 }
569
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700570 list_for_each_safe(pos, next, &list) {
571 int ret;
572
573 info = list_entry(pos, struct shmem_inode_info, shrinklist);
574 inode = &info->vfs_inode;
575
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700576 if (nr_to_split && split >= nr_to_split)
577 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700578
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700579 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700580 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
581 if (!page)
582 goto drop;
583
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700584 /* No huge page at the end of the file: nothing to split */
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700585 if (!PageTransHuge(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700586 put_page(page);
587 goto drop;
588 }
589
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700590 /*
591 * Leave the inode on the list if we failed to lock
592 * the page at this time.
593 *
594 * Waiting for the lock may lead to deadlock in the
595 * reclaim path.
596 */
597 if (!trylock_page(page)) {
598 put_page(page);
599 goto leave;
600 }
601
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700602 ret = split_huge_page(page);
603 unlock_page(page);
604 put_page(page);
605
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700606 /* If split failed leave the inode on the list */
607 if (ret)
608 goto leave;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700609
610 split++;
611drop:
612 list_del_init(&info->shrinklist);
613 removed++;
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700614leave:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700615 iput(inode);
616 }
617
618 spin_lock(&sbinfo->shrinklist_lock);
619 list_splice_tail(&list, &sbinfo->shrinklist);
620 sbinfo->shrinklist_len -= removed;
621 spin_unlock(&sbinfo->shrinklist_lock);
622
623 return split;
624}
625
626static long shmem_unused_huge_scan(struct super_block *sb,
627 struct shrink_control *sc)
628{
629 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
630
631 if (!READ_ONCE(sbinfo->shrinklist_len))
632 return SHRINK_STOP;
633
634 return shmem_unused_huge_shrink(sbinfo, sc, 0);
635}
636
637static long shmem_unused_huge_count(struct super_block *sb,
638 struct shrink_control *sc)
639{
640 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
641 return READ_ONCE(sbinfo->shrinklist_len);
642}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700643#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700644
645#define shmem_huge SHMEM_HUGE_DENY
646
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700647static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
648 struct shrink_control *sc, unsigned long nr_to_split)
649{
650 return 0;
651}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700652#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700653
Yang Shi89fdcd22018-06-07 17:06:59 -0700654static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
655{
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700656 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Yang Shi89fdcd22018-06-07 17:06:59 -0700657 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
658 shmem_huge != SHMEM_HUGE_DENY)
659 return true;
660 return false;
661}
662
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700663/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700664 * Like add_to_page_cache_locked, but error if expected item has gone.
665 */
666static int shmem_add_to_page_cache(struct page *page,
667 struct address_space *mapping,
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700668 pgoff_t index, void *expected, gfp_t gfp,
669 struct mm_struct *charge_mm)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700670{
Matthew Wilcox552446a42017-12-01 13:25:14 -0500671 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
672 unsigned long i = 0;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -0700673 unsigned long nr = compound_nr(page);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700674 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700675
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700676 VM_BUG_ON_PAGE(PageTail(page), page);
677 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800678 VM_BUG_ON_PAGE(!PageLocked(page), page);
679 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700680 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700681
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700682 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700683 page->mapping = mapping;
684 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700685
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700686 if (!PageSwapCache(page)) {
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -0700687 error = mem_cgroup_charge(page, charge_mm, gfp);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700688 if (error) {
689 if (PageTransHuge(page)) {
690 count_vm_event(THP_FILE_FALLBACK);
691 count_vm_event(THP_FILE_FALLBACK_CHARGE);
692 }
693 goto error;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700694 }
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700695 }
696 cgroup_throttle_swaprate(page, gfp);
697
Matthew Wilcox552446a42017-12-01 13:25:14 -0500698 do {
699 void *entry;
700 xas_lock_irq(&xas);
701 entry = xas_find_conflict(&xas);
702 if (entry != expected)
703 xas_set_err(&xas, -EEXIST);
704 xas_create_range(&xas);
705 if (xas_error(&xas))
706 goto unlock;
707next:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700708 xas_store(&xas, page);
Matthew Wilcox552446a42017-12-01 13:25:14 -0500709 if (++i < nr) {
710 xas_next(&xas);
711 goto next;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700712 }
Matthew Wilcox552446a42017-12-01 13:25:14 -0500713 if (PageTransHuge(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700714 count_vm_event(THP_FILE_ALLOC);
Mel Gorman11fb9982016-07-28 15:46:20 -0700715 __inc_node_page_state(page, NR_SHMEM_THPS);
Matthew Wilcox552446a42017-12-01 13:25:14 -0500716 }
717 mapping->nrpages += nr;
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700718 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
719 __mod_lruvec_page_state(page, NR_SHMEM, nr);
Matthew Wilcox552446a42017-12-01 13:25:14 -0500720unlock:
721 xas_unlock_irq(&xas);
722 } while (xas_nomem(&xas, gfp));
723
724 if (xas_error(&xas)) {
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700725 error = xas_error(&xas);
726 goto error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700727 }
Matthew Wilcox552446a42017-12-01 13:25:14 -0500728
729 return 0;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700730error:
731 page->mapping = NULL;
732 page_ref_sub(page, nr);
733 return error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700734}
735
736/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700737 * Like delete_from_page_cache, but substitutes swap for page.
738 */
739static void shmem_delete_from_page_cache(struct page *page, void *radswap)
740{
741 struct address_space *mapping = page->mapping;
742 int error;
743
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700744 VM_BUG_ON_PAGE(PageCompound(page), page);
745
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700746 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500747 error = shmem_replace_entry(mapping, page->index, page, radswap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700748 page->mapping = NULL;
749 mapping->nrpages--;
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700750 __dec_lruvec_page_state(page, NR_FILE_PAGES);
751 __dec_lruvec_page_state(page, NR_SHMEM);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700752 xa_unlock_irq(&mapping->i_pages);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300753 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700754 BUG_ON(error);
755}
756
757/*
Matthew Wilcoxc121d3b2017-12-04 03:13:54 -0500758 * Remove swap entry from page cache, free the swap and its page cache.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700759 */
760static int shmem_free_swap(struct address_space *mapping,
761 pgoff_t index, void *radswap)
762{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700763 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700764
Matthew Wilcox55f3f7e2018-11-26 16:08:43 -0500765 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700766 if (old != radswap)
767 return -ENOENT;
768 free_swap_and_cache(radix_to_swp_entry(radswap));
769 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700770}
771
772/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800773 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800774 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800775 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700776 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800777 * as long as the inode doesn't go away and racy results are not a problem.
778 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800779unsigned long shmem_partial_swap_usage(struct address_space *mapping,
780 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800781{
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500782 XA_STATE(xas, &mapping->i_pages, start);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800783 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800784 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800785
786 rcu_read_lock();
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500787 xas_for_each(&xas, page, end - 1) {
788 if (xas_retry(&xas, page))
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700789 continue;
Matthew Wilcox3159f942017-11-03 13:30:42 -0400790 if (xa_is_value(page))
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800791 swapped++;
792
793 if (need_resched()) {
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500794 xas_pause(&xas);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800795 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800796 }
797 }
798
799 rcu_read_unlock();
800
801 return swapped << PAGE_SHIFT;
802}
803
804/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800805 * Determine (in bytes) how many of the shmem object's pages mapped by the
806 * given vma is swapped out.
807 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700808 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
Vlastimil Babka48131e02016-01-14 15:19:23 -0800809 * as long as the inode doesn't go away and racy results are not a problem.
810 */
811unsigned long shmem_swap_usage(struct vm_area_struct *vma)
812{
813 struct inode *inode = file_inode(vma->vm_file);
814 struct shmem_inode_info *info = SHMEM_I(inode);
815 struct address_space *mapping = inode->i_mapping;
816 unsigned long swapped;
817
818 /* Be careful as we don't hold info->lock */
819 swapped = READ_ONCE(info->swapped);
820
821 /*
822 * The easier cases are when the shmem object has nothing in swap, or
823 * the vma maps it whole. Then we can simply use the stats that we
824 * already track.
825 */
826 if (!swapped)
827 return 0;
828
829 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
830 return swapped << PAGE_SHIFT;
831
832 /* Here comes the more involved part */
833 return shmem_partial_swap_usage(mapping,
834 linear_page_index(vma, vma->vm_start),
835 linear_page_index(vma, vma->vm_end));
836}
837
838/*
Hugh Dickins24513262012-01-20 14:34:21 -0800839 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
840 */
841void shmem_unlock_mapping(struct address_space *mapping)
842{
843 struct pagevec pvec;
844 pgoff_t indices[PAGEVEC_SIZE];
845 pgoff_t index = 0;
846
Mel Gorman86679822017-11-15 17:37:52 -0800847 pagevec_init(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800848 /*
849 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
850 */
851 while (!mapping_unevictable(mapping)) {
852 /*
853 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
854 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
855 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700856 pvec.nr = find_get_entries(mapping, index,
857 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800858 if (!pvec.nr)
859 break;
860 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700861 pagevec_remove_exceptionals(&pvec);
Kuo-Hsin Yang64e3d122018-11-06 13:23:24 +0000862 check_move_unevictable_pages(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800863 pagevec_release(&pvec);
864 cond_resched();
865 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700866}
867
868/*
Hugh Dickins71725ed2020-04-06 20:07:57 -0700869 * Check whether a hole-punch or truncation needs to split a huge page,
870 * returning true if no split was required, or the split has been successful.
871 *
872 * Eviction (or truncation to 0 size) should never need to split a huge page;
873 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
874 * head, and then succeeded to trylock on tail.
875 *
876 * A split can only succeed when there are no additional references on the
877 * huge page: so the split below relies upon find_get_entries() having stopped
878 * when it found a subpage of the huge page, without getting further references.
879 */
880static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
881{
882 if (!PageTransCompound(page))
883 return true;
884
885 /* Just proceed to delete a huge page wholly within the range punched */
886 if (PageHead(page) &&
887 page->index >= start && page->index + HPAGE_PMD_NR <= end)
888 return true;
889
890 /* Try to split huge page, so we can truly punch the hole or truncate */
891 return split_huge_page(page) >= 0;
892}
893
894/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -0500895 * Remove range of pages and swap entries from page cache, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700896 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700897 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700898static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
899 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700901 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300903 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
904 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
905 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
906 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700907 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700908 pgoff_t indices[PAGEVEC_SIZE];
909 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700910 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700911 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700913 if (lend == -1)
914 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700915
Mel Gorman86679822017-11-15 17:37:52 -0800916 pagevec_init(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700917 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700918 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700919 pvec.nr = find_get_entries(mapping, index,
920 min(end - index, (pgoff_t)PAGEVEC_SIZE),
921 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700922 if (!pvec.nr)
923 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700924 for (i = 0; i < pagevec_count(&pvec); i++) {
925 struct page *page = pvec.pages[i];
926
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700927 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700928 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700929 break;
930
Matthew Wilcox3159f942017-11-03 13:30:42 -0400931 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700932 if (unfalloc)
933 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700934 nr_swaps_freed += !shmem_free_swap(mapping,
935 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700936 continue;
937 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700938
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700939 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
940
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700941 if (!trylock_page(page))
942 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700943
Hugh Dickins71725ed2020-04-06 20:07:57 -0700944 if ((!unfalloc || !PageUptodate(page)) &&
945 page_mapping(page) == mapping) {
946 VM_BUG_ON_PAGE(PageWriteback(page), page);
947 if (shmem_punch_compound(page, start, end))
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700948 truncate_inode_page(mapping, page);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700949 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700950 unlock_page(page);
951 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700952 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800953 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700954 cond_resched();
955 index++;
956 }
957
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700958 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700959 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700960 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700961 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300962 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700963 if (start > end) {
964 top = partial_end;
965 partial_end = 0;
966 }
967 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700968 set_page_dirty(page);
969 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300970 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700971 }
972 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700973 if (partial_end) {
974 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700975 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700976 if (page) {
977 zero_user_segment(page, 0, partial_end);
978 set_page_dirty(page);
979 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300980 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700981 }
982 }
983 if (start >= end)
984 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700985
986 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700987 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700988 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700989
990 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700991 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700992 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700993 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700994 /* If all gone or hole-punch or unfalloc, we're done */
995 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700996 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700997 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700998 index = start;
999 continue;
1000 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001001 for (i = 0; i < pagevec_count(&pvec); i++) {
1002 struct page *page = pvec.pages[i];
1003
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001004 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001005 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001006 break;
1007
Matthew Wilcox3159f942017-11-03 13:30:42 -04001008 if (xa_is_value(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001009 if (unfalloc)
1010 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001011 if (shmem_free_swap(mapping, index, page)) {
1012 /* Swap was replaced by page: retry */
1013 index--;
1014 break;
1015 }
1016 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001017 continue;
1018 }
1019
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001020 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001021
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001022 if (!unfalloc || !PageUptodate(page)) {
Hugh Dickins71725ed2020-04-06 20:07:57 -07001023 if (page_mapping(page) != mapping) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001024 /* Page was replaced by swap: retry */
1025 unlock_page(page);
1026 index--;
1027 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001028 }
Hugh Dickins71725ed2020-04-06 20:07:57 -07001029 VM_BUG_ON_PAGE(PageWriteback(page), page);
1030 if (shmem_punch_compound(page, start, end))
1031 truncate_inode_page(mapping, page);
Hugh Dickins0783ac95b2020-04-20 18:14:07 -07001032 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
Hugh Dickins71725ed2020-04-06 20:07:57 -07001033 /* Wipe the page and don't get stuck */
1034 clear_highpage(page);
1035 flush_dcache_page(page);
1036 set_page_dirty(page);
1037 if (index <
1038 round_up(start, HPAGE_PMD_NR))
1039 start = index + 1;
1040 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001041 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001042 unlock_page(page);
1043 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001044 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -08001045 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001046 index++;
1047 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001048
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001049 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001050 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001052 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001053}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001055void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1056{
1057 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001058 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059}
Hugh Dickins94c1e622011-06-27 16:18:03 -07001060EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
David Howellsa528d352017-01-31 16:46:22 +00001062static int shmem_getattr(const struct path *path, struct kstat *stat,
1063 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -07001064{
David Howellsa528d352017-01-31 16:46:22 +00001065 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -07001066 struct shmem_inode_info *info = SHMEM_I(inode);
Yang Shi89fdcd22018-06-07 17:06:59 -07001067 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
Yu Zhao44a30222015-09-08 15:03:33 -07001068
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001069 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001070 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001071 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001072 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001073 }
Yu Zhao44a30222015-09-08 15:03:33 -07001074 generic_fillattr(inode, stat);
Yang Shi89fdcd22018-06-07 17:06:59 -07001075
1076 if (is_huge_enabled(sb_info))
1077 stat->blksize = HPAGE_PMD_SIZE;
1078
Yu Zhao44a30222015-09-08 15:03:33 -07001079 return 0;
1080}
1081
Hugh Dickins94c1e622011-06-27 16:18:03 -07001082static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
David Howells75c3cfa2015-03-17 22:26:12 +00001084 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -07001085 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001086 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 int error;
1088
Jan Kara31051c82016-05-26 16:55:18 +02001089 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001090 if (error)
1091 return error;
1092
Hugh Dickins94c1e622011-06-27 16:18:03 -07001093 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1094 loff_t oldsize = inode->i_size;
1095 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001096
David Herrmann40e041a2014-08-08 14:25:27 -07001097 /* protected by i_mutex */
1098 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1099 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1100 return -EPERM;
1101
Hugh Dickins94c1e622011-06-27 16:18:03 -07001102 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001103 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1104 oldsize, newsize);
1105 if (error)
1106 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001107 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001108 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001109 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001110 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001111 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001112 if (oldsize > holebegin)
1113 unmap_mapping_range(inode->i_mapping,
1114 holebegin, 0, 1);
1115 if (info->alloced)
1116 shmem_truncate_range(inode,
1117 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001118 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001119 if (oldsize > holebegin)
1120 unmap_mapping_range(inode->i_mapping,
1121 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001122
1123 /*
1124 * Part of the huge page can be beyond i_size: subject
1125 * to shrink under memory pressure.
