blob: 4f11b55063631976af81e4221c7366b768db6690 [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>
Theodore Ts'oe408e692022-07-14 21:59:12 -040031#include <linux/fileattr.h>
Matt Mackall853ac432009-01-06 14:40:20 -080032#include <linux/mm.h>
Arnd Bergmann46c9a942018-08-17 15:45:09 -070033#include <linux/random.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010034#include <linux/sched/signal.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040035#include <linux/export.h>
Matthew Wilcox (Oracle)5ff21212023-02-06 19:08:50 +000036#include <linux/shmem_fs.h>
Matt Mackall853ac432009-01-06 14:40:20 -080037#include <linux/swap.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080038#include <linux/uio.h>
Mike Kravetz749df872017-09-06 16:24:16 -070039#include <linux/hugetlb.h>
Al Viro626c3922019-09-08 20:28:06 -040040#include <linux/fs_parser.h>
Miaohe Lin86a2f3f2021-09-02 14:54:15 -070041#include <linux/swapfile.h>
Jeff Layton36f05ca2022-09-09 09:00:31 -040042#include <linux/iversion.h>
NeilBrown014bb1d2022-05-09 18:20:47 -070043#include "swap.h"
Andrea Arcangeli95cc09d2017-02-22 15:43:31 -080044
Alexey Dobriyan68279f92023-10-11 19:55:00 +030045static struct vfsmount *shm_mnt __ro_after_init;
Matt Mackall853ac432009-01-06 14:40:20 -080046
47#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070048/*
49 * This virtual memory filesystem is heavily based on the ramfs. It
50 * extends ramfs by the ability to use swap and honor resource limits
51 * which makes it a completely usable filesystem.
52 */
53
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070054#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070055#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010056#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080057#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/string.h>
60#include <linux/slab.h>
61#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#include <linux/writeback.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070063#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070064#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070065#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070066#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <linux/security.h>
68#include <linux/swapops.h>
69#include <linux/mempolicy.h>
70#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000071#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070072#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070073#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080074#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040075#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070076#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070077#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070078#include <uapi/linux/memfd.h>
Mike Rapoport4c27fe42017-02-22 15:43:25 -080079#include <linux/rmap.h>
Amir Goldstein2b4db792017-05-18 15:29:33 +030080#include <linux/uuid.h>
Carlos Maiolinoe09764c2023-07-25 16:45:08 +020081#include <linux/quotaops.h>
Kent Overstreet1e2f2d32023-12-15 15:51:54 -050082#include <linux/rcupdate_wait.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070083
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080084#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Mel Gormandd56b042015-11-06 16:28:43 -080086#include "internal.h"
87
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030088#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/* Pretend that each entry is of this size in directory's i_size */
92#define BOGO_DIRENT_SIZE 20
93
Hugh Dickinse07c4692023-08-08 21:32:21 -070094/* Pretend that one inode + its dentry occupy this much memory */
95#define BOGO_INODE_SIZE 1024
96
Hugh Dickins69f07ec2011-08-03 16:21:26 -070097/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
98#define SHORT_SYMLINK_LEN 128
99
Hugh Dickins1aac1402012-05-29 15:06:42 -0700100/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -0700101 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
Jan Kara96087032021-04-12 15:50:21 +0200102 * inode->i_private (with i_rwsem making sure that it has only one user at
Hugh Dickinsf00cdc62014-06-23 13:22:06 -0700103 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700104 */
105struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -0700106 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -0700107 pgoff_t start; /* start of range currently being fallocated */
108 pgoff_t next; /* the next page offset to be fallocated */
109 pgoff_t nr_falloced; /* how many new pages have been fallocated */
110 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
111};
112
Al Viro0b5071d2019-09-08 18:49:18 -0400113struct shmem_options {
114 unsigned long long blocks;
115 unsigned long long inodes;
116 struct mempolicy *mpol;
117 kuid_t uid;
118 kgid_t gid;
119 umode_t mode;
Chris Downea3271f2020-08-06 23:20:25 -0700120 bool full_inums;
Al Viro0b5071d2019-09-08 18:49:18 -0400121 int huge;
122 int seen;
Luis Chamberlain2c6efe92023-03-09 15:05:45 -0800123 bool noswap;
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200124 unsigned short quota_types;
Lukas Czernerde4c0e72023-07-25 16:45:09 +0200125 struct shmem_quota_limits qlimits;
Al Viro0b5071d2019-09-08 18:49:18 -0400126#define SHMEM_SEEN_BLOCKS 1
127#define SHMEM_SEEN_INODES 2
128#define SHMEM_SEEN_HUGE 4
Chris Downea3271f2020-08-06 23:20:25 -0700129#define SHMEM_SEEN_INUMS 8
Luis Chamberlain2c6efe92023-03-09 15:05:45 -0800130#define SHMEM_SEEN_NOSWAP 16
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200131#define SHMEM_SEEN_QUOTA 32
Al Viro0b5071d2019-09-08 18:49:18 -0400132};
133
Baolin Wang4b989952024-06-11 18:11:07 +0800134#ifdef CONFIG_TRANSPARENT_HUGEPAGE
135static unsigned long huge_shmem_orders_always __read_mostly;
136static unsigned long huge_shmem_orders_madvise __read_mostly;
137static unsigned long huge_shmem_orders_inherit __read_mostly;
138static unsigned long huge_shmem_orders_within_size __read_mostly;
139#endif
140
Andrew Mortonb76db7352008-02-08 04:21:49 -0800141#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800142static unsigned long shmem_default_max_blocks(void)
143{
Arun KSca79b0c2018-12-28 00:34:29 -0800144 return totalram_pages() / 2;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800145}
146
147static unsigned long shmem_default_max_inodes(void)
148{
Arun KSca79b0c2018-12-28 00:34:29 -0800149 unsigned long nr_pages = totalram_pages();
150
Hugh Dickinse07c4692023-08-08 21:32:21 -0700151 return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
152 ULONG_MAX / BOGO_INODE_SIZE);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800153}
Andrew Mortonb76db7352008-02-08 04:21:49 -0800154#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800155
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -0700156static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
Hugh Dickins054a9f72023-09-29 20:31:27 -0700157 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
Baolin Wang736f0e02024-08-12 15:42:07 +0800158 struct vm_area_struct *vma, vm_fault_t *fault_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161{
162 return sb->s_fs_info;
163}
164
165/*
166 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
167 * for shared memory and for shared anonymous (/dev/zero) mappings
168 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
169 * consistent with the pre-accounting of private mappings ...