1126 */
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001127 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001128 spin_lock(&sbinfo->shrinklist_lock);
Cong Wangd0413532017-08-10 15:24:24 -07001129 /*
1130 * _careful to defend against unlocked access to
1131 * ->shrink_list in shmem_unused_huge_shrink()
1132 */
1133 if (list_empty_careful(&info->shrinklist)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001134 list_add_tail(&info->shrinklist,
1135 &sbinfo->shrinklist);
1136 sbinfo->shrinklist_len++;
1137 }
1138 spin_unlock(&sbinfo->shrinklist_lock);
1139 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
1142
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001143 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001144 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001145 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 return error;
1147}
1148
Al Viro1f895f72010-06-05 19:10:41 -04001149static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001152 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001154 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 shmem_unacct_size(info->flags, inode->i_size);
1156 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001157 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001158 if (!list_empty(&info->shrinklist)) {
1159 spin_lock(&sbinfo->shrinklist_lock);
1160 if (!list_empty(&info->shrinklist)) {
1161 list_del_init(&info->shrinklist);
1162 sbinfo->shrinklist_len--;
1163 }
1164 spin_unlock(&sbinfo->shrinklist_lock);
1165 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001166 while (!list_empty(&info->swaplist)) {
1167 /* Wait while shmem_unuse() is scanning this inode... */
1168 wait_var_event(&info->stop_eviction,
1169 !atomic_read(&info->stop_eviction));
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001170 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001171 /* ...but beware of the race if we peeked too early */
1172 if (!atomic_read(&info->stop_eviction))
1173 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001174 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 }
Al Viro3ed47db2016-01-22 18:08:52 -05001176 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001177
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001178 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001179 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001180 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001181 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182}
1183
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001184extern struct swap_info_struct *swap_info[];
1185
1186static int shmem_find_swap_entries(struct address_space *mapping,
1187 pgoff_t start, unsigned int nr_entries,
1188 struct page **entries, pgoff_t *indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001189 unsigned int type, bool frontswap)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001190{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001191 XA_STATE(xas, &mapping->i_pages, start);
1192 struct page *page;
Hugh Dickins87039542019-04-18 17:49:58 -07001193 swp_entry_t entry;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001194 unsigned int ret = 0;
1195
1196 if (!nr_entries)
1197 return 0;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001198
1199 rcu_read_lock();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001200 xas_for_each(&xas, page, ULONG_MAX) {
1201 if (xas_retry(&xas, page))
Mike Kravetz5b9c98f2018-06-07 17:05:53 -07001202 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001203
1204 if (!xa_is_value(page))
1205 continue;
1206
Hugh Dickins87039542019-04-18 17:49:58 -07001207 entry = radix_to_swp_entry(page);
1208 if (swp_type(entry) != type)
1209 continue;
1210 if (frontswap &&
1211 !frontswap_test(swap_info[type], swp_offset(entry)))
1212 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001213
1214 indices[ret] = xas.xa_index;
1215 entries[ret] = page;
1216
1217 if (need_resched()) {
1218 xas_pause(&xas);
1219 cond_resched_rcu();
1220 }
1221 if (++ret == nr_entries)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001222 break;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001223 }
Matthew Wilcox478922e2016-12-14 15:08:52 -08001224 rcu_read_unlock();
Matthew Wilcoxe21a2952017-11-22 08:36:00 -05001225
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001226 return ret;
1227}
1228
1229/*
1230 * Move the swapped pages for an inode to page cache. Returns the count
1231 * of pages swapped in, or the error in case of failure.
1232 */
1233static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1234 pgoff_t *indices)
1235{
1236 int i = 0;
1237 int ret = 0;
1238 int error = 0;
1239 struct address_space *mapping = inode->i_mapping;
1240
1241 for (i = 0; i < pvec.nr; i++) {
1242 struct page *page = pvec.pages[i];
1243
1244 if (!xa_is_value(page))
1245 continue;
1246 error = shmem_swapin_page(inode, indices[i],
1247 &page, SGP_CACHE,
1248 mapping_gfp_mask(mapping),
1249 NULL, NULL);
1250 if (error == 0) {
1251 unlock_page(page);
1252 put_page(page);
1253 ret++;
1254 }
1255 if (error == -ENOMEM)
1256 break;
1257 error = 0;
1258 }
1259 return error ? error : ret;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001260}
1261
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001262/*
1263 * If swap found in inode, free it and move page from swapcache to filecache.
1264 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001265static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1266 bool frontswap, unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001268 struct address_space *mapping = inode->i_mapping;
1269 pgoff_t start = 0;
1270 struct pagevec pvec;
1271 pgoff_t indices[PAGEVEC_SIZE];
1272 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1273 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001275 pagevec_init(&pvec);
1276 do {
1277 unsigned int nr_entries = PAGEVEC_SIZE;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001278
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001279 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1280 nr_entries = *fs_pages_to_unuse;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001281
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001282 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1283 pvec.pages, indices,
Hugh Dickins87039542019-04-18 17:49:58 -07001284 type, frontswap);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001285 if (pvec.nr == 0) {
1286 ret = 0;
1287 break;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001288 }
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001289
1290 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1291 if (ret < 0)
1292 break;
1293
1294 if (frontswap_partial) {
1295 *fs_pages_to_unuse -= ret;
1296 if (*fs_pages_to_unuse == 0) {
1297 ret = FRONTSWAP_PAGES_UNUSED;
1298 break;
1299 }
1300 }
1301
1302 start = indices[pvec.nr - 1];
1303 } while (true);
1304
1305 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306}
1307
1308/*
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001309 * Read all the shared memory data that resides in the swap
1310 * device 'type' back into memory, so the swap device can be
1311 * unused.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 */
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001313int shmem_unuse(unsigned int type, bool frontswap,
1314 unsigned long *fs_pages_to_unuse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001316 struct shmem_inode_info *info, *next;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001317 int error = 0;
1318
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001319 if (list_empty(&shmem_swaplist))
1320 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001322 mutex_lock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001323 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1324 if (!info->swapped) {
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001325 list_del_init(&info->swaplist);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001326 continue;
1327 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001328 /*
1329 * Drop the swaplist mutex while searching the inode for swap;
1330 * but before doing so, make sure shmem_evict_inode() will not
1331 * remove placeholder inode from swaplist, nor let it be freed
1332 * (igrab() would protect from unlink, but not from unmount).
1333 */
1334 atomic_inc(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001335 mutex_unlock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001336
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001337 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001338 fs_pages_to_unuse);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001339 cond_resched();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001340
1341 mutex_lock(&shmem_swaplist_mutex);
1342 next = list_next_entry(info, swaplist);
1343 if (!info->swapped)
1344 list_del_init(&info->swaplist);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001345 if (atomic_dec_and_test(&info->stop_eviction))
1346 wake_up_var(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001347 if (error)
Hugh Dickins778dd892011-05-11 15:13:37 -07001348 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001350 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001351
Hugh Dickins778dd892011-05-11 15:13:37 -07001352 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
1355/*
1356 * Move the page from the page cache to the swap cache.
1357 */
1358static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1359{
1360 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001363 swp_entry_t swap;
1364 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001366 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 mapping = page->mapping;
1369 index = page->index;
1370 inode = mapping->host;
1371 info = SHMEM_I(inode);
1372 if (info->flags & VM_LOCKED)
1373 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001374 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 goto redirty;
1376
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001377 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001378 * Our capabilities prevent regular writeback or sync from ever calling
1379 * shmem_writepage; but a stacking filesystem might use ->writepage of
1380 * its underlying filesystem, in which case tmpfs should write out to
1381 * swap only in response to memory pressure, and not for the writeback
1382 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001383 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001384 if (!wbc->for_reclaim) {
1385 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1386 goto redirty;
1387 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001388
1389 /*
1390 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1391 * value into swapfile.c, the only way we can correctly account for a
1392 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001393 *
1394 * That's okay for a page already fallocated earlier, but if we have
1395 * not yet completed the fallocation, then (a) we want to keep track
1396 * of this page in case we have to undo it, and (b) it may not be a
1397 * good idea to continue anyway, once we're pushing into swap. So
1398 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001399 */
1400 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001401 if (inode->i_private) {
1402 struct shmem_falloc *shmem_falloc;
1403 spin_lock(&inode->i_lock);
1404 shmem_falloc = inode->i_private;
1405 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001406 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001407 index >= shmem_falloc->start &&
1408 index < shmem_falloc->next)
1409 shmem_falloc->nr_unswapped++;
1410 else
1411 shmem_falloc = NULL;
1412 spin_unlock(&inode->i_lock);
1413 if (shmem_falloc)
1414 goto redirty;
1415 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001416 clear_highpage(page);
1417 flush_dcache_page(page);
1418 SetPageUptodate(page);
1419 }
1420
Huang Ying38d8b4e2017-07-06 15:37:18 -07001421 swap = get_swap_page(page);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001422 if (!swap.val)
1423 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001424
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001425 /*
1426 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001427 * if it's not already there. Do it now before the page is
1428 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001429 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001430 * we've incremented swapped, because shmem_unuse_inode() will
1431 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001432 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001433 mutex_lock(&shmem_swaplist_mutex);
1434 if (list_empty(&info->swaplist))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001435 list_add(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001436
Yang Shi4afab1c2019-11-30 17:58:07 -08001437 if (add_to_swap_cache(page, swap,
Joonsoo Kim3852f672020-08-11 18:30:47 -07001438 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1439 NULL) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001440 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001441 shmem_recalc_inode(inode);
1442 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001443 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001444
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001445 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001446 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1447
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001448 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001449 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001450 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 return 0;
1452 }
1453
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001454 mutex_unlock(&shmem_swaplist_mutex);
Minchan Kim75f6d6d2017-07-06 15:37:21 -07001455 put_swap_page(page, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456redirty:
1457 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001458 if (wbc->for_reclaim)
1459 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1460 unlock_page(page);
1461 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463
Hugh Dickins75edd342016-05-19 17:12:44 -07001464#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001465static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001466{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001467 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001468
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001469 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001470 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001471
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001472 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001473
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001474 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001475}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001476
1477static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1478{
1479 struct mempolicy *mpol = NULL;
1480 if (sbinfo->mpol) {
1481 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1482 mpol = sbinfo->mpol;
1483 mpol_get(mpol);
1484 spin_unlock(&sbinfo->stat_lock);
1485 }
1486 return mpol;
1487}
Hugh Dickins75edd342016-05-19 17:12:44 -07001488#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1489static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1490{
1491}
1492static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1493{
1494 return NULL;
1495}
1496#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1497#ifndef CONFIG_NUMA
1498#define vm_policy vm_private_data
1499#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001500
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001501static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1502 struct shmem_inode_info *info, pgoff_t index)
1503{
1504 /* Create a pseudo vma that just contains the policy */
Kirill A. Shutemov2c4541e2018-07-26 16:37:30 -07001505 vma_init(vma, NULL);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001506 /* Bias interleave by inode number to distribute better across nodes */
1507 vma->vm_pgoff = index + info->vfs_inode.i_ino;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001508 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1509}
1510
1511static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1512{
1513 /* Drop reference taken by mpol_shared_policy_lookup() */
1514 mpol_cond_put(vma->vm_policy);
1515}
1516
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001517static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1518 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001521 struct page *page;
Will Deaconcab48b22021-01-14 15:42:14 +00001522 struct vm_fault vmf = {
1523 .vma = &pvma,
1524 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001526 shmem_pseudo_vma_init(&pvma, info, index);
Minchan Kime9e9b7e2018-04-05 16:23:42 -07001527 page = swap_cluster_readahead(swap, gfp, &vmf);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001528 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001529
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001530 return page;
1531}
Mel Gorman18a2f372012-12-05 14:01:41 -08001532
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001533static struct page *shmem_alloc_hugepage(gfp_t gfp,
1534 struct shmem_inode_info *info, pgoff_t index)
1535{
1536 struct vm_area_struct pvma;
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001537 struct address_space *mapping = info->vfs_inode.i_mapping;
1538 pgoff_t hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001539 struct page *page;
1540
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001541 hindex = round_down(index, HPAGE_PMD_NR);
Matthew Wilcox7b8d0462017-12-01 22:13:06 -05001542 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1543 XA_PRESENT))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001544 return NULL;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001545
1546 shmem_pseudo_vma_init(&pvma, info, hindex);
1547 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
David Rientjes19deb762019-09-04 12:54:20 -07001548 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001549 shmem_pseudo_vma_destroy(&pvma);
1550 if (page)
1551 prep_transhuge_page(page);
David Rientjesdcdf11e2020-04-06 20:04:25 -07001552 else
1553 count_vm_event(THP_FILE_FALLBACK);
Mel Gorman18a2f372012-12-05 14:01:41 -08001554 return page;
1555}
1556
1557static struct page *shmem_alloc_page(gfp_t gfp,
1558 struct shmem_inode_info *info, pgoff_t index)
1559{
1560 struct vm_area_struct pvma;
1561 struct page *page;
1562
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001563 shmem_pseudo_vma_init(&pvma, info, index);
1564 page = alloc_page_vma(gfp, &pvma, 0);
1565 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001566
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001567 return page;
1568}
1569
1570static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
Mike Rapoport0f079692017-09-06 16:22:59 -07001571 struct inode *inode,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001572 pgoff_t index, bool huge)
1573{
Mike Rapoport0f079692017-09-06 16:22:59 -07001574 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001575 struct page *page;
1576 int nr;
1577 int err = -ENOSPC;
1578
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001579 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001580 huge = false;
1581 nr = huge ? HPAGE_PMD_NR : 1;
1582
Mike Rapoport0f079692017-09-06 16:22:59 -07001583 if (!shmem_inode_acct_block(inode, nr))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001584 goto failed;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001585
1586 if (huge)
1587 page = shmem_alloc_hugepage(gfp, info, index);
1588 else
1589 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001590 if (page) {
1591 __SetPageLocked(page);
1592 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001593 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001594 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001595
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001596 err = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07001597 shmem_inode_unacct_blocks(inode, nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001598failed:
1599 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001603 * When a page is moved from swapcache to shmem filecache (either by the
1604 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1605 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1606 * ignorance of the mapping it belongs to. If that mapping has special
1607 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1608 * we may need to copy to a suitable page before moving to filecache.
1609 *
1610 * In a future release, this may well be extended to respect cpuset and
1611 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1612 * but for now it is a simple matter of zone.
1613 */
1614static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1615{
1616 return page_zonenum(page) > gfp_zone(gfp);
1617}
1618
1619static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1620 struct shmem_inode_info *info, pgoff_t index)
1621{
1622 struct page *oldpage, *newpage;
1623 struct address_space *swap_mapping;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001624 swp_entry_t entry;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001625 pgoff_t swap_index;
1626 int error;
1627
1628 oldpage = *pagep;
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001629 entry.val = page_private(oldpage);
1630 swap_index = swp_offset(entry);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001631 swap_mapping = page_mapping(oldpage);
1632
1633 /*
1634 * We have arrived here because our zones are constrained, so don't
1635 * limit chance of success by further cpuset and node constraints.
1636 */
1637 gfp &= ~GFP_CONSTRAINT_MASK;
1638 newpage = shmem_alloc_page(gfp, info, index);
1639 if (!newpage)
1640 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001641
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001642 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001643 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001644 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001645
Hugh Dickins9956edf2016-11-10 10:46:11 -08001646 __SetPageLocked(newpage);
1647 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001648 SetPageUptodate(newpage);
Yu Zhaoc1cb20d2018-11-30 14:09:03 -08001649 set_page_private(newpage, entry.val);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001650 SetPageSwapCache(newpage);
1651
1652 /*
1653 * Our caller will very soon move newpage out of swapcache, but it's
1654 * a nice clean interface for us to replace oldpage by newpage there.
1655 */
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001656 xa_lock_irq(&swap_mapping->i_pages);
Matthew Wilcox62f945b2017-11-17 10:22:37 -05001657 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001658 if (!error) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07001659 mem_cgroup_migrate(oldpage, newpage);
1660 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1661 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001662 }
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001663 xa_unlock_irq(&swap_mapping->i_pages);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001664
Hugh Dickins0142ef62012-06-07 14:21:09 -07001665 if (unlikely(error)) {
1666 /*
1667 * Is this possible? I think not, now that our callers check
1668 * both PageSwapCache and page_private after getting page lock;
1669 * but be defensive. Reverse old to newpage for clear and free.
1670 */
1671 oldpage = newpage;
1672 } else {
Johannes Weiner6058eae2020-06-03 16:02:40 -07001673 lru_cache_add(newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001674 *pagep = newpage;
1675 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001676
1677 ClearPageSwapCache(oldpage);
1678 set_page_private(oldpage, 0);
1679
1680 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001681 put_page(oldpage);
1682 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001683 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001684}
1685
1686/*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001687 * Swap in the page pointed to by *pagep.
1688 * Caller has to make sure that *pagep contains a valid swapped page.
1689 * Returns 0 and the page in pagep if success. On failure, returns the
Randy Dunlapaf44c122020-08-11 18:33:17 -07001690 * error code and NULL in *pagep.
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001691 */
1692static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1693 struct page **pagep, enum sgp_type sgp,
1694 gfp_t gfp, struct vm_area_struct *vma,
1695 vm_fault_t *fault_type)
1696{
1697 struct address_space *mapping = inode->i_mapping;
1698 struct shmem_inode_info *info = SHMEM_I(inode);
1699 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001700 struct page *page;
1701 swp_entry_t swap;
1702 int error;
1703
1704 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1705 swap = radix_to_swp_entry(*pagep);
1706 *pagep = NULL;
1707
1708 /* Look it up and read it in.. */
1709 page = lookup_swap_cache(swap, NULL, 0);
1710 if (!page) {
1711 /* Or update major stats only when swapin succeeds?? */
1712 if (fault_type) {
1713 *fault_type |= VM_FAULT_MAJOR;
1714 count_vm_event(PGMAJFAULT);
1715 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1716 }
1717 /* Here we actually start the io */
1718 page = shmem_swapin(swap, gfp, info, index);
1719 if (!page) {
1720 error = -ENOMEM;
1721 goto failed;
1722 }
1723 }
1724
1725 /* We have to do this with page locked to prevent races */
1726 lock_page(page);
1727 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1728 !shmem_confirm_swap(mapping, index, swap)) {
1729 error = -EEXIST;
1730 goto unlock;
1731 }
1732 if (!PageUptodate(page)) {
1733 error = -EIO;
1734 goto failed;
1735 }
1736 wait_on_page_writeback(page);
1737
Steven Price8a848022020-05-13 16:37:49 +01001738 /*
1739 * Some architectures may have to restore extra metadata to the
1740 * physical page after reading from swap.
1741 */
1742 arch_swap_restore(swap, page);
1743
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001744 if (shmem_should_replace_page(page, gfp)) {
1745 error = shmem_replace_page(&page, gfp, info, index);
1746 if (error)
1747 goto failed;
1748 }
1749
Johannes Weiner3fea5a42020-06-03 16:01:41 -07001750 error = shmem_add_to_page_cache(page, mapping, index,
1751 swp_to_radix_entry(swap), gfp,
1752 charge_mm);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001753 if (error)
1754 goto failed;
1755
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001756 spin_lock_irq(&info->lock);
1757 info->swapped--;
1758 shmem_recalc_inode(inode);
1759 spin_unlock_irq(&info->lock);
1760
1761 if (sgp == SGP_WRITE)
1762 mark_page_accessed(page);
1763
1764 delete_from_swap_cache(page);
1765 set_page_dirty(page);
1766 swap_free(swap);
1767
1768 *pagep = page;
1769 return 0;
1770failed:
1771 if (!shmem_confirm_swap(mapping, index, swap))
1772 error = -EEXIST;
1773unlock:
1774 if (page) {
1775 unlock_page(page);
1776 put_page(page);
1777 }
1778
1779 return error;
1780}
1781
1782/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001783 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 *
1785 * If we allocate a new one we do not mark it dirty. That's up to the
1786 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001787 * entry since a page cannot live in both the swap and page cache.
1788 *
Axel Rasmussend6721232021-03-18 17:01:51 +11001789 * vma, vmf, and fault_type are only supplied by shmem_fault: otherwise they
1790 * are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001792static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001793 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
Souptick Joarder2b740302018-08-23 17:01:36 -07001794 struct vm_area_struct *vma, struct vm_fault *vmf,
1795 vm_fault_t *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796{
1797 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann23f919d2016-12-12 16:42:28 -08001798 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001800 struct mm_struct *charge_mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001801 struct page *page;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001802 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001803 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001805 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001806 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001808 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001810 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1811 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812repeat:
Hugh Dickins75edd342016-05-19 17:12:44 -07001813 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001814 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001815 return -EINVAL;
1816 }
1817
1818 sbinfo = SHMEM_SB(inode->i_sb);
1819 charge_mm = vma ? vma->vm_mm : current->mm;
1820
1821 page = find_lock_entry(mapping, index);
1822 if (xa_is_value(page)) {
1823 error = shmem_swapin_page(inode, index, &page,
1824 sgp, gfp, vma, fault_type);
1825 if (error == -EEXIST)
1826 goto repeat;
1827
1828 *pagep = page;
1829 return error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001830 }
1831
Axel Rasmussend6721232021-03-18 17:01:51 +11001832 if (page && vma && userfaultfd_minor(vma)) {
1833 unlock_page(page);
Axel Rasmussen02732572021-03-22 13:48:35 -07001834 put_page(page);
Axel Rasmussend6721232021-03-18 17:01:51 +11001835 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1836 return 0;
1837 }
1838
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001839 if (page)
1840 hindex = page->index;
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001841 if (page && sgp == SGP_WRITE)
1842 mark_page_accessed(page);
1843
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001844 /* fallocated page? */
1845 if (page && !PageUptodate(page)) {
1846 if (sgp != SGP_READ)
1847 goto clear;
1848 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001849 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001850 page = NULL;
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001851 hindex = index;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001852 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001853 if (page || sgp == SGP_READ)
1854 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
1856 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001857 * Fast cache lookup did not find it:
1858 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001860
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001861 if (vma && userfaultfd_missing(vma)) {
1862 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1863 return 0;
1864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001866 /* shmem_symlink() */
1867 if (mapping->a_ops != &shmem_aops)
1868 goto alloc_nohuge;
1869 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1870 goto alloc_nohuge;
1871 if (shmem_huge == SHMEM_HUGE_FORCE)
1872 goto alloc_huge;
1873 switch (sbinfo->huge) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001874 case SHMEM_HUGE_NEVER:
1875 goto alloc_nohuge;
Kees Cook27d80fa22020-04-06 20:07:51 -07001876 case SHMEM_HUGE_WITHIN_SIZE: {
1877 loff_t i_size;
1878 pgoff_t off;
1879
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001880 off = round_up(index, HPAGE_PMD_NR);
1881 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1882 if (i_size >= HPAGE_PMD_SIZE &&
1883 i_size >> PAGE_SHIFT >= off)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001884 goto alloc_huge;
Kees Cook27d80fa22020-04-06 20:07:51 -07001885
1886 fallthrough;
1887 }
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001888 case SHMEM_HUGE_ADVISE:
1889 if (sgp_huge == SGP_HUGE)
1890 goto alloc_huge;
1891 /* TODO: implement fadvise() hints */
1892 goto alloc_nohuge;
1893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001895alloc_huge:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001896 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1897 if (IS_ERR(page)) {
1898alloc_nohuge:
1899 page = shmem_alloc_and_acct_page(gfp, inode,
1900 index, false);
1901 }
1902 if (IS_ERR(page)) {
1903 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001904
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001905 error = PTR_ERR(page);
1906 page = NULL;
1907 if (error != -ENOSPC)
1908 goto unlock;
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001909 /*
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001910 * Try to reclaim some space by splitting a huge page
1911 * beyond i_size on the filesystem.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001912 */
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001913 while (retry--) {
1914 int ret;
1915
1916 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1917 if (ret == SHRINK_STOP)
1918 break;
1919 if (ret)
1920 goto alloc_nohuge;
1921 }
1922 goto unlock;
1923 }
1924
1925 if (PageTransHuge(page))
1926 hindex = round_down(index, HPAGE_PMD_NR);
1927 else
1928 hindex = index;
1929
1930 if (sgp == SGP_WRITE)
1931 __SetPageReferenced(page);
1932
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001933 error = shmem_add_to_page_cache(page, mapping, hindex,
Johannes Weiner3fea5a42020-06-03 16:01:41 -07001934 NULL, gfp & GFP_RECLAIM_MASK,
1935 charge_mm);
1936 if (error)
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001937 goto unacct;
Johannes Weiner6058eae2020-06-03 16:02:40 -07001938 lru_cache_add(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001939
1940 spin_lock_irq(&info->lock);
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001941 info->alloced += compound_nr(page);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001942 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1943 shmem_recalc_inode(inode);
1944 spin_unlock_irq(&info->lock);
1945 alloced = true;
1946
1947 if (PageTransHuge(page) &&
1948 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1949 hindex + HPAGE_PMD_NR - 1) {
1950 /*
1951 * Part of the huge page is beyond i_size: subject
1952 * to shrink under memory pressure.
1953 */
1954 spin_lock(&sbinfo->shrinklist_lock);
1955 /*
1956 * _careful to defend against unlocked access to
1957 * ->shrink_list in shmem_unused_huge_shrink()
1958 */
1959 if (list_empty_careful(&info->shrinklist)) {
1960 list_add_tail(&info->shrinklist,
1961 &sbinfo->shrinklist);
1962 sbinfo->shrinklist_len++;
1963 }
1964 spin_unlock(&sbinfo->shrinklist_lock);
1965 }
1966
1967 /*
1968 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1969 */
1970 if (sgp == SGP_FALLOC)
1971 sgp = SGP_WRITE;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001972clear:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001973 /*
1974 * Let SGP_WRITE caller clear ends if write does not fill page;
1975 * but SGP_FALLOC on a page fallocated earlier must initialize
1976 * it now, lest undo on failure cancel our earlier guarantee.
1977 */
1978 if (sgp != SGP_WRITE && !PageUptodate(page)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08001979 int i;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001980
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001981 for (i = 0; i < compound_nr(page); i++) {
1982 clear_highpage(page + i);
1983 flush_dcache_page(page + i);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001984 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07001985 SetPageUptodate(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001986 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001987
Hugh Dickins54af60422011-08-03 16:21:24 -07001988 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001989 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001990 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001991 if (alloced) {
1992 ClearPageDirty(page);
1993 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001994 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001995 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001996 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001997 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001998 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001999 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07002000 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07002001out:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002002 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07002003 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07002004
Nick Piggind0217ac2007-07-19 01:47:03 -07002005 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002006 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 */
Hugh Dickins54af60422011-08-03 16:21:24 -07002008unacct:
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07002009 shmem_inode_unacct_blocks(inode, compound_nr(page));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002010
2011 if (PageTransHuge(page)) {
2012 unlock_page(page);
2013 put_page(page);
2014 goto alloc_nohuge;
2015 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07002016unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07002017 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07002018 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002019 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07002020 }
2021 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002022 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07002023 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002024 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07002025 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002026 }
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002027 if (error == -EEXIST)
Hugh Dickins54af60422011-08-03 16:21:24 -07002028 goto repeat;
2029 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002032/*
2033 * This is like autoremove_wake_function, but it removes the wait queue
2034 * entry unconditionally - even if something else had already woken the
2035 * target.