170 */
171static inline int shmem_acct_size(unsigned long flags, loff_t size)
172{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000173 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000174 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
177static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000179 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 vm_unacct_memory(VM_ACCT(size));
181}
182
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700183static inline int shmem_reacct_size(unsigned long flags,
184 loff_t oldsize, loff_t newsize)
185{
186 if (!(flags & VM_NORESERVE)) {
187 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
188 return security_vm_enough_memory_mm(current->mm,
189 VM_ACCT(newsize) - VM_ACCT(oldsize));
190 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
191 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 }
193 return 0;
194}
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196/*
197 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700198 * pages are allocated, in order to allow large sparse files.
Hugh Dickins4199f512023-09-29 20:30:03 -0700199 * shmem_get_folio reports shmem_acct_blocks failure as -ENOSPC not -ENOMEM,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 */
Hugh Dickins4199f512023-09-29 20:30:03 -0700202static inline int shmem_acct_blocks(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700204 if (!(flags & VM_NORESERVE))
205 return 0;
206
207 return security_vm_enough_memory_mm(current->mm,
208 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
210
211static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000213 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300214 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
Hugh Dickins4199f512023-09-29 20:30:03 -0700217static int shmem_inode_acct_blocks(struct inode *inode, long pages)
Mike Rapoport0f079692017-09-06 16:22:59 -0700218{
219 struct shmem_inode_info *info = SHMEM_I(inode);
220 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Lukas Czernerc7e263a2023-07-25 16:45:04 +0200221 int err = -ENOSPC;
Mike Rapoport0f079692017-09-06 16:22:59 -0700222
Hugh Dickins4199f512023-09-29 20:30:03 -0700223 if (shmem_acct_blocks(info->flags, pages))
Lukas Czernerc7e263a2023-07-25 16:45:04 +0200224 return err;
Mike Rapoport0f079692017-09-06 16:22:59 -0700225
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700226 might_sleep(); /* when quotas */
Mike Rapoport0f079692017-09-06 16:22:59 -0700227 if (sbinfo->max_blocks) {
Hugh Dickinsbeb98682023-09-29 20:42:45 -0700228 if (!percpu_counter_limited_add(&sbinfo->used_blocks,
229 sbinfo->max_blocks, pages))
Mike Rapoport0f079692017-09-06 16:22:59 -0700230 goto unacct;
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200231
232 err = dquot_alloc_block_nodirty(inode, pages);
Hugh Dickinsbeb98682023-09-29 20:42:45 -0700233 if (err) {
234 percpu_counter_sub(&sbinfo->used_blocks, pages);
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200235 goto unacct;
Hugh Dickinsbeb98682023-09-29 20:42:45 -0700236 }
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200237 } else {
238 err = dquot_alloc_block_nodirty(inode, pages);
239 if (err)
240 goto unacct;
Mike Rapoport0f079692017-09-06 16:22:59 -0700241 }
242
Lukas Czernerc7e263a2023-07-25 16:45:04 +0200243 return 0;
Mike Rapoport0f079692017-09-06 16:22:59 -0700244
245unacct:
246 shmem_unacct_blocks(info->flags, pages);
Lukas Czernerc7e263a2023-07-25 16:45:04 +0200247 return err;
Mike Rapoport0f079692017-09-06 16:22:59 -0700248}
249
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700250static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
Mike Rapoport0f079692017-09-06 16:22:59 -0700251{
252 struct shmem_inode_info *info = SHMEM_I(inode);
253 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
254
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700255 might_sleep(); /* when quotas */
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200256 dquot_free_block_nodirty(inode, pages);
257
Mike Rapoport0f079692017-09-06 16:22:59 -0700258 if (sbinfo->max_blocks)
259 percpu_counter_sub(&sbinfo->used_blocks, pages);
260 shmem_unacct_blocks(info->flags, pages);
261}
262
Hugh Dickins759b9772007-03-05 00:30:28 -0800263static const struct super_operations shmem_ops;
Christoph Hellwigaefacb22024-02-19 07:27:10 +0100264static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800265static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800266static const struct inode_operations shmem_inode_operations;
267static const struct inode_operations shmem_dir_inode_operations;
268static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +0400269static const struct vm_operations_struct shmem_vm_ops;
Pasha Tatashind09e8ca2022-11-15 02:06:01 +0000270static const struct vm_operations_struct shmem_anon_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700271static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Christoph Hellwigaefacb22024-02-19 07:27:10 +0100273bool shmem_mapping(struct address_space *mapping)
274{
275 return mapping->a_ops == &shmem_aops;
276}
277EXPORT_SYMBOL_GPL(shmem_mapping);
278
Pasha Tatashind09e8ca2022-11-15 02:06:01 +0000279bool vma_is_anon_shmem(struct vm_area_struct *vma)
280{
281 return vma->vm_ops == &shmem_anon_vm_ops;
282}
283
Mike Rapoportb0506e42017-02-22 15:43:28 -0800284bool vma_is_shmem(struct vm_area_struct *vma)
285{
Pasha Tatashind09e8ca2022-11-15 02:06:01 +0000286 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
Mike Rapoportb0506e42017-02-22 15:43:28 -0800287}
288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800290static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200292#ifdef CONFIG_TMPFS_QUOTA
293
294static int shmem_enable_quotas(struct super_block *sb,
295 unsigned short quota_types)
296{
297 int type, err = 0;
298
299 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
300 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
301 if (!(quota_types & (1 << type)))
302 continue;
303 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
304 DQUOT_USAGE_ENABLED |
305 DQUOT_LIMITS_ENABLED);
306 if (err)
307 goto out_err;
308 }
309 return 0;
310
311out_err:
312 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
313 type, err);
314 for (type--; type >= 0; type--)
315 dquot_quota_off(sb, type);
316 return err;
317}
318
319static void shmem_disable_quotas(struct super_block *sb)
320{
321 int type;
322
323 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
324 dquot_quota_off(sb, type);
325}
326
Jan Karaccb49012024-02-06 15:08:19 +0100327static struct dquot __rcu **shmem_get_dquots(struct inode *inode)
Carlos Maiolinoe09764c2023-07-25 16:45:08 +0200328{
329 return SHMEM_I(inode)->i_dquot;
330}
331#endif /* CONFIG_TMPFS_QUOTA */
332
Chris Downe809d5f2020-08-06 23:20:20 -0700333/*
334 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
335 * produces a novel ino for the newly allocated inode.
336 *
337 * It may also be called when making a hard link to permit the space needed by
338 * each dentry. However, in that case, no new inode number is needed since that
339 * internally draws from another pool of inode numbers (currently global
340 * get_next_ino()). This case is indicated by passing NULL as inop.
341 */
342#define SHMEM_INO_BATCH 1024
343static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800344{
345 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -0700346 ino_t ino;
347
348 if (!(sb->s_flags & SB_KERNMOUNT)) {
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700349 raw_spin_lock(&sbinfo->stat_lock);
Byron Stanoszekbb3e96d2020-09-18 21:20:18 -0700350 if (sbinfo->max_inodes) {
Hugh Dickinse07c4692023-08-08 21:32:21 -0700351 if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700352 raw_spin_unlock(&sbinfo->stat_lock);
Byron Stanoszekbb3e96d2020-09-18 21:20:18 -0700353 return -ENOSPC;
354 }
Hugh Dickinse07c4692023-08-08 21:32:21 -0700355 sbinfo->free_ispace -= BOGO_INODE_SIZE;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800356 }
Chris Downe809d5f2020-08-06 23:20:20 -0700357 if (inop) {
358 ino = sbinfo->next_ino++;
359 if (unlikely(is_zero_ino(ino)))
360 ino = sbinfo->next_ino++;
Chris Downea3271f2020-08-06 23:20:25 -0700361 if (unlikely(!sbinfo->full_inums &&
362 ino > UINT_MAX)) {
Chris Downe809d5f2020-08-06 23:20:20 -0700363 /*
364 * Emulate get_next_ino uint wraparound for
365 * compatibility
366 */
Chris Downea3271f2020-08-06 23:20:25 -0700367 if (IS_ENABLED(CONFIG_64BIT))
368 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
369 __func__, MINOR(sb->s_dev));
370 sbinfo->next_ino = 1;
371 ino = sbinfo->next_ino++;
Chris Downe809d5f2020-08-06 23:20:20 -0700372 }
373 *inop = ino;
374 }
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700375 raw_spin_unlock(&sbinfo->stat_lock);
Chris Downe809d5f2020-08-06 23:20:20 -0700376 } else if (inop) {
377 /*
378 * __shmem_file_setup, one of our callers, is lock-free: it
379 * doesn't hold stat_lock in shmem_reserve_inode since
380 * max_inodes is always 0, and is called from potentially
381 * unknown contexts. As such, use a per-cpu batched allocator
382 * which doesn't require the per-sb stat_lock unless we are at
383 * the batch boundary.
Chris Downea3271f2020-08-06 23:20:25 -0700384 *
385 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
386 * shmem mounts are not exposed to userspace, so we don't need
387 * to worry about things like glibc compatibility.
Chris Downe809d5f2020-08-06 23:20:20 -0700388 */
389 ino_t *next_ino;
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700390
Chris Downe809d5f2020-08-06 23:20:20 -0700391 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
392 ino = *next_ino;
393 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700394 raw_spin_lock(&sbinfo->stat_lock);
Chris Downe809d5f2020-08-06 23:20:20 -0700395 ino = sbinfo->next_ino;
396 sbinfo->next_ino += SHMEM_INO_BATCH;
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700397 raw_spin_unlock(&sbinfo->stat_lock);
Chris Downe809d5f2020-08-06 23:20:20 -0700398 if (unlikely(is_zero_ino(ino)))
399 ino++;
400 }
401 *inop = ino;
402 *next_ino = ++ino;
403 put_cpu();
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800404 }
Chris Downe809d5f2020-08-06 23:20:20 -0700405
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800406 return 0;
407}
408
Hugh Dickins2daf18a2023-08-08 21:33:56 -0700409static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800410{
411 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
412 if (sbinfo->max_inodes) {
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700413 raw_spin_lock(&sbinfo->stat_lock);
Hugh Dickins2daf18a2023-08-08 21:33:56 -0700414 sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -0700415 raw_spin_unlock(&sbinfo->stat_lock);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800416 }
417}
418
Randy Dunlap46711812008-03-19 17:00:41 -0700419/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700420 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 * @inode: inode to recalc
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700422 * @alloced: the change in number of pages allocated to inode
423 * @swapped: the change in number of pages swapped from inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 *
425 * We have to calculate the free blocks since the mm can drop
426 * undirtied hole pages behind our back.