2036 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02002037static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002038{
2039 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02002040 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002041 return ret;
2042}
2043
Souptick Joarder20acce62018-06-07 17:09:17 -07002044static vm_fault_t shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045{
Dave Jiang11bac802017-02-24 14:56:41 -08002046 struct vm_area_struct *vma = vmf->vma;
Al Viro496ad9a2013-01-23 17:07:38 -05002047 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002048 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002049 enum sgp_type sgp;
Souptick Joarder20acce62018-06-07 17:09:17 -07002050 int err;
2051 vm_fault_t ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002053 /*
2054 * Trinity finds that probing a hole which tmpfs is punching can
2055 * prevent the hole-punch from ever completing: which in turn
2056 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07002057 * faulting pages into the hole while it's being punched. Although
2058 * shmem_undo_range() does remove the additions, it may be unable to
2059 * keep up, as each new page needs its own unmap_mapping_range() call,
2060 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2061 *
2062 * It does not matter if we sometimes reach this check just before the
2063 * hole-punch begins, so that one fault then races with the punch:
2064 * we just need to make racing faults a rare case.
2065 *
2066 * The implementation below would be much simpler if we just used a
2067 * standard mutex or completion: but we cannot take i_mutex in fault,
2068 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002069 */
2070 if (unlikely(inode->i_private)) {
2071 struct shmem_falloc *shmem_falloc;
2072
2073 spin_lock(&inode->i_lock);
2074 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002075 if (shmem_falloc &&
2076 shmem_falloc->waitq &&
2077 vmf->pgoff >= shmem_falloc->start &&
2078 vmf->pgoff < shmem_falloc->next) {
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002079 struct file *fpin;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002080 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002081 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002082
2083 ret = VM_FAULT_NOPAGE;
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002084 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2085 if (fpin)
Hugh Dickins8e205f72014-07-23 14:00:10 -07002086 ret = VM_FAULT_RETRY;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002087
2088 shmem_falloc_waitq = shmem_falloc->waitq;
2089 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2090 TASK_UNINTERRUPTIBLE);
2091 spin_unlock(&inode->i_lock);
2092 schedule();
2093
2094 /*
2095 * shmem_falloc_waitq points into the shmem_fallocate()
2096 * stack of the hole-punching task: shmem_falloc_waitq
2097 * is usually invalid by the time we reach here, but
2098 * finish_wait() does not dereference it in that case;
2099 * though i_lock needed lest racing with wake_up_all().
2100 */
2101 spin_lock(&inode->i_lock);
2102 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2103 spin_unlock(&inode->i_lock);
Kirill A. Shutemov8897c1b2019-11-30 17:50:26 -08002104
2105 if (fpin)
2106 fput(fpin);
Hugh Dickins8e205f72014-07-23 14:00:10 -07002107 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002108 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07002109 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002110 }
2111
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002112 sgp = SGP_CACHE;
Michal Hocko18600332017-07-10 15:48:02 -07002113
2114 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2115 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002116 sgp = SGP_NOHUGE;
Michal Hocko18600332017-07-10 15:48:02 -07002117 else if (vma->vm_flags & VM_HUGEPAGE)
2118 sgp = SGP_HUGE;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07002119
Souptick Joarder20acce62018-06-07 17:09:17 -07002120 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08002121 gfp, vma, vmf, &ret);
Souptick Joarder20acce62018-06-07 17:09:17 -07002122 if (err)
2123 return vmf_error(err);
Hugh Dickins68da9f02011-07-25 17:12:34 -07002124 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125}
2126
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002127unsigned long shmem_get_unmapped_area(struct file *file,
2128 unsigned long uaddr, unsigned long len,
2129 unsigned long pgoff, unsigned long flags)
2130{
2131 unsigned long (*get_area)(struct file *,
2132 unsigned long, unsigned long, unsigned long, unsigned long);
2133 unsigned long addr;
2134 unsigned long offset;
2135 unsigned long inflated_len;
2136 unsigned long inflated_addr;
2137 unsigned long inflated_offset;
2138
2139 if (len > TASK_SIZE)
2140 return -ENOMEM;
2141
2142 get_area = current->mm->get_unmapped_area;
2143 addr = get_area(file, uaddr, len, pgoff, flags);
2144
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002145 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002146 return addr;
2147 if (IS_ERR_VALUE(addr))
2148 return addr;
2149 if (addr & ~PAGE_MASK)
2150 return addr;
2151 if (addr > TASK_SIZE - len)
2152 return addr;
2153
2154 if (shmem_huge == SHMEM_HUGE_DENY)
2155 return addr;
2156 if (len < HPAGE_PMD_SIZE)
2157 return addr;
2158 if (flags & MAP_FIXED)
2159 return addr;
2160 /*
2161 * Our priority is to support MAP_SHARED mapped hugely;
2162 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002163 * But if caller specified an address hint and we allocated area there
2164 * successfully, respect that as before.
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002165 */
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002166 if (uaddr == addr)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002167 return addr;
2168
2169 if (shmem_huge != SHMEM_HUGE_FORCE) {
2170 struct super_block *sb;
2171
2172 if (file) {
2173 VM_BUG_ON(file->f_op != &shmem_file_operations);
2174 sb = file_inode(file)->i_sb;
2175 } else {
2176 /*
2177 * Called directly from mm/mmap.c, or drivers/char/mem.c
2178 * for "/dev/zero", to create a shared anonymous object.
2179 */
2180 if (IS_ERR(shm_mnt))
2181 return addr;
2182 sb = shm_mnt->mnt_sb;
2183 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002184 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002185 return addr;
2186 }
2187
2188 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2189 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2190 return addr;
2191 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2192 return addr;
2193
2194 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2195 if (inflated_len > TASK_SIZE)
2196 return addr;
2197 if (inflated_len < len)
2198 return addr;
2199
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002200 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002201 if (IS_ERR_VALUE(inflated_addr))
2202 return addr;
2203 if (inflated_addr & ~PAGE_MASK)
2204 return addr;
2205
2206 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2207 inflated_addr += offset - inflated_offset;
2208 if (inflated_offset > offset)
2209 inflated_addr += HPAGE_PMD_SIZE;
2210
2211 if (inflated_addr > TASK_SIZE - len)
2212 return addr;
2213 return inflated_addr;
2214}
2215
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002217static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218{
Al Viro496ad9a2013-01-23 17:07:38 -05002219 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002220 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221}
2222
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002223static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2224 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225{
Al Viro496ad9a2013-01-23 17:07:38 -05002226 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002227 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002229 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2230 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231}
2232#endif
2233
2234int shmem_lock(struct file *file, int lock, struct user_struct *user)
2235{
Al Viro496ad9a2013-01-23 17:07:38 -05002236 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 struct shmem_inode_info *info = SHMEM_I(inode);
2238 int retval = -ENOMEM;
2239
Hugh Dickinsea0dfeb2020-04-20 18:14:14 -07002240 /*
2241 * What serializes the accesses to info->flags?
2242 * ipc_lock_object() when called from shmctl_do_lock(),
2243 * no serialization needed when called from shm_destroy().
2244 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 if (lock && !(info->flags & VM_LOCKED)) {
2246 if (!user_shm_lock(inode->i_size, user))
2247 goto out_nomem;
2248 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002249 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 }
2251 if (!lock && (info->flags & VM_LOCKED) && user) {
2252 user_shm_unlock(inode->i_size, user);
2253 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002254 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 }
2256 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002257
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258out_nomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 return retval;
2260}
2261
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002262static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263{
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002264 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2265
2266 if (info->seals & F_SEAL_FUTURE_WRITE) {
2267 /*
2268 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2269 * "future write" seal active.
2270 */
2271 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2272 return -EPERM;
2273
2274 /*
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002275 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2276 * MAP_SHARED and read-only, take care to not allow mprotect to
2277 * revert protections on such mappings. Do this only for shared
2278 * mappings. For private mappings, don't need to mask
2279 * VM_MAYWRITE as we still want them to be COW-writable.
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002280 */
Nicolas Geoffray05d351102019-11-30 17:53:28 -08002281 if (vma->vm_flags & VM_SHARED)
2282 vma->vm_flags &= ~(VM_MAYWRITE);
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002283 }
2284
Catalin Marinas51b0bff2019-11-29 12:45:08 +00002285 /* arm64 - allow memory tagging on RAM-based files */
2286 vma->vm_flags |= VM_MTE_ALLOWED;
2287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 file_accessed(file);
2289 vma->vm_ops = &shmem_vm_ops;
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002290 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002291 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2292 (vma->vm_end & HPAGE_PMD_MASK)) {
2293 khugepaged_enter(vma, vma->vm_flags);
2294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 return 0;
2296}
2297
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002298static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002299 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300{
2301 struct inode *inode;
2302 struct shmem_inode_info *info;
2303 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -07002304 ino_t ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305
Chris Downe809d5f2020-08-06 23:20:20 -07002306 if (shmem_reserve_inode(sb, &ino))
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002307 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309 inode = new_inode(sb);
2310 if (inode) {
Chris Downe809d5f2020-08-06 23:20:20 -07002311 inode->i_ino = ino;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002312 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002314 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
Arnd Bergmann46c9a942018-08-17 15:45:09 -07002315 inode->i_generation = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 info = SHMEM_I(inode);
2317 memset(info, 0, (char *)inode - (char *)info);
2318 spin_lock_init(&info->lock);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07002319 atomic_set(&info->stop_eviction, 0);
David Herrmann40e041a2014-08-08 14:25:27 -07002320 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002321 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002322 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002324 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002325 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
2327 switch (mode & S_IFMT) {
2328 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002329 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 init_special_inode(inode, mode, dev);
2331 break;
2332 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002333 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 inode->i_op = &shmem_inode_operations;
2335 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002336 mpol_shared_policy_init(&info->policy,
2337 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 break;
2339 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002340 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 /* Some things misbehave if size == 0 on a directory */
2342 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2343 inode->i_op = &shmem_dir_inode_operations;
2344 inode->i_fop = &simple_dir_operations;
2345 break;
2346 case S_IFLNK:
2347 /*
2348 * Must not load anything in the rbtree,
2349 * mpol_free_shared_policy will not be called.
2350 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002351 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 break;
2353 }
Joel Fernandes (Google)b45d71f2018-09-20 12:22:39 -07002354
2355 lockdep_annotate_inode_mutex_key(inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002356 } else
2357 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 return inode;
2359}
2360
Johannes Weiner0cd61442014-04-03 14:47:46 -07002361bool shmem_mapping(struct address_space *mapping)
2362{
Hugh Dickinsf8005452017-02-22 15:41:42 -08002363 return mapping->a_ops == &shmem_aops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002364}
2365
Axel Rasmussend6721232021-03-18 17:01:51 +11002366#ifdef CONFIG_USERFAULTFD
2367int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
2368 struct vm_area_struct *dst_vma,
2369 unsigned long dst_addr, unsigned long src_addr,
2370 enum mcopy_atomic_mode mode, struct page **pagep)
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002371{
Axel Rasmussend6721232021-03-18 17:01:51 +11002372 bool is_continue = (mode == MCOPY_ATOMIC_CONTINUE);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002373 struct inode *inode = file_inode(dst_vma->vm_file);
2374 struct shmem_inode_info *info = SHMEM_I(inode);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002375 struct address_space *mapping = inode->i_mapping;
2376 gfp_t gfp = mapping_gfp_mask(mapping);
2377 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002378 spinlock_t *ptl;
2379 void *page_kaddr;
2380 struct page *page;
2381 pte_t _dst_pte, *dst_pte;
2382 int ret;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002383 pgoff_t offset, max_off;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002384
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002385 ret = -ENOMEM;
Mike Rapoport0f079692017-09-06 16:22:59 -07002386 if (!shmem_inode_acct_block(inode, 1))
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002387 goto out;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002388
Axel Rasmussend6721232021-03-18 17:01:51 +11002389 if (is_continue) {
2390 ret = -EFAULT;
2391 page = find_lock_page(mapping, pgoff);
2392 if (!page)
2393 goto out_unacct_blocks;
2394 } else if (!*pagep) {
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002395 page = shmem_alloc_page(gfp, info, pgoff);
2396 if (!page)
Mike Rapoport0f079692017-09-06 16:22:59 -07002397 goto out_unacct_blocks;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002398
Axel Rasmussend6721232021-03-18 17:01:51 +11002399 if (mode == MCOPY_ATOMIC_NORMAL) { /* mcopy_atomic */
Mike Rapoport8d103962017-09-06 16:23:02 -07002400 page_kaddr = kmap_atomic(page);
2401 ret = copy_from_user(page_kaddr,
2402 (const void __user *)src_addr,
2403 PAGE_SIZE);
2404 kunmap_atomic(page_kaddr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002405
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002406 /* fallback to copy_from_user outside mmap_lock */
Mike Rapoport8d103962017-09-06 16:23:02 -07002407 if (unlikely(ret)) {
2408 *pagep = page;
2409 shmem_inode_unacct_blocks(inode, 1);
2410 /* don't free the page */
Andrea Arcangeli9e368252018-11-30 14:09:25 -08002411 return -ENOENT;
Mike Rapoport8d103962017-09-06 16:23:02 -07002412 }
Axel Rasmussend6721232021-03-18 17:01:51 +11002413 } else { /* zeropage */
Mike Rapoport8d103962017-09-06 16:23:02 -07002414 clear_highpage(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002415 }
2416 } else {
2417 page = *pagep;
2418 *pagep = NULL;
2419 }
2420
Axel Rasmussend6721232021-03-18 17:01:51 +11002421 if (!is_continue) {
2422 VM_BUG_ON(PageSwapBacked(page));
2423 VM_BUG_ON(PageLocked(page));
2424 __SetPageLocked(page);
2425 __SetPageSwapBacked(page);
2426 __SetPageUptodate(page);
2427 }
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002428
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002429 ret = -EFAULT;
2430 offset = linear_page_index(dst_vma, dst_addr);
2431 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2432 if (unlikely(offset >= max_off))
2433 goto out_release;
2434
Axel Rasmussend6721232021-03-18 17:01:51 +11002435 /* If page wasn't already in the page cache, add it. */
2436 if (!is_continue) {
2437 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2438 gfp & GFP_RECLAIM_MASK, dst_mm);
2439 if (ret)
2440 goto out_release;
2441 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002442
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002443 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2444 if (dst_vma->vm_flags & VM_WRITE)
2445 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002446 else {
2447 /*
2448 * We don't set the pte dirty if the vma has no
2449 * VM_WRITE permission, so mark the page dirty or it
2450 * could be freed from under us. We could do it
2451 * unconditionally before unlock_page(), but doing it
2452 * only if VM_WRITE is not set is faster.