427 *
428 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
429 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 */
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700431static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 struct shmem_inode_info *info = SHMEM_I(inode);
434 long freed;
435
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700436 spin_lock(&info->lock);
437 info->alloced += alloced;
438 info->swapped += swapped;
439 freed = info->alloced - info->swapped -
440 READ_ONCE(inode->i_mapping->nrpages);
441 /*
442 * Special case: whereas normally shmem_recalc_inode() is called
443 * after i_mapping->nrpages has already been adjusted (up or down),
444 * shmem_writepage() has to raise swapped before nrpages is lowered -
445 * to stop a racing shmem_recalc_inode() from thinking that a page has
446 * been freed. Compensate here, to avoid the need for a followup call.
447 */
448 if (swapped > 0)
449 freed += swapped;
450 if (freed > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 info->alloced -= freed;
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700452 spin_unlock(&info->lock);
453
454 /* The quota case may block */
455 if (freed > 0)
Mike Rapoport0f079692017-09-06 16:22:59 -0700456 shmem_inode_unacct_blocks(inode, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
458
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700459bool shmem_charge(struct inode *inode, long pages)
460{
Hugh Dickins509f0062023-07-25 16:45:10 +0200461 struct address_space *mapping = inode->i_mapping;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700462
Hugh Dickins4199f512023-09-29 20:30:03 -0700463 if (shmem_inode_acct_blocks(inode, pages))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700464 return false;
Mike Rapoportb1cc94a2017-09-06 16:22:56 -0700465
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800466 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
Hugh Dickins509f0062023-07-25 16:45:10 +0200467 xa_lock_irq(&mapping->i_pages);
468 mapping->nrpages += pages;
469 xa_unlock_irq(&mapping->i_pages);
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800470
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700471 shmem_recalc_inode(inode, pages, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700472 return true;
473}
474
475void shmem_uncharge(struct inode *inode, long pages)
476{
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700477 /* pages argument is currently unused: keep it to help debugging */
Matthew Wilcox (Oracle)6ffcd822022-06-28 20:41:40 -0400478 /* nrpages adjustment done by __filemap_remove_folio() or caller */
Hugh Dickinsaaa52e32018-11-30 14:10:29 -0800479
Hugh Dickins3c1b7522023-08-03 22:46:11 -0700480 shmem_recalc_inode(inode, 0, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700481}
482
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700483/*
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500484 * Replace item expected in xarray by a new item, while holding xa_lock.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700485 */
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500486static int shmem_replace_entry(struct address_space *mapping,
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700487 pgoff_t index, void *expected, void *replacement)
488{
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500489 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700490 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700491
492 VM_BUG_ON(!expected);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700493 VM_BUG_ON(!replacement);
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500494 item = xas_load(&xas);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700495 if (item != expected)
496 return -ENOENT;
Matthew Wilcox62f945b2017-11-17 10:22:37 -0500497 xas_store(&xas, replacement);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700498 return 0;
499}
500
501/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700502 * Sometimes, before we decide whether to proceed or to fail, we must check
503 * that an entry was not already brought back from swap by a racing thread.
504 *
Matthew Wilcox (Oracle)32f51ea2024-08-21 20:34:37 +0100505 * Checking folio is not enough: by the time a swapcache folio is locked, it
506 * might be reused, and again be swapcache, using the same swap as before.
Hugh Dickinsd1899222012-07-11 14:02:47 -0700507 */
508static bool shmem_confirm_swap(struct address_space *mapping,
509 pgoff_t index, swp_entry_t swap)
510{
Matthew Wilcoxa12831b2017-11-22 08:34:58 -0500511 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
Hugh Dickinsd1899222012-07-11 14:02:47 -0700512}
513
514/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700515 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
516 *
517 * SHMEM_HUGE_NEVER:
518 * disables huge pages for the mount;
519 * SHMEM_HUGE_ALWAYS:
520 * enables huge pages for the mount;
521 * SHMEM_HUGE_WITHIN_SIZE:
522 * only allocate huge pages if the page will be fully within i_size,
523 * also respect fadvise()/madvise() hints;
524 * SHMEM_HUGE_ADVISE:
525 * only allocate huge pages if requested with fadvise()/madvise();
526 */
527
528#define SHMEM_HUGE_NEVER 0
529#define SHMEM_HUGE_ALWAYS 1
530#define SHMEM_HUGE_WITHIN_SIZE 2
531#define SHMEM_HUGE_ADVISE 3
532
533/*
534 * Special values.
535 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
536 *
537 * SHMEM_HUGE_DENY:
538 * disables huge on shm_mnt and all mounts, for emergency use;
539 * SHMEM_HUGE_FORCE:
540 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
541 *
542 */
543#define SHMEM_HUGE_DENY (-1)
544#define SHMEM_HUGE_FORCE (-2)
545
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700546#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700547/* ifdef here to avoid bloating shmem.o when not necessary */
548
Hugh Dickins5e6e5a12021-09-02 14:54:37 -0700549static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700550
Baolin Wangd58a2a52024-07-22 13:43:18 +0800551static bool __shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
Rik van Riele1e4cfd2024-09-03 11:19:28 -0400552 loff_t write_end, bool shmem_huge_force,
553 struct vm_area_struct *vma,
Baolin Wangd58a2a52024-07-22 13:43:18 +0800554 unsigned long vm_flags)
Hugh Dickinsc8520232021-09-02 14:54:31 -0700555{
Baolin Wang6beeab82024-07-22 13:43:19 +0800556 struct mm_struct *mm = vma ? vma->vm_mm : NULL;
Hugh Dickinsc8520232021-09-02 14:54:31 -0700557 loff_t i_size;
Hugh Dickinsc8520232021-09-02 14:54:31 -0700558
Xavier Rochef7cd16a2022-03-22 14:39:55 -0700559 if (!S_ISREG(inode->i_mode))
560 return false;
David Stevens2cf13382023-01-13 11:30:11 +0900561 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
Hugh Dickinsc8520232021-09-02 14:54:31 -0700562 return false;
Zach O'Keefe7c6c6cc2022-09-22 15:40:37 -0700563 if (shmem_huge == SHMEM_HUGE_DENY)
564 return false;
Zach O'Keefe3de0c262022-12-24 00:20:35 -0800565 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
566 return true;
Hugh Dickins5e6e5a12021-09-02 14:54:37 -0700567
568 switch (SHMEM_SB(inode->i_sb)->huge) {
Hugh Dickinsc8520232021-09-02 14:54:31 -0700569 case SHMEM_HUGE_ALWAYS:
570 return true;
571 case SHMEM_HUGE_WITHIN_SIZE:
Liu Yuntaode6ee652021-09-24 15:43:32 -0700572 index = round_up(index + 1, HPAGE_PMD_NR);
Rik van Riele1e4cfd2024-09-03 11:19:28 -0400573 i_size = max(write_end, i_size_read(inode));
574 i_size = round_up(i_size, PAGE_SIZE);
Liu Yuntaode6ee652021-09-24 15:43:32 -0700575 if (i_size >> PAGE_SHIFT >= index)
Hugh Dickinsc8520232021-09-02 14:54:31 -0700576 return true;
577 fallthrough;
578 case SHMEM_HUGE_ADVISE:
David Stevens2cf13382023-01-13 11:30:11 +0900579 if (mm && (vm_flags & VM_HUGEPAGE))
Hugh Dickins5e6e5a12021-09-02 14:54:37 -0700580 return true;
581 fallthrough;
Hugh Dickinsc8520232021-09-02 14:54:31 -0700582 default:
Hugh Dickinsc8520232021-09-02 14:54:31 -0700583 return false;
584 }
585}
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700586
Baolin Wang6beeab82024-07-22 13:43:19 +0800587static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
Rik van Riele1e4cfd2024-09-03 11:19:28 -0400588 loff_t write_end, bool shmem_huge_force,
589 struct vm_area_struct *vma, unsigned long vm_flags)
Gavin Shan9fd154b2024-06-27 10:39:52 +1000590{
591 if (HPAGE_PMD_ORDER > MAX_PAGECACHE_ORDER)
592 return false;
593
Rik van Riele1e4cfd2024-09-03 11:19:28 -0400594 return __shmem_huge_global_enabled(inode, index, write_end,
595 shmem_huge_force, vma, vm_flags);
Gavin Shan9fd154b2024-06-27 10:39:52 +1000596}
597
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700598#if defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700599static int shmem_parse_huge(const char *str)
600{
601 if (!strcmp(str, "never"))
602 return SHMEM_HUGE_NEVER;
603 if (!strcmp(str, "always"))
604 return SHMEM_HUGE_ALWAYS;
605 if (!strcmp(str, "within_size"))
606 return SHMEM_HUGE_WITHIN_SIZE;
607 if (!strcmp(str, "advise"))
608 return SHMEM_HUGE_ADVISE;
609 if (!strcmp(str, "deny"))
610 return SHMEM_HUGE_DENY;
611 if (!strcmp(str, "force"))
612 return SHMEM_HUGE_FORCE;
613 return -EINVAL;
614}
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700615#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700616
Arnd Bergmanne5f22492019-07-16 16:25:51 -0700617#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700618static const char *shmem_format_huge(int huge)
619{
620 switch (huge) {
621 case SHMEM_HUGE_NEVER:
622 return "never";
623 case SHMEM_HUGE_ALWAYS:
624 return "always";
625 case SHMEM_HUGE_WITHIN_SIZE:
626 return "within_size";
627 case SHMEM_HUGE_ADVISE:
628 return "advise";
629 case SHMEM_HUGE_DENY:
630 return "deny";
631 case SHMEM_HUGE_FORCE:
632 return "force";
633 default:
634 VM_BUG_ON(1);
635 return "bad_val";
636 }
637}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800638#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700639
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700640static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
Hugh Dickins15444052024-08-25 16:25:39 -0700641 struct shrink_control *sc, unsigned long nr_to_free)
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700642{
643 LIST_HEAD(list), *pos, *next;
644 struct inode *inode;
645 struct shmem_inode_info *info;
Matthew Wilcox (Oracle)05624572022-05-12 20:23:03 -0700646 struct folio *folio;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700647 unsigned long batch = sc ? sc->nr_to_scan : 128;
Hugh Dickins15444052024-08-25 16:25:39 -0700648 unsigned long split = 0, freed = 0;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700649
650 if (list_empty(&sbinfo->shrinklist))
651 return SHRINK_STOP;
652
653 spin_lock(&sbinfo->shrinklist_lock);
654 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
655 info = list_entry(pos, struct shmem_inode_info, shrinklist);
656
657 /* pin the inode */
658 inode = igrab(&info->vfs_inode);
659
660 /* inode is about to be evicted */
661 if (!inode) {
662 list_del_init(&info->shrinklist);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700663 goto next;
664 }
665
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700666 list_move(&info->shrinklist, &list);
667next:
Gang Li62c98272022-01-14 14:05:23 -0800668 sbinfo->shrinklist_len--;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700669 if (!--batch)
670 break;
671 }
672 spin_unlock(&sbinfo->shrinklist_lock);
673
674 list_for_each_safe(pos, next, &list) {
Hugh Dickins15444052024-08-25 16:25:39 -0700675 pgoff_t next, end;
676 loff_t i_size;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700677 int ret;
678
679 info = list_entry(pos, struct shmem_inode_info, shrinklist);
680 inode = &info->vfs_inode;
681
Hugh Dickins15444052024-08-25 16:25:39 -0700682 if (nr_to_free && freed >= nr_to_free)
Gang Li62c98272022-01-14 14:05:23 -0800683 goto move_back;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700684
Hugh Dickins15444052024-08-25 16:25:39 -0700685 i_size = i_size_read(inode);
686 folio = filemap_get_entry(inode->i_mapping, i_size / PAGE_SIZE);
687 if (!folio || xa_is_value(folio))
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700688 goto drop;
689
Hugh Dickins15444052024-08-25 16:25:39 -0700690 /* No large folio at the end of the file: nothing to split */
Matthew Wilcox (Oracle)05624572022-05-12 20:23:03 -0700691 if (!folio_test_large(folio)) {
692 folio_put(folio);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700693 goto drop;
694 }
695
Hugh Dickins15444052024-08-25 16:25:39 -0700696 /* Check if there is anything to gain from splitting */
697 next = folio_next_index(folio);
698 end = shmem_fallocend(inode, DIV_ROUND_UP(i_size, PAGE_SIZE));
699 if (end <= folio->index || end >= next) {
700 folio_put(folio);
701 goto drop;
702 }
703
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700704 /*
Gang Li62c98272022-01-14 14:05:23 -0800705 * Move the inode on the list back to shrinklist if we failed
706 * to lock the page at this time.