2453 */
2454 set_page_dirty(page);
2455 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002456
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002457 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002458
2459 ret = -EFAULT;
2460 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2461 if (unlikely(offset >= max_off))
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002462 goto out_release_unlock;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002463
2464 ret = -EEXIST;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002465 if (!pte_none(*dst_pte))
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002466 goto out_release_unlock;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002467
Axel Rasmussend6721232021-03-18 17:01:51 +11002468 if (!is_continue) {
2469 lru_cache_add(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002470
Axel Rasmussend6721232021-03-18 17:01:51 +11002471 spin_lock_irq(&info->lock);
2472 info->alloced++;
2473 inode->i_blocks += BLOCKS_PER_PAGE;
2474 shmem_recalc_inode(inode);
2475 spin_unlock_irq(&info->lock);
2476 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002477
2478 inc_mm_counter(dst_mm, mm_counter_file(page));
2479 page_add_file_rmap(page, false);
2480 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2481
2482 /* No need to invalidate - it was non-present before */
2483 update_mmu_cache(dst_vma, dst_addr, dst_pte);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002484 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002485 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002486 ret = 0;
2487out:
2488 return ret;
Johannes Weiner3fea5a42020-06-03 16:01:41 -07002489out_release_unlock:
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002490 pte_unmap_unlock(dst_pte, ptl);
Andrea Arcangelidcf7fe92018-11-30 14:09:43 -08002491 ClearPageDirty(page);
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002492 delete_from_page_cache(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002493out_release:
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002494 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002495 put_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002496out_unacct_blocks:
Mike Rapoport0f079692017-09-06 16:22:59 -07002497 shmem_inode_unacct_blocks(inode, 1);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002498 goto out;
2499}
Axel Rasmussend6721232021-03-18 17:01:51 +11002500#endif /* CONFIG_USERFAULTFD */
Mike Rapoport8d103962017-09-06 16:23:02 -07002501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002503static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002504static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002506#ifdef CONFIG_TMPFS_XATTR
2507static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2508#else
2509#define shmem_initxattrs NULL
2510#endif
2511
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002513shmem_write_begin(struct file *file, struct address_space *mapping,
2514 loff_t pos, unsigned len, unsigned flags,
2515 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516{
Nick Piggin800d15a2007-10-16 01:25:03 -07002517 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002518 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002519 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002520
2521 /* i_mutex is held by caller */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002522 if (unlikely(info->seals & (F_SEAL_GROW |
2523 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2524 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
David Herrmann40e041a2014-08-08 14:25:27 -07002525 return -EPERM;
2526 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2527 return -EPERM;
2528 }
2529
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002530 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002531}
2532
2533static int
2534shmem_write_end(struct file *file, struct address_space *mapping,
2535 loff_t pos, unsigned len, unsigned copied,
2536 struct page *page, void *fsdata)
2537{
2538 struct inode *inode = mapping->host;
2539
Hugh Dickinsd3602442008-02-04 22:28:44 -08002540 if (pos + copied > inode->i_size)
2541 i_size_write(inode, pos + copied);
2542
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002543 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002544 struct page *head = compound_head(page);
2545 if (PageTransCompound(page)) {
2546 int i;
2547
2548 for (i = 0; i < HPAGE_PMD_NR; i++) {
2549 if (head + i == page)
2550 continue;
2551 clear_highpage(head + i);
2552 flush_dcache_page(head + i);
2553 }
2554 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002555 if (copied < PAGE_SIZE) {
2556 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002557 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002558 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002559 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002560 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002561 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002562 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002563 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002564 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002565
Nick Piggin800d15a2007-10-16 01:25:03 -07002566 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567}
2568
Al Viro2ba5bbe2014-04-02 20:00:02 -04002569static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570{
Al Viro6e58e792014-02-03 17:07:03 -05002571 struct file *file = iocb->ki_filp;
2572 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002574 pgoff_t index;
2575 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002576 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002577 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002578 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002579 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002580
2581 /*
2582 * Might this read be for a stacking filesystem? Then when reading
2583 * holes of a sparse file, we actually need to allocate those pages,
2584 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2585 */
Al Viro777eda22014-12-17 04:46:46 -05002586 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002587 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002589 index = *ppos >> PAGE_SHIFT;
2590 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
2592 for (;;) {
2593 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002594 pgoff_t end_index;
2595 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 loff_t i_size = i_size_read(inode);
2597
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002598 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 if (index > end_index)
2600 break;
2601 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002602 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 if (nr <= offset)
2604 break;
2605 }
2606
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002607 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002608 if (error) {
2609 if (error == -EINVAL)
2610 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 break;
2612 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002613 if (page) {
2614 if (sgp == SGP_CACHE)
2615 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002616 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
2619 /*
2620 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002621 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002623 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002625 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002627 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 if (nr <= offset) {
2629 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002630 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 break;
2632 }
2633 }
2634 nr -= offset;
2635
2636 if (page) {
2637 /*
2638 * If users can be writing to this page using arbitrary
2639 * virtual addresses, take care about potential aliasing
2640 * before reading the page on the kernel side.
2641 */
2642 if (mapping_writably_mapped(mapping))
2643 flush_dcache_page(page);
2644 /*
2645 * Mark the page accessed if we read the beginning.
2646 */
2647 if (!offset)
2648 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002649 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002651 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653
2654 /*
2655 * Ok, we have the page, and it's up-to-date, so
2656 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002658 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002659 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002661 index += offset >> PAGE_SHIFT;
2662 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002664 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002665 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 break;
Al Viro6e58e792014-02-03 17:07:03 -05002667 if (ret < nr) {
2668 error = -EFAULT;
2669 break;
2670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 cond_resched();
2672 }
2673
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002674 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002675 file_accessed(file);
2676 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677}
2678
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002679/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -05002680 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002681 */
2682static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002683 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002684{
2685 struct page *page;
2686 struct pagevec pvec;
2687 pgoff_t indices[PAGEVEC_SIZE];
2688 bool done = false;
2689 int i;
2690
Mel Gorman86679822017-11-15 17:37:52 -08002691 pagevec_init(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002692 pvec.nr = 1; /* start small: we may be there already */
2693 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002694 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002695 pvec.nr, pvec.pages, indices);
2696 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002697 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002698 index = end;
2699 break;
2700 }
2701 for (i = 0; i < pvec.nr; i++, index++) {
2702 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002703 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002704 done = true;
2705 break;
2706 }
2707 index = indices[i];
2708 }
2709 page = pvec.pages[i];
Matthew Wilcox3159f942017-11-03 13:30:42 -04002710 if (page && !xa_is_value(page)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002711 if (!PageUptodate(page))
2712 page = NULL;
2713 }
2714 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002715 (page && whence == SEEK_DATA) ||
2716 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002717 done = true;
2718 break;
2719 }
2720 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002721 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002722 pagevec_release(&pvec);
2723 pvec.nr = PAGEVEC_SIZE;
2724 cond_resched();
2725 }
2726 return index;
2727}
2728
Andrew Morton965c8e52012-12-17 15:59:39 -08002729static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002730{
2731 struct address_space *mapping = file->f_mapping;
2732 struct inode *inode = mapping->host;
2733 pgoff_t start, end;
2734 loff_t new_offset;
2735
Andrew Morton965c8e52012-12-17 15:59:39 -08002736 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2737 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002738 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002739 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002740 /* We're holding i_mutex so we can access i_size directly */
2741
Yufen Yu1a413642018-11-16 15:08:39 -08002742 if (offset < 0 || offset >= inode->i_size)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002743 offset = -ENXIO;
2744 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002745 start = offset >> PAGE_SHIFT;
2746 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002747 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002748 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002749 if (new_offset > offset) {
2750 if (new_offset < inode->i_size)
2751 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002752 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002753 offset = -ENXIO;
2754 else
2755 offset = inode->i_size;
2756 }
2757 }
2758
Hugh Dickins387aae62013-08-04 11:30:25 -07002759 if (offset >= 0)
2760 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002761 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002762 return offset;
2763}
2764
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002765static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2766 loff_t len)
2767{
Al Viro496ad9a2013-01-23 17:07:38 -05002768 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002769 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002770 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002771 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002772 pgoff_t start, index, end;
2773 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002774
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002775 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2776 return -EOPNOTSUPP;
2777
Al Viro59551022016-01-22 15:40:57 -05002778 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002779
2780 if (mode & FALLOC_FL_PUNCH_HOLE) {
2781 struct address_space *mapping = file->f_mapping;
2782 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2783 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002784 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002785
David Herrmann40e041a2014-08-08 14:25:27 -07002786 /* protected by i_mutex */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002787 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
David Herrmann40e041a2014-08-08 14:25:27 -07002788 error = -EPERM;
2789 goto out;
2790 }
2791
Hugh Dickins8e205f72014-07-23 14:00:10 -07002792 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junaa71ecd2019-11-30 17:58:11 -08002793 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002794 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2795 spin_lock(&inode->i_lock);
2796 inode->i_private = &shmem_falloc;
2797 spin_unlock(&inode->i_lock);
2798
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002799 if ((u64)unmap_end > (u64)unmap_start)
2800 unmap_mapping_range(mapping, unmap_start,
2801 1 + unmap_end - unmap_start, 0);
2802 shmem_truncate_range(inode, offset, offset + len - 1);
2803 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002804
2805 spin_lock(&inode->i_lock);
2806 inode->i_private = NULL;
2807 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02002808 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002809 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002810 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002811 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002812 }
2813
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002814 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2815 error = inode_newsize_ok(inode, offset + len);
2816 if (error)
2817 goto out;
2818
David Herrmann40e041a2014-08-08 14:25:27 -07002819 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2820 error = -EPERM;
2821 goto out;
2822 }
2823
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002824 start = offset >> PAGE_SHIFT;
2825 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002826 /* Try to avoid a swapstorm if len is impossible to satisfy */
2827 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2828 error = -ENOSPC;
2829 goto out;
2830 }
2831
Hugh Dickins8e205f72014-07-23 14:00:10 -07002832 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002833 shmem_falloc.start = start;
2834 shmem_falloc.next = start;
2835 shmem_falloc.nr_falloced = 0;
2836 shmem_falloc.nr_unswapped = 0;
2837 spin_lock(&inode->i_lock);
2838 inode->i_private = &shmem_falloc;
2839 spin_unlock(&inode->i_lock);
2840
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002841 for (index = start; index < end; index++) {
2842 struct page *page;
2843
2844 /*
2845 * Good, the fallocate(2) manpage permits EINTR: we may have
2846 * been interrupted because we are using up too much memory.
2847 */
2848 if (signal_pending(current))
2849 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002850 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2851 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002852 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002853 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002854 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002855 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002856 if (index > start) {
2857 shmem_undo_range(inode,
2858 (loff_t)start << PAGE_SHIFT,
2859 ((loff_t)index << PAGE_SHIFT) - 1, true);
2860 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002861 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002862 }
2863
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002864 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002865 * Inform shmem_writepage() how far we have reached.
2866 * No need for lock or barrier: we have the page lock.
2867 */
2868 shmem_falloc.next++;
2869 if (!PageUptodate(page))
2870 shmem_falloc.nr_falloced++;
2871
2872 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002873 * If !PageUptodate, leave it that way so that freeable pages
2874 * can be recognized if we need to rollback on error later.
2875 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002876 * than free the pages we are allocating (and SGP_CACHE pages
2877 * might still be clean: we now need to mark those dirty too).
2878 */
2879 set_page_dirty(page);
2880 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002881 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002882 cond_resched();
2883 }
2884
2885 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2886 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002887 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002888undone:
2889 spin_lock(&inode->i_lock);
2890 inode->i_private = NULL;
2891 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002892out:
Al Viro59551022016-01-22 15:40:57 -05002893 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002894 return error;
2895}
2896
David Howells726c3342006-06-23 02:02:58 -07002897static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898{
David Howells726c3342006-06-23 02:02:58 -07002899 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
2901 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002902 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002904 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002906 buf->f_bavail =
2907 buf->f_bfree = sbinfo->max_blocks -
2908 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002909 }
2910 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 buf->f_files = sbinfo->max_inodes;
2912 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 }
2914 /* else leave those fields 0 like simple_statfs */
2915 return 0;
2916}
2917
2918/*
2919 * File creation. Allocate an inode, and we're done..
2920 */
2921static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002922shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002924 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 int error = -ENOSPC;
2926
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002927 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002929 error = simple_acl_create(dir, inode);
2930 if (error)
2931 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002932 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002933 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002934 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002935 if (error && error != -EOPNOTSUPP)
2936 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002937
Al Viro718deb62009-12-16 19:35:36 -05002938 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002940 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 d_instantiate(dentry, inode);
2942 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 }
2944 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002945out_iput:
2946 iput(inode);
2947 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948}
2949
Al Viro60545d02013-06-07 01:20:27 -04002950static int
2951shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2952{
2953 struct inode *inode;
2954 int error = -ENOSPC;
2955
2956 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2957 if (inode) {
2958 error = security_inode_init_security(inode, dir,
2959 NULL,
2960 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002961 if (error && error != -EOPNOTSUPP)
2962 goto out_iput;
2963 error = simple_acl_create(dir, inode);
2964 if (error)
2965 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04002966 d_tmpfile(dentry, inode);
2967 }
2968 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002969out_iput:
2970 iput(inode);
2971 return error;
Al Viro60545d02013-06-07 01:20:27 -04002972}
2973
Al Viro18bb1db2011-07-26 01:41:39 -04002974static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975{
2976 int error;
2977
2978 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2979 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002980 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 return 0;
2982}
2983
Al Viro4acdaf22011-07-26 01:42:34 -04002984static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002985 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986{
2987 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2988}
2989
2990/*
2991 * Link a file..