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700707 *
708 * Waiting for the lock may lead to deadlock in the
709 * reclaim path.
710 */
Matthew Wilcox (Oracle)05624572022-05-12 20:23:03 -0700711 if (!folio_trylock(folio)) {
712 folio_put(folio);
Gang Li62c98272022-01-14 14:05:23 -0800713 goto move_back;
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700714 }
715
Matthew Wilcox (Oracle)d788f5b2022-09-02 20:46:00 +0100716 ret = split_folio(folio);
Matthew Wilcox (Oracle)05624572022-05-12 20:23:03 -0700717 folio_unlock(folio);
718 folio_put(folio);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700719
Gang Li62c98272022-01-14 14:05:23 -0800720 /* If split failed move the inode on the list back to shrinklist */
Kirill A. Shutemovb3cd54b2018-03-22 16:17:35 -0700721 if (ret)
Gang Li62c98272022-01-14 14:05:23 -0800722 goto move_back;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700723
Hugh Dickins15444052024-08-25 16:25:39 -0700724 freed += next - end;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700725 split++;
726drop:
727 list_del_init(&info->shrinklist);
Gang Li62c98272022-01-14 14:05:23 -0800728 goto put;
729move_back:
730 /*
731 * Make sure the inode is either on the global list or deleted
732 * from any local list before iput() since it could be deleted
733 * in another thread once we put the inode (then the local list
734 * is corrupted).
735 */
736 spin_lock(&sbinfo->shrinklist_lock);
737 list_move(&info->shrinklist, &sbinfo->shrinklist);
738 sbinfo->shrinklist_len++;
739 spin_unlock(&sbinfo->shrinklist_lock);
740put:
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700741 iput(inode);
742 }
743
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700744 return split;
745}
746
747static long shmem_unused_huge_scan(struct super_block *sb,
748 struct shrink_control *sc)
749{
750 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
751
752 if (!READ_ONCE(sbinfo->shrinklist_len))
753 return SHRINK_STOP;
754
755 return shmem_unused_huge_shrink(sbinfo, sc, 0);
756}
757
758static long shmem_unused_huge_count(struct super_block *sb,
759 struct shrink_control *sc)
760{
761 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
762 return READ_ONCE(sbinfo->shrinklist_len);
763}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700764#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700765
766#define shmem_huge SHMEM_HUGE_DENY
767
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700768static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
Hugh Dickins15444052024-08-25 16:25:39 -0700769 struct shrink_control *sc, unsigned long nr_to_free)
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700770{
771 return 0;
772}
Baolin Wang6beeab82024-07-22 13:43:19 +0800773
774static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
Rik van Riele1e4cfd2024-09-03 11:19:28 -0400775 loff_t write_end, bool shmem_huge_force,
776 struct vm_area_struct *vma, unsigned long vm_flags)
Baolin Wang6beeab82024-07-22 13:43:19 +0800777{
778 return false;
779}
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -0700780#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700781
782/*
Hugh Dickins9be7d5b2023-09-29 20:28:50 -0700783 * Somewhat like filemap_add_folio, but error if expected item has gone.
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700784 */
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700785static int shmem_add_to_page_cache(struct folio *folio,
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700786 struct address_space *mapping,
Hugh Dickins054a9f72023-09-29 20:31:27 -0700787 pgoff_t index, void *expected, gfp_t gfp)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700788{
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700789 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
790 long nr = folio_nr_pages(folio);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700791
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700792 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
793 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
794 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700795
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700796 folio_ref_add(folio, nr);
797 folio->mapping = mapping;
798 folio->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700799
Hugh Dickins054a9f72023-09-29 20:31:27 -0700800 gfp &= GFP_RECLAIM_MASK;
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700801 folio_throttle_swaprate(folio, gfp);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700802
Matthew Wilcox552446a2017-12-01 13:25:14 -0500803 do {
Matthew Wilcox552446a2017-12-01 13:25:14 -0500804 xas_lock_irq(&xas);
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -0400805 if (expected != xas_find_conflict(&xas)) {
Matthew Wilcox552446a2017-12-01 13:25:14 -0500806 xas_set_err(&xas, -EEXIST);
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -0400807 goto unlock;
808 }
809 if (expected && xas_find_conflict(&xas)) {
810 xas_set_err(&xas, -EEXIST);
811 goto unlock;
812 }
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700813 xas_store(&xas, folio);
Matthew Wilcox552446a2017-12-01 13:25:14 -0500814 if (xas_error(&xas))
815 goto unlock;
Hugh Dickins3022fd72023-09-29 20:32:40 -0700816 if (folio_test_pmd_mappable(folio))
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700817 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -0700818 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
819 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
Hugh Dickins3022fd72023-09-29 20:32:40 -0700820 mapping->nrpages += nr;
Matthew Wilcox552446a2017-12-01 13:25:14 -0500821unlock:
822 xas_unlock_irq(&xas);
823 } while (xas_nomem(&xas, gfp));
824
825 if (xas_error(&xas)) {
Hugh Dickins054a9f72023-09-29 20:31:27 -0700826 folio->mapping = NULL;
827 folio_ref_sub(folio, nr);
828 return xas_error(&xas);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700829 }
Matthew Wilcox552446a2017-12-01 13:25:14 -0500830
831 return 0;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700832}
833
834/*
Hugh Dickins9be7d5b2023-09-29 20:28:50 -0700835 * Somewhat like filemap_remove_folio, but substitutes swap for @folio.