2992 */
2993static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2994{
David Howells75c3cfa2015-03-17 22:26:12 +00002995 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong29b00e602019-02-22 22:35:32 -08002996 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997
2998 /*
2999 * No ordinary (disk based) filesystem counts links as inodes;
3000 * but each new link needs a new dentry, pinning lowmem, and
3001 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wong1062af92019-02-21 08:48:09 -08003002 * But if an O_TMPFILE file is linked into the tmpfs, the
3003 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 */
Darrick J. Wong1062af92019-02-21 08:48:09 -08003005 if (inode->i_nlink) {
Chris Downe809d5f2020-08-06 23:20:20 -07003006 ret = shmem_reserve_inode(inode->i_sb, NULL);
Darrick J. Wong1062af92019-02-21 08:48:09 -08003007 if (ret)
3008 goto out;
3009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010
3011 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003012 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07003013 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04003014 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 dget(dentry); /* Extra pinning count for the created dentry */
3016 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003017out:
3018 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019}
3020
3021static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3022{
David Howells75c3cfa2015-03-17 22:26:12 +00003023 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003025 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3026 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027
3028 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003029 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003030 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 dput(dentry); /* Undo the count from "create" - this does all the work */
3032 return 0;
3033}
3034
3035static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3036{
3037 if (!simple_empty(dentry))
3038 return -ENOTEMPTY;
3039
David Howells75c3cfa2015-03-17 22:26:12 +00003040 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003041 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 return shmem_unlink(dir, dentry);
3043}
3044
Miklos Szeredi37456772014-07-23 15:15:34 +02003045static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3046{
David Howellse36cb0b2015-01-29 12:02:35 +00003047 bool old_is_dir = d_is_dir(old_dentry);
3048 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003049
3050 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3051 if (old_is_dir) {
3052 drop_nlink(old_dir);
3053 inc_nlink(new_dir);
3054 } else {
3055 drop_nlink(new_dir);
3056 inc_nlink(old_dir);
3057 }
3058 }
3059 old_dir->i_ctime = old_dir->i_mtime =
3060 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003061 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003062 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02003063
3064 return 0;
3065}
3066
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003067static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3068{
3069 struct dentry *whiteout;
3070 int error;
3071
3072 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3073 if (!whiteout)
3074 return -ENOMEM;
3075
3076 error = shmem_mknod(old_dir, whiteout,
3077 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3078 dput(whiteout);
3079 if (error)
3080 return error;
3081
3082 /*
3083 * Cheat and hash the whiteout while the old dentry is still in
3084 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3085 *
3086 * d_lookup() will consistently find one of them at this point,
3087 * not sure which one, but that isn't even important.
3088 */
3089 d_rehash(whiteout);
3090 return 0;
3091}
3092
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093/*
3094 * The VFS layer already does all the dentry stuff for rename,
3095 * we just have to decrement the usage count for the target if
3096 * it exists so that the VFS layer correctly free's it when it
3097 * gets overwritten.
3098 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003099static 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 -07003100{
David Howells75c3cfa2015-03-17 22:26:12 +00003101 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 int they_are_dirs = S_ISDIR(inode->i_mode);
3103
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003104 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003105 return -EINVAL;
3106
Miklos Szeredi37456772014-07-23 15:15:34 +02003107 if (flags & RENAME_EXCHANGE)
3108 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3109
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110 if (!simple_empty(new_dentry))
3111 return -ENOTEMPTY;
3112
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003113 if (flags & RENAME_WHITEOUT) {
3114 int error;
3115
3116 error = shmem_whiteout(old_dir, old_dentry);
3117 if (error)
3118 return error;
3119 }
3120
David Howells75c3cfa2015-03-17 22:26:12 +00003121 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003123 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003124 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003125 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003128 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003129 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 }
3131
3132 old_dir->i_size -= BOGO_DIRENT_SIZE;
3133 new_dir->i_size += BOGO_DIRENT_SIZE;
3134 old_dir->i_ctime = old_dir->i_mtime =
3135 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003136 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 return 0;
3138}
3139
3140static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3141{
3142 int error;
3143 int len;
3144 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003145 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146
3147 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003148 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 return -ENAMETOOLONG;
3150
Joe Perches0825a6f2018-06-14 15:27:58 -07003151 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3152 VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 if (!inode)
3154 return -ENOSPC;
3155
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003156 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003157 shmem_initxattrs, NULL);
Mateusz Nosek343c3d72020-04-06 20:07:54 -07003158 if (error && error != -EOPNOTSUPP) {
3159 iput(inode);
3160 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003161 }
3162
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003164 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003165 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3166 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003167 iput(inode);
3168 return -ENOMEM;
3169 }
3170 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003172 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003173 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174 if (error) {
3175 iput(inode);
3176 return error;
3177 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003178 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003180 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003181 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003183 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003184 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003187 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 d_instantiate(dentry, inode);
3189 dget(dentry);
3190 return 0;
3191}
3192
Al Virofceef392015-12-29 15:58:39 -05003193static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194{
Al Virofceef392015-12-29 15:58:39 -05003195 mark_page_accessed(arg);
3196 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197}
3198
Al Viro6b255392015-11-17 10:20:54 -05003199static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003200 struct inode *inode,
3201 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003204 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003205 if (!dentry) {
3206 page = find_get_page(inode->i_mapping, 0);
3207 if (!page)
3208 return ERR_PTR(-ECHILD);
3209 if (!PageUptodate(page)) {
3210 put_page(page);
3211 return ERR_PTR(-ECHILD);
3212 }
3213 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003214 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003215 if (error)
3216 return ERR_PTR(error);
3217 unlock_page(page);
3218 }
Al Virofceef392015-12-29 15:58:39 -05003219 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003220 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221}
3222
Eric Parisb09e0fa2011-05-24 17:12:39 -07003223#ifdef CONFIG_TMPFS_XATTR
3224/*
3225 * Superblocks without xattr inode operations may get some security.* xattr
3226 * support from the LSM "for free". As soon as we have any other xattrs
3227 * like ACLs, we also need to implement the security.* handlers at
3228 * filesystem level, though.
3229 */
3230
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003231/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003232 * Callback for security_inode_init_security() for acquiring xattrs.
3233 */
3234static int shmem_initxattrs(struct inode *inode,
3235 const struct xattr *xattr_array,
3236 void *fs_info)
3237{
3238 struct shmem_inode_info *info = SHMEM_I(inode);
3239 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003240 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003241 size_t len;
3242
3243 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003244 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003245 if (!new_xattr)
3246 return -ENOMEM;
3247
3248 len = strlen(xattr->name) + 1;
3249 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3250 GFP_KERNEL);
3251 if (!new_xattr->name) {
Chengguang Xu3bef7352020-07-23 21:15:14 -07003252 kvfree(new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003253 return -ENOMEM;
3254 }
3255
3256 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3257 XATTR_SECURITY_PREFIX_LEN);
3258 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3259 xattr->name, len);
3260
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003261 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003262 }
3263
3264 return 0;
3265}
3266
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003267static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003268 struct dentry *unused, struct inode *inode,
Mark Salyzyn3484eba2019-11-04 08:57:10 -08003269 const char *name, void *buffer, size_t size,
3270 int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003271{
Al Virob2968212016-04-10 20:48:24 -04003272 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003273
3274 name = xattr_full_name(handler, name);
3275 return simple_xattr_get(&info->xattrs, name, buffer, size);
3276}
3277
3278static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003279 struct dentry *unused, struct inode *inode,
3280 const char *name, const void *value,
3281 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003282{
Al Viro59301222016-05-27 10:19:30 -04003283 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003284
3285 name = xattr_full_name(handler, name);
Daniel Xua46a2292020-03-12 13:03:15 -07003286 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003287}
3288
3289static const struct xattr_handler shmem_security_xattr_handler = {
3290 .prefix = XATTR_SECURITY_PREFIX,
3291 .get = shmem_xattr_handler_get,
3292 .set = shmem_xattr_handler_set,
3293};
3294
3295static const struct xattr_handler shmem_trusted_xattr_handler = {
3296 .prefix = XATTR_TRUSTED_PREFIX,
3297 .get = shmem_xattr_handler_get,
3298 .set = shmem_xattr_handler_set,
3299};
3300
Eric Parisb09e0fa2011-05-24 17:12:39 -07003301static const struct xattr_handler *shmem_xattr_handlers[] = {
3302#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003303 &posix_acl_access_xattr_handler,
3304 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003305#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003306 &shmem_security_xattr_handler,
3307 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003308 NULL
3309};
3310
Eric Parisb09e0fa2011-05-24 17:12:39 -07003311static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3312{
David Howells75c3cfa2015-03-17 22:26:12 +00003313 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003314 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003315}
3316#endif /* CONFIG_TMPFS_XATTR */
3317
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003318static const struct inode_operations shmem_short_symlink_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003319 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003320#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003321 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003322#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323};
3324
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003325static const struct inode_operations shmem_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003326 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003327#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003328 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003329#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003330};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003331
David M. Grimes91828a42006-10-17 00:09:45 -07003332static struct dentry *shmem_get_parent(struct dentry *child)
3333{
3334 return ERR_PTR(-ESTALE);
3335}
3336
3337static int shmem_match(struct inode *ino, void *vfh)
3338{
3339 __u32 *fh = vfh;
3340 __u64 inum = fh[2];
3341 inum = (inum << 32) | fh[1];
3342 return ino->i_ino == inum && fh[0] == ino->i_generation;
3343}
3344
Amir Goldstein12ba7802018-06-07 17:07:15 -07003345/* Find any alias of inode, but prefer a hashed alias */
3346static struct dentry *shmem_find_alias(struct inode *inode)
3347{
3348 struct dentry *alias = d_find_alias(inode);
3349
3350 return alias ?: d_find_any_alias(inode);
3351}
3352
3353
Christoph Hellwig480b1162007-10-21 16:42:13 -07003354static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3355 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003356{
David M. Grimes91828a42006-10-17 00:09:45 -07003357 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003358 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003359 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003360
Christoph Hellwig480b1162007-10-21 16:42:13 -07003361 if (fh_len < 3)
3362 return NULL;
3363
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003364 inum = fid->raw[2];
3365 inum = (inum << 32) | fid->raw[1];
3366
Christoph Hellwig480b1162007-10-21 16:42:13 -07003367 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3368 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003369 if (inode) {
Amir Goldstein12ba7802018-06-07 17:07:15 -07003370 dentry = shmem_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003371 iput(inode);
3372 }
3373
Christoph Hellwig480b1162007-10-21 16:42:13 -07003374 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003375}
3376
Al Virob0b03822012-04-02 14:34:06 -04003377static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3378 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003379{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303380 if (*len < 3) {
3381 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003382 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303383 }
David M. Grimes91828a42006-10-17 00:09:45 -07003384
Al Viro1d3382cb2010-10-23 15:19:20 -04003385 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003386 /* Unfortunately insert_inode_hash is not idempotent,
3387 * so as we hash inodes here rather than at creation
3388 * time, we need a lock to ensure we only try
3389 * to do it once
3390 */
3391 static DEFINE_SPINLOCK(lock);
3392 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003393 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003394 __insert_inode_hash(inode,
3395 inode->i_ino + inode->i_generation);
3396 spin_unlock(&lock);
3397 }
3398
3399 fh[0] = inode->i_generation;
3400 fh[1] = inode->i_ino;
3401 fh[2] = ((__u64)inode->i_ino) >> 32;
3402
3403 *len = 3;
3404 return 1;
3405}
3406
Christoph Hellwig39655162007-10-21 16:42:17 -07003407static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003408 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003409 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003410 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003411};
3412
Al Viro626c3922019-09-08 20:28:06 -04003413enum shmem_param {
3414 Opt_gid,
3415 Opt_huge,
3416 Opt_mode,
3417 Opt_mpol,
3418 Opt_nr_blocks,
3419 Opt_nr_inodes,
3420 Opt_size,
3421 Opt_uid,
Chris Downea3271f2020-08-06 23:20:25 -07003422 Opt_inode32,
3423 Opt_inode64,
Al Viro626c3922019-09-08 20:28:06 -04003424};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425
Al Viro5eede622019-12-16 13:33:32 -05003426static const struct constant_table shmem_param_enums_huge[] = {
Al Viro2710c957a2019-09-06 22:12:08 -04003427 {"never", SHMEM_HUGE_NEVER },
3428 {"always", SHMEM_HUGE_ALWAYS },
3429 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3430 {"advise", SHMEM_HUGE_ADVISE },
Al Viro2710c957a2019-09-06 22:12:08 -04003431 {}
3432};
3433
Al Virod7167b12019-09-07 07:23:15 -04003434const struct fs_parameter_spec shmem_fs_parameters[] = {
Al Viro626c3922019-09-08 20:28:06 -04003435 fsparam_u32 ("gid", Opt_gid),
Al Viro2710c957a2019-09-06 22:12:08 -04003436 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
Al Viro626c3922019-09-08 20:28:06 -04003437 fsparam_u32oct("mode", Opt_mode),
3438 fsparam_string("mpol", Opt_mpol),
3439 fsparam_string("nr_blocks", Opt_nr_blocks),
3440 fsparam_string("nr_inodes", Opt_nr_inodes),
3441 fsparam_string("size", Opt_size),
3442 fsparam_u32 ("uid", Opt_uid),
Chris Downea3271f2020-08-06 23:20:25 -07003443 fsparam_flag ("inode32", Opt_inode32),
3444 fsparam_flag ("inode64", Opt_inode64),
Al Viro626c3922019-09-08 20:28:06 -04003445 {}
3446};
3447
David Howellsf3235622019-03-25 16:38:31 +00003448static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
Al Viro626c3922019-09-08 20:28:06 -04003449{
David Howellsf3235622019-03-25 16:38:31 +00003450 struct shmem_options *ctx = fc->fs_private;
Al Viro626c3922019-09-08 20:28:06 -04003451 struct fs_parse_result result;
3452 unsigned long long size;
3453 char *rest;
3454 int opt;
3455
Al Virod7167b12019-09-07 07:23:15 -04003456 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
David Howellsf3235622019-03-25 16:38:31 +00003457 if (opt < 0)
Al Viro626c3922019-09-08 20:28:06 -04003458 return opt;
Al Viro626c3922019-09-08 20:28:06 -04003459
3460 switch (opt) {
3461 case Opt_size:
3462 size = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003463 if (*rest == '%') {
3464 size <<= PAGE_SHIFT;
3465 size *= totalram_pages();
3466 do_div(size, 100);
3467 rest++;
3468 }
3469 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003470 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003471 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3472 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003473 break;
3474 case Opt_nr_blocks:
3475 ctx->blocks = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003476 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003477 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003478 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04003479 break;
3480 case Opt_nr_inodes:
3481 ctx->inodes = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04003482 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04003483 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04003484 ctx->seen |= SHMEM_SEEN_INODES;
Al Viro626c3922019-09-08 20:28:06 -04003485 break;
3486 case Opt_mode:
3487 ctx->mode = result.uint_32 & 07777;
3488 break;
3489 case Opt_uid:
3490 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003491 if (!uid_valid(ctx->uid))
Al Viro626c3922019-09-08 20:28:06 -04003492 goto bad_value;
3493 break;
3494 case Opt_gid:
3495 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
Al Viroe04dc422019-09-08 19:20:12 -04003496 if (!gid_valid(ctx->gid))
Al Viro626c3922019-09-08 20:28:06 -04003497 goto bad_value;
3498 break;
3499 case Opt_huge:
3500 ctx->huge = result.uint_32;
3501 if (ctx->huge != SHMEM_HUGE_NEVER &&
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003502 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Al Viro626c3922019-09-08 20:28:06 -04003503 has_transparent_hugepage()))
3504 goto unsupported_parameter;
Al Viroe04dc422019-09-08 19:20:12 -04003505 ctx->seen |= SHMEM_SEEN_HUGE;
Al Viro626c3922019-09-08 20:28:06 -04003506 break;
3507 case Opt_mpol:
3508 if (IS_ENABLED(CONFIG_NUMA)) {
3509 mpol_put(ctx->mpol);
3510 ctx->mpol = NULL;
3511 if (mpol_parse_str(param->string, &ctx->mpol))
3512 goto bad_value;
3513 break;
3514 }
3515 goto unsupported_parameter;
Chris Downea3271f2020-08-06 23:20:25 -07003516 case Opt_inode32:
3517 ctx->full_inums = false;
3518 ctx->seen |= SHMEM_SEEN_INUMS;
3519 break;
3520 case Opt_inode64:
3521 if (sizeof(ino_t) < 8) {
3522 return invalfc(fc,
3523 "Cannot use inode64 with <64bit inums in kernel\n");
3524 }
3525 ctx->full_inums = true;
3526 ctx->seen |= SHMEM_SEEN_INUMS;
3527 break;
Al Viroe04dc422019-09-08 19:20:12 -04003528 }
3529 return 0;
3530
Al Viro626c3922019-09-08 20:28:06 -04003531unsupported_parameter:
Al Virof35aa2b2019-12-21 21:35:55 -05003532 return invalfc(fc, "Unsupported parameter '%s'", param->key);
Al Viro626c3922019-09-08 20:28:06 -04003533bad_value:
Al Virof35aa2b2019-12-21 21:35:55 -05003534 return invalfc(fc, "Bad value for '%s'", param->key);
Al Viroe04dc422019-09-08 19:20:12 -04003535}
3536
David Howellsf3235622019-03-25 16:38:31 +00003537static int shmem_parse_options(struct fs_context *fc, void *data)
Al Viroe04dc422019-09-08 19:20:12 -04003538{
David Howellsf3235622019-03-25 16:38:31 +00003539 char *options = data;
3540
Al Viro33f37c62019-10-09 22:48:01 -04003541 if (options) {
3542 int err = security_sb_eat_lsm_opts(options, &fc->security);
3543 if (err)
3544 return err;
3545 }
3546
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003547 while (options != NULL) {
Al Viro626c3922019-09-08 20:28:06 -04003548 char *this_char = options;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003549 for (;;) {
3550 /*
3551 * NUL-terminate this option: unfortunately,
3552 * mount options form a comma-separated list,
3553 * but mpol's nodelist may also contain commas.