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700836 */
Matthew Wilcox (Oracle)4cd400f2022-09-02 20:46:03 +0100837static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700838{
Matthew Wilcox (Oracle)4cd400f2022-09-02 20:46:03 +0100839 struct address_space *mapping = folio->mapping;
840 long nr = folio_nr_pages(folio);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700841 int error;
842
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700843 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox (Oracle)4cd400f2022-09-02 20:46:03 +0100844 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
845 folio->mapping = NULL;
846 mapping->nrpages -= nr;
847 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
848 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700849 xa_unlock_irq(&mapping->i_pages);
Baolin Wang872339c2024-08-12 15:42:08 +0800850 folio_put_refs(folio, nr);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700851 BUG_ON(error);
852}
853
854/*
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +0800855 * Remove swap entry from page cache, free the swap and its page cache. Returns
856 * the number of pages being freed. 0 means entry not found in XArray (0 pages
857 * being freed).
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700858 */
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +0800859static long shmem_free_swap(struct address_space *mapping,
860 pgoff_t index, void *radswap)
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700861{
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +0800862 int order = xa_get_order(&mapping->i_pages, index);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700863 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700864
Matthew Wilcox55f3f7e2018-11-26 16:08:43 -0500865 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
Johannes Weiner6dbaf22c2014-04-03 14:47:41 -0700866 if (old != radswap)
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +0800867 return 0;
868 free_swap_and_cache_nr(radix_to_swp_entry(radswap), 1 << order);
869
870 return 1 << order;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700871}
872
873/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800874 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800875 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800876 *
Jan Kara96087032021-04-12 15:50:21 +0200877 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800878 * as long as the inode doesn't go away and racy results are not a problem.
879 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800880unsigned long shmem_partial_swap_usage(struct address_space *mapping,
881 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800882{
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500883 XA_STATE(xas, &mapping->i_pages, start);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800884 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800885 unsigned long swapped = 0;
Hugh Dickinse5548f82023-08-22 22:14:47 -0700886 unsigned long max = end - 1;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800887
888 rcu_read_lock();
Hugh Dickinse5548f82023-08-22 22:14:47 -0700889 xas_for_each(&xas, page, max) {
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500890 if (xas_retry(&xas, page))
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700891 continue;
Matthew Wilcox3159f942017-11-03 13:30:42 -0400892 if (xa_is_value(page))
Shakeel Butt354a5952024-09-06 16:05:12 -0700893 swapped += 1 << xas_get_order(&xas);
Hugh Dickinse5548f82023-08-22 22:14:47 -0700894 if (xas.xa_index == max)
895 break;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800896 if (need_resched()) {
Matthew Wilcox7ae34242017-12-04 03:28:00 -0500897 xas_pause(&xas);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800898 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800899 }
900 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800901 rcu_read_unlock();
902
903 return swapped << PAGE_SHIFT;
904}
905
906/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800907 * Determine (in bytes) how many of the shmem object's pages mapped by the
908 * given vma is swapped out.
909 *
Jan Kara96087032021-04-12 15:50:21 +0200910 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
Vlastimil Babka48131e02016-01-14 15:19:23 -0800911 * as long as the inode doesn't go away and racy results are not a problem.
912 */
913unsigned long shmem_swap_usage(struct vm_area_struct *vma)
914{
915 struct inode *inode = file_inode(vma->vm_file);
916 struct shmem_inode_info *info = SHMEM_I(inode);
917 struct address_space *mapping = inode->i_mapping;
918 unsigned long swapped;
919
920 /* Be careful as we don't hold info->lock */
921 swapped = READ_ONCE(info->swapped);
922
923 /*
924 * The easier cases are when the shmem object has nothing in swap, or
925 * the vma maps it whole. Then we can simply use the stats that we
926 * already track.
927 */
928 if (!swapped)
929 return 0;
930
931 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
932 return swapped << PAGE_SHIFT;
933
934 /* Here comes the more involved part */
Peter Xu02399c82021-11-05 13:36:02 -0700935 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
936 vma->vm_pgoff + vma_pages(vma));
Vlastimil Babka48131e02016-01-14 15:19:23 -0800937}
938
939/*
Hugh Dickins24513262012-01-20 14:34:21 -0800940 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
941 */
942void shmem_unlock_mapping(struct address_space *mapping)
943{
Matthew Wilcox (Oracle)105c9882022-06-04 17:40:17 -0400944 struct folio_batch fbatch;
Hugh Dickins24513262012-01-20 14:34:21 -0800945 pgoff_t index = 0;
946
Matthew Wilcox (Oracle)105c9882022-06-04 17:40:17 -0400947 folio_batch_init(&fbatch);
Hugh Dickins24513262012-01-20 14:34:21 -0800948 /*
949 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
950 */
Matthew Wilcox (Oracle)105c9882022-06-04 17:40:17 -0400951 while (!mapping_unevictable(mapping) &&
952 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
953 check_move_unevictable_folios(&fbatch);
954 folio_batch_release(&fbatch);
Hugh Dickins24513262012-01-20 14:34:21 -0800955 cond_resched();
956 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700957}
958
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400959static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
Hugh Dickins71725ed2020-04-06 20:07:57 -0700960{
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400961 struct folio *folio;
Hugh Dickins71725ed2020-04-06 20:07:57 -0700962
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400963 /*
Matthew Wilcox (Oracle)a7f58622022-09-02 20:46:21 +0100964 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
Hugh Dickins81914af2023-03-19 22:19:21 -0700965 * beyond i_size, and reports fallocated folios as holes.