3554 */
3555 options = strchr(options, ',');
3556 if (options == NULL)
3557 break;
3558 options++;
3559 if (!isdigit(*options)) {
3560 options[-1] = '\0';
3561 break;
3562 }
3563 }
Al Viro626c3922019-09-08 20:28:06 -04003564 if (*this_char) {
3565 char *value = strchr(this_char,'=');
David Howellsf3235622019-03-25 16:38:31 +00003566 size_t len = 0;
Al Viro626c3922019-09-08 20:28:06 -04003567 int err;
3568
3569 if (value) {
3570 *value++ = '\0';
David Howellsf3235622019-03-25 16:38:31 +00003571 len = strlen(value);
Al Viro626c3922019-09-08 20:28:06 -04003572 }
David Howellsf3235622019-03-25 16:38:31 +00003573 err = vfs_parse_fs_string(fc, this_char, value, len);
3574 if (err < 0)
3575 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577 }
3578 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579}
3580
David Howellsf3235622019-03-25 16:38:31 +00003581/*
3582 * Reconfigure a shmem filesystem.
3583 *
3584 * Note that we disallow change from limited->unlimited blocks/inodes while any
3585 * are in use; but we must separately disallow unlimited->limited, because in
3586 * that case we have no record of how much is already in use.
3587 */
3588static int shmem_reconfigure(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589{
David Howellsf3235622019-03-25 16:38:31 +00003590 struct shmem_options *ctx = fc->fs_private;
3591 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003592 unsigned long inodes;
David Howellsf3235622019-03-25 16:38:31 +00003593 const char *err;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003594
3595 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003596 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
David Howellsf3235622019-03-25 16:38:31 +00003597 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3598 if (!sbinfo->max_blocks) {
3599 err = "Cannot retroactively limit size";
Al Viro0b5071d2019-09-08 18:49:18 -04003600 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003601 }
Al Viro0b5071d2019-09-08 18:49:18 -04003602 if (percpu_counter_compare(&sbinfo->used_blocks,
David Howellsf3235622019-03-25 16:38:31 +00003603 ctx->blocks) > 0) {
3604 err = "Too small a size for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003605 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003606 }
Al Viro0b5071d2019-09-08 18:49:18 -04003607 }
David Howellsf3235622019-03-25 16:38:31 +00003608 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3609 if (!sbinfo->max_inodes) {
3610 err = "Cannot retroactively limit inodes";
Al Viro0b5071d2019-09-08 18:49:18 -04003611 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003612 }
3613 if (ctx->inodes < inodes) {
3614 err = "Too few inodes for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04003615 goto out;
David Howellsf3235622019-03-25 16:38:31 +00003616 }
Al Viro0b5071d2019-09-08 18:49:18 -04003617 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003618
Chris Downea3271f2020-08-06 23:20:25 -07003619 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3620 sbinfo->next_ino > UINT_MAX) {
3621 err = "Current inum too high to switch to 32-bit inums";
3622 goto out;
3623 }
3624
David Howellsf3235622019-03-25 16:38:31 +00003625 if (ctx->seen & SHMEM_SEEN_HUGE)
3626 sbinfo->huge = ctx->huge;
Chris Downea3271f2020-08-06 23:20:25 -07003627 if (ctx->seen & SHMEM_SEEN_INUMS)
3628 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00003629 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3630 sbinfo->max_blocks = ctx->blocks;
3631 if (ctx->seen & SHMEM_SEEN_INODES) {
3632 sbinfo->max_inodes = ctx->inodes;
3633 sbinfo->free_inodes = ctx->inodes - inodes;
Al Viro0b5071d2019-09-08 18:49:18 -04003634 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003635
Greg Thelen5f001102013-02-22 16:36:01 -08003636 /*
3637 * Preserve previous mempolicy unless mpol remount option was specified.
3638 */
David Howellsf3235622019-03-25 16:38:31 +00003639 if (ctx->mpol) {
Greg Thelen5f001102013-02-22 16:36:01 -08003640 mpol_put(sbinfo->mpol);
David Howellsf3235622019-03-25 16:38:31 +00003641 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3642 ctx->mpol = NULL;
Greg Thelen5f001102013-02-22 16:36:01 -08003643 }
David Howellsf3235622019-03-25 16:38:31 +00003644 spin_unlock(&sbinfo->stat_lock);
3645 return 0;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003646out:
3647 spin_unlock(&sbinfo->stat_lock);
Al Virof35aa2b2019-12-21 21:35:55 -05003648 return invalfc(fc, "%s", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003650
Al Viro34c80b12011-12-08 21:32:45 -05003651static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003652{
Al Viro34c80b12011-12-08 21:32:45 -05003653 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003654
3655 if (sbinfo->max_blocks != shmem_default_max_blocks())
3656 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003657 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003658 if (sbinfo->max_inodes != shmem_default_max_inodes())
3659 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
Joe Perches0825a6f2018-06-14 15:27:58 -07003660 if (sbinfo->mode != (0777 | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003661 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003662 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3663 seq_printf(seq, ",uid=%u",
3664 from_kuid_munged(&init_user_ns, sbinfo->uid));
3665 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3666 seq_printf(seq, ",gid=%u",
3667 from_kgid_munged(&init_user_ns, sbinfo->gid));
Chris Downea3271f2020-08-06 23:20:25 -07003668
3669 /*
3670 * Showing inode{64,32} might be useful even if it's the system default,
3671 * since then people don't have to resort to checking both here and
3672 * /proc/config.gz to confirm 64-bit inums were successfully applied
3673 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3674 *
3675 * We hide it when inode64 isn't the default and we are using 32-bit
3676 * inodes, since that probably just means the feature isn't even under
3677 * consideration.
3678 *
3679 * As such:
3680 *
3681 * +-----------------+-----------------+
3682 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3683 * +------------------+-----------------+-----------------+
3684 * | full_inums=true | show | show |
3685 * | full_inums=false | show | hide |
3686 * +------------------+-----------------+-----------------+
3687 *
3688 */
3689 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3690 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003691#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003692 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3693 if (sbinfo->huge)
3694 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3695#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003696 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003697 return 0;
3698}
David Herrmann9183df22014-08-08 14:25:29 -07003699
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003700#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701
3702static void shmem_put_super(struct super_block *sb)
3703{
Hugh Dickins602586a2010-08-17 15:23:56 -07003704 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3705
Chris Downe809d5f2020-08-06 23:20:20 -07003706 free_percpu(sbinfo->ino_batch);
Hugh Dickins602586a2010-08-17 15:23:56 -07003707 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003708 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003709 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710 sb->s_fs_info = NULL;
3711}
3712
David Howellsf3235622019-03-25 16:38:31 +00003713static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714{
David Howellsf3235622019-03-25 16:38:31 +00003715 struct shmem_options *ctx = fc->fs_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003717 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003718 int err = -ENOMEM;
3719
3720 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003721 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003722 L1_CACHE_BYTES), GFP_KERNEL);
3723 if (!sbinfo)
3724 return -ENOMEM;
3725
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003726 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003728#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 /*
3730 * Per default we only allow half of the physical ram per
3731 * tmpfs instance, limiting inodes to one per page of lowmem;
3732 * but the internal instance is left unlimited.
3733 */
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003734 if (!(sb->s_flags & SB_KERNMOUNT)) {
David Howellsf3235622019-03-25 16:38:31 +00003735 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3736 ctx->blocks = shmem_default_max_blocks();
3737 if (!(ctx->seen & SHMEM_SEEN_INODES))
3738 ctx->inodes = shmem_default_max_inodes();
Chris Downea3271f2020-08-06 23:20:25 -07003739 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3740 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
Al Viroca4e0512013-08-31 12:57:10 -04003741 } else {
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003742 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 }
David M. Grimes91828a42006-10-17 00:09:45 -07003744 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003745 sb->s_flags |= SB_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746#else
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003747 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748#endif
David Howellsf3235622019-03-25 16:38:31 +00003749 sbinfo->max_blocks = ctx->blocks;
3750 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
Chris Downe809d5f2020-08-06 23:20:20 -07003751 if (sb->s_flags & SB_KERNMOUNT) {
3752 sbinfo->ino_batch = alloc_percpu(ino_t);
3753 if (!sbinfo->ino_batch)
3754 goto failed;
3755 }
David Howellsf3235622019-03-25 16:38:31 +00003756 sbinfo->uid = ctx->uid;
3757 sbinfo->gid = ctx->gid;
Chris Downea3271f2020-08-06 23:20:25 -07003758 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00003759 sbinfo->mode = ctx->mode;
3760 sbinfo->huge = ctx->huge;
3761 sbinfo->mpol = ctx->mpol;
3762 ctx->mpol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003764 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003765 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003766 goto failed;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003767 spin_lock_init(&sbinfo->shrinklist_lock);
3768 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003769
Hugh Dickins285b2c42011-08-03 16:21:20 -07003770 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003771 sb->s_blocksize = PAGE_SIZE;
3772 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 sb->s_magic = TMPFS_MAGIC;
3774 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003775 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003776#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003777 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003778#endif
3779#ifdef CONFIG_TMPFS_POSIX_ACL
Linus Torvalds1751e8a2017-11-27 13:05:09 -08003780 sb->s_flags |= SB_POSIXACL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003781#endif
Amir Goldstein2b4db792017-05-18 15:29:33 +03003782 uuid_gen(&sb->s_uuid);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003783
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003784 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 if (!inode)
3786 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003787 inode->i_uid = sbinfo->uid;
3788 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003789 sb->s_root = d_make_root(inode);
3790 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003791 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 return 0;
3793
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794failed:
3795 shmem_put_super(sb);
3796 return err;
3797}
3798
David Howellsf3235622019-03-25 16:38:31 +00003799static int shmem_get_tree(struct fs_context *fc)
3800{
3801 return get_tree_nodev(fc, shmem_fill_super);
3802}
3803
3804static void shmem_free_fc(struct fs_context *fc)
3805{
3806 struct shmem_options *ctx = fc->fs_private;
3807
3808 if (ctx) {
3809 mpol_put(ctx->mpol);
3810 kfree(ctx);
3811 }
3812}
3813
3814static const struct fs_context_operations shmem_fs_context_ops = {
3815 .free = shmem_free_fc,
3816 .get_tree = shmem_get_tree,
3817#ifdef CONFIG_TMPFS
3818 .parse_monolithic = shmem_parse_options,
3819 .parse_param = shmem_parse_one,
3820 .reconfigure = shmem_reconfigure,
3821#endif
3822};
3823
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003824static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825
3826static struct inode *shmem_alloc_inode(struct super_block *sb)
3827{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003828 struct shmem_inode_info *info;
3829 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3830 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003832 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833}
3834
Al Viro74b1da52019-04-15 23:19:05 -04003835static void shmem_free_in_core_inode(struct inode *inode)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003836{
Al Viro84e710d2016-04-15 00:58:55 -04003837 if (S_ISLNK(inode->i_mode))
3838 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003839 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3840}
3841
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842static void shmem_destroy_inode(struct inode *inode)
3843{
Al Viro09208d12011-07-26 03:15:03 -04003844 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846}
3847
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003848static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003850 struct shmem_inode_info *info = foo;
3851 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852}
3853
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003854static void shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855{
3856 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3857 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003858 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859}
3860
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003861static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003863 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864}
3865
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003866static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003868 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003870 .write_begin = shmem_write_begin,
3871 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003873#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003874 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003875#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003876 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877};
3878
Helge Deller15ad7cd2006-12-06 20:40:36 -08003879static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003881 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003883 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003884 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003885 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003886 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003887 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003888 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003889 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890#endif
3891};
3892
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003893static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003894 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003895 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003896#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003897 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003898 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003899#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900};
3901
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003902static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903#ifdef CONFIG_TMPFS
3904 .create = shmem_create,
3905 .lookup = simple_lookup,
3906 .link = shmem_link,
3907 .unlink = shmem_unlink,
3908 .symlink = shmem_symlink,
3909 .mkdir = shmem_mkdir,
3910 .rmdir = shmem_rmdir,
3911 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003912 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003913 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003915#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003916 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003917#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003918#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003919 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003920 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003921#endif
3922};
3923
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003924static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003925#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003926 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003927#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003928#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003929 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003930 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003931#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932};
3933
Hugh Dickins759b9772007-03-05 00:30:28 -08003934static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935 .alloc_inode = shmem_alloc_inode,
Al Viro74b1da52019-04-15 23:19:05 -04003936 .free_inode = shmem_free_in_core_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 .destroy_inode = shmem_destroy_inode,
3938#ifdef CONFIG_TMPFS
3939 .statfs = shmem_statfs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003940 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941#endif
Al Viro1f895f72010-06-05 19:10:41 -04003942 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943 .drop_inode = generic_delete_inode,
3944 .put_super = shmem_put_super,
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07003945#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003946 .nr_cached_objects = shmem_unused_huge_count,
3947 .free_cached_objects = shmem_unused_huge_scan,
3948#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949};
3950
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003951static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003952 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003953 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954#ifdef CONFIG_NUMA
3955 .set_policy = shmem_set_policy,
3956 .get_policy = shmem_get_policy,
3957#endif
Vinayak Menon35eacb52021-03-18 15:20:17 +05303958#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
3959 .allow_speculation = filemap_allow_speculation,
3960#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961};
3962
David Howellsf3235622019-03-25 16:38:31 +00003963int shmem_init_fs_context(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964{
David Howellsf3235622019-03-25 16:38:31 +00003965 struct shmem_options *ctx;
3966
3967 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3968 if (!ctx)
3969 return -ENOMEM;
3970
3971 ctx->mode = 0777 | S_ISVTX;
3972 ctx->uid = current_fsuid();
3973 ctx->gid = current_fsgid();
3974
3975 fc->fs_private = ctx;
3976 fc->ops = &shmem_fs_context_ops;
3977 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978}
3979
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003980static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981 .owner = THIS_MODULE,
3982 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00003983 .init_fs_context = shmem_init_fs_context,
3984#ifdef CONFIG_TMPFS
Al Virod7167b12019-09-07 07:23:15 -04003985 .parameters = shmem_fs_parameters,
David Howellsf3235622019-03-25 16:38:31 +00003986#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 .kill_sb = kill_litter_super,
Matthew Wilcox (Oracle)01c70262020-10-15 20:06:00 -07003988 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003991int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992{
3993 int error;
3994
weiping zhang9a8ec03e2017-11-15 17:38:18 -08003995 shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003997 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003999 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 goto out2;
4001 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07004002
Al Viroca4e0512013-08-31 12:57:10 -04004003 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004 if (IS_ERR(shm_mnt)) {
4005 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07004006 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007 goto out1;
4008 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004009
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004010#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004011 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004012 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4013 else
4014 shmem_huge = 0; /* just in case it was patched */
4015#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 return 0;
4017
4018out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004019 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004021 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 shm_mnt = ERR_PTR(error);
4023 return error;
4024}
Matt Mackall853ac432009-01-06 14:40:20 -08004025
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004026#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004027static ssize_t shmem_enabled_show(struct kobject *kobj,
4028 struct kobj_attribute *attr, char *buf)
4029{
Colin Ian King26083eb62019-11-30 17:58:04 -08004030 static const int values[] = {
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004031 SHMEM_HUGE_ALWAYS,
4032 SHMEM_HUGE_WITHIN_SIZE,
4033 SHMEM_HUGE_ADVISE,
4034 SHMEM_HUGE_NEVER,
4035 SHMEM_HUGE_DENY,
4036 SHMEM_HUGE_FORCE,
4037 };
4038 int i, count;
4039
4040 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4041 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4042
4043 count += sprintf(buf + count, fmt,
4044 shmem_format_huge(values[i]));
4045 }
4046 buf[count - 1] = '\n';
4047 return count;
4048}
4049
4050static ssize_t shmem_enabled_store(struct kobject *kobj,
4051 struct kobj_attribute *attr, const char *buf, size_t count)
4052{
4053 char tmp[16];
4054 int huge;
4055
4056 if (count + 1 > sizeof(tmp))
4057 return -EINVAL;
4058 memcpy(tmp, buf, count);
4059 tmp[count] = '\0';
4060 if (count && tmp[count - 1] == '\n')
4061 tmp[count - 1] = '\0';
4062
4063 huge = shmem_parse_huge(tmp);
4064 if (huge == -EINVAL)
4065 return -EINVAL;
4066 if (!has_transparent_hugepage() &&
4067 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4068 return -EINVAL;
4069
4070 shmem_huge = huge;
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004071 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004072 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4073 return count;
4074}
4075
4076struct kobj_attribute shmem_enabled_attr =
4077 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004078#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004079
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004080#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004081bool shmem_huge_enabled(struct vm_area_struct *vma)
4082{
4083 struct inode *inode = file_inode(vma->vm_file);
4084 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4085 loff_t i_size;
4086 pgoff_t off;
4087
Yang Shic0630662019-07-18 15:57:27 -07004088 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4089 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4090 return false;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004091 if (shmem_huge == SHMEM_HUGE_FORCE)
4092 return true;
4093 if (shmem_huge == SHMEM_HUGE_DENY)
4094 return false;
4095 switch (sbinfo->huge) {
4096 case SHMEM_HUGE_NEVER:
4097 return false;
4098 case SHMEM_HUGE_ALWAYS:
4099 return true;
4100 case SHMEM_HUGE_WITHIN_SIZE:
4101 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4102 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4103 if (i_size >= HPAGE_PMD_SIZE &&
4104 i_size >> PAGE_SHIFT >= off)
4105 return true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07004106 fallthrough;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004107 case SHMEM_HUGE_ADVISE:
4108 /* TODO: implement fadvise() hints */
4109 return (vma->vm_flags & VM_HUGEPAGE);
4110 default:
4111 VM_BUG_ON(1);
4112 return false;
4113 }
4114}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004115#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004116
Matt Mackall853ac432009-01-06 14:40:20 -08004117#else /* !CONFIG_SHMEM */
4118
4119/*
4120 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4121 *
4122 * This is intended for small system where the benefits of the full
4123 * shmem code (swap-backed and resource-limited) are outweighed by
4124 * their complexity. On systems without swap this code should be
4125 * effectively equivalent, but much lighter weight.
4126 */
4127
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004128static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004129 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00004130 .init_fs_context = ramfs_init_fs_context,
Al Virod7167b12019-09-07 07:23:15 -04004131 .parameters = ramfs_fs_parameters,
Matt Mackall853ac432009-01-06 14:40:20 -08004132 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004133 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004134};
4135
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004136int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004137{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004138 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004139
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004140 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004141 BUG_ON(IS_ERR(shm_mnt));
4142
4143 return 0;
4144}
4145
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08004146int shmem_unuse(unsigned int type, bool frontswap,
4147 unsigned long *fs_pages_to_unuse)
Matt Mackall853ac432009-01-06 14:40:20 -08004148{
4149 return 0;
4150}
4151
Hugh Dickins3f96b792009-09-21 17:03:37 -07004152int shmem_lock(struct file *file, int lock, struct user_struct *user)
4153{
4154 return 0;
4155}
4156
Hugh Dickins24513262012-01-20 14:34:21 -08004157void shmem_unlock_mapping(struct address_space *mapping)
4158{
4159}
4160
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004161#ifdef CONFIG_MMU
4162unsigned long shmem_get_unmapped_area(struct file *file,
4163 unsigned long addr, unsigned long len,
4164 unsigned long pgoff, unsigned long flags)
4165{
4166 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4167}
4168#endif
4169
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004170void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004171{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004172 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004173}
4174EXPORT_SYMBOL_GPL(shmem_truncate_range);
4175
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004176#define shmem_vm_ops generic_file_vm_ops
4177#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004178#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004179#define shmem_acct_size(flags, size) 0
4180#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004181
4182#endif /* CONFIG_SHMEM */
4183
4184/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185
Matthew Auld703321b2017-10-06 23:18:13 +01004186static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
Eric Parisc7277092013-12-02 11:24:19 +00004187 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189 struct inode *inode;
Al Viro93dec2d2018-07-08 23:02:03 -04004190 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191
Matthew Auld703321b2017-10-06 23:18:13 +01004192 if (IS_ERR(mnt))
4193 return ERR_CAST(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194
Hugh Dickins285b2c42011-08-03 16:21:20 -07004195 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196 return ERR_PTR(-EINVAL);
4197
4198 if (shmem_acct_size(flags, size))
4199 return ERR_PTR(-ENOMEM);
4200
Al Viro93dec2d2018-07-08 23:02:03 -04004201 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4202 flags);
Al Virodac2d1f2018-06-09 09:27:41 -04004203 if (unlikely(!inode)) {
4204 shmem_unacct_size(flags, size);
4205 return ERR_PTR(-ENOSPC);
4206 }
Eric Parisc7277092013-12-02 11:24:19 +00004207 inode->i_flags |= i_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004208 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004209 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004210 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
Al Viro93dec2d2018-07-08 23:02:03 -04004211 if (!IS_ERR(res))
4212 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4213 &shmem_file_operations);
Al Viro26567cd2013-03-01 20:22:53 -05004214 if (IS_ERR(res))
Al Viro93dec2d2018-07-08 23:02:03 -04004215 iput(inode);
Al Viro6b4d0b22013-02-14 21:37:26 -05004216 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217}
Eric Parisc7277092013-12-02 11:24:19 +00004218
4219/**
4220 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4221 * kernel internal. There will be NO LSM permission checks against the
4222 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004223 * higher layer. The users are the big_key and shm implementations. LSM
4224 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004225 * @name: name for dentry (to be seen in /proc/<pid>/maps
4226 * @size: size to be set for the file
4227 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4228 */
4229struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4230{
Matthew Auld703321b2017-10-06 23:18:13 +01004231 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
Eric Parisc7277092013-12-02 11:24:19 +00004232}
4233
4234/**
4235 * shmem_file_setup - get an unlinked file living in tmpfs
4236 * @name: name for dentry (to be seen in /proc/<pid>/maps
4237 * @size: size to be set for the file
4238 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4239 */
4240struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4241{
Matthew Auld703321b2017-10-06 23:18:13 +01004242 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
Eric Parisc7277092013-12-02 11:24:19 +00004243}
Keith Packard395e0dd2008-06-20 00:08:06 -07004244EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245
Randy Dunlap46711812008-03-19 17:00:41 -07004246/**
Matthew Auld703321b2017-10-06 23:18:13 +01004247 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4248 * @mnt: the tmpfs mount where the file will be created
4249 * @name: name for dentry (to be seen in /proc/<pid>/maps
4250 * @size: size to be set for the file
4251 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4252 */
4253struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4254 loff_t size, unsigned long flags)
4255{
4256 return __shmem_file_setup(mnt, name, size, flags, 0);
4257}
4258EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4259
4260/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261 * shmem_zero_setup - setup a shared anonymous mapping
Peter Collingbourne45e55302020-08-06 23:23:37 -07004262 * @vma: the vma to be mmapped is prepared by do_mmap
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 */
4264int shmem_zero_setup(struct vm_area_struct *vma)
4265{
4266 struct file *file;
4267 loff_t size = vma->vm_end - vma->vm_start;
4268
Hugh Dickins66fc1302015-06-14 09:48:09 -07004269 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07004270 * Cloning a new file under mmap_lock leads to a lock ordering conflict
Hugh Dickins66fc1302015-06-14 09:48:09 -07004271 * between XFS directory reading and selinux: since this file is only
4272 * accessible to the user through its mapping, use S_PRIVATE flag to
4273 * bypass file security, in the same way as shmem_kernel_file_setup().
4274 */
Matthew Auld703321b2017-10-06 23:18:13 +01004275 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276 if (IS_ERR(file))
4277 return PTR_ERR(file);
4278
4279 if (vma->vm_file)
4280 fput(vma->vm_file);
4281 vma->vm_file = file;
4282 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004283
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004284 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004285 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4286 (vma->vm_end & HPAGE_PMD_MASK)) {
4287 khugepaged_enter(vma, vma->vm_flags);
4288 }
4289
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290 return 0;
4291}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004292
4293/**
4294 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4295 * @mapping: the page's address_space
4296 * @index: the page index
4297 * @gfp: the page allocator flags to use if allocating
4298 *
4299 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4300 * with any new page allocations done using the specified allocation flags.
4301 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4302 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4303 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4304 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004305 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4306 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004307 */
4308struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4309 pgoff_t index, gfp_t gfp)
4310{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004311#ifdef CONFIG_SHMEM
4312 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004313 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004314 int error;
4315
4316 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004317 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
Mike Rapoportcfda0522017-02-22 15:43:37 -08004318 gfp, NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004319 if (error)
4320 page = ERR_PTR(error);
4321 else
4322 unlock_page(page);
4323 return page;
4324#else
4325 /*
4326 * The tiny !SHMEM case uses ramfs without swap
4327 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004328 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004329#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004330}
4331EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);