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400966 */
Hugh Dickins81914af2023-03-19 22:19:21 -0700967 folio = filemap_get_entry(inode->i_mapping, index);
968 if (!folio)
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400969 return folio;
Hugh Dickins81914af2023-03-19 22:19:21 -0700970 if (!xa_is_value(folio)) {
971 folio_lock(folio);
972 if (folio->mapping == inode->i_mapping)
973 return folio;
974 /* The folio has been swapped out */
975 folio_unlock(folio);
976 folio_put(folio);
977 }
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400978 /*
Hugh Dickins81914af2023-03-19 22:19:21 -0700979 * But read a folio back from swap if any of it is within i_size
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -0400980 * (although in some cases this is just a waste of time).
981 */
Matthew Wilcox (Oracle)a7f58622022-09-02 20:46:21 +0100982 folio = NULL;
Rik van Riele1e4cfd2024-09-03 11:19:28 -0400983 shmem_get_folio(inode, index, 0, &folio, SGP_READ);
Matthew Wilcox (Oracle)a7f58622022-09-02 20:46:21 +0100984 return folio;
Hugh Dickins71725ed2020-04-06 20:07:57 -0700985}
986
987/*
Matthew Wilcox7f4446e2017-12-04 03:31:13 -0500988 * Remove range of pages and swap entries from page cache, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700989 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700990 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700991static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
992 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700994 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300996 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
997 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -0400998 struct folio_batch fbatch;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700999 pgoff_t indices[PAGEVEC_SIZE];
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001000 struct folio *folio;
1001 bool same_folio;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001002 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -07001003 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001004 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001006 if (lend == -1)
1007 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001008
Hugh Dickinsd144bf62021-09-02 14:54:21 -07001009 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
1010 info->fallocend = start;
1011
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -05001012 folio_batch_init(&fbatch);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001013 index = start;
Vishal Moola (Oracle)3392ca122022-10-17 09:17:59 -07001014 while (index < end && find_lock_entries(mapping, &index, end - 1,
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -05001015 &fbatch, indices)) {
1016 for (i = 0; i < folio_batch_count(&fbatch); i++) {
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001017 folio = fbatch.folios[i];
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001018
Matthew Wilcox (Oracle)7b774aa2021-12-03 08:50:01 -05001019 if (xa_is_value(folio)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001020 if (unfalloc)
1021 continue;
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +08001022 nr_swaps_freed += shmem_free_swap(mapping,
Vishal Moola (Oracle)3392ca122022-10-17 09:17:59 -07001023 indices[i], folio);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001024 continue;
1025 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001026
Matthew Wilcox (Oracle)7b774aa2021-12-03 08:50:01 -05001027 if (!unfalloc || !folio_test_uptodate(folio))
Matthew Wilcox (Oracle)1e84a3d2021-12-02 16:01:55 -05001028 truncate_inode_folio(mapping, folio);
Matthew Wilcox (Oracle)7b774aa2021-12-03 08:50:01 -05001029 folio_unlock(folio);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001030 }
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -05001031 folio_batch_remove_exceptionals(&fbatch);
1032 folio_batch_release(&fbatch);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001033 cond_resched();
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001034 }
1035
Hugh Dickins44bcabd2022-12-04 16:51:50 -08001036 /*
1037 * When undoing a failed fallocate, we want none of the partial folio
1038 * zeroing and splitting below, but shall want to truncate the whole
1039 * folio when !uptodate indicates that it was added by this fallocate,
1040 * even when [lstart, lend] covers only a part of the folio.
1041 */
1042 if (unfalloc)
1043 goto whole_folios;
1044
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001045 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1046 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1047 if (folio) {
1048 same_folio = lend < folio_pos(folio) + folio_size(folio);
1049 folio_mark_dirty(folio);
1050 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
Sidhartha Kumar87b11f82023-06-27 10:43:49 -07001051 start = folio_next_index(folio);
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001052 if (same_folio)
1053 end = folio->index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001054 }
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001055 folio_unlock(folio);
1056 folio_put(folio);
1057 folio = NULL;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001058 }
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001059
1060 if (!same_folio)
1061 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1062 if (folio) {
1063 folio_mark_dirty(folio);
1064 if (!truncate_inode_partial_folio(folio, lstart, lend))
1065 end = folio->index;
1066 folio_unlock(folio);
1067 folio_put(folio);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001068 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001069
Hugh Dickins44bcabd2022-12-04 16:51:50 -08001070whole_folios:
1071
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001072 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001073 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001074 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001075
Vishal Moola (Oracle)9fb6bee2022-10-17 09:18:00 -07001076 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
Matthew Wilcox (Oracle)cf2039a2021-02-25 17:16:11 -08001077 indices)) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001078 /* If all gone or hole-punch or unfalloc, we're done */
1079 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001080 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001081 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001082 index = start;
1083 continue;
1084 }
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001085 for (i = 0; i < folio_batch_count(&fbatch); i++) {
Matthew Wilcox (Oracle)b9a8a412020-05-27 17:59:22 -04001086 folio = fbatch.folios[i];
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001087
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001088 if (xa_is_value(folio)) {
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +08001089 long swaps_freed;
1090
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001091 if (unfalloc)
1092 continue;
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +08001093 swaps_freed = shmem_free_swap(mapping, indices[i], folio);
1094 if (!swaps_freed) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001095 /* Swap was replaced by page: retry */
Vishal Moola (Oracle)9fb6bee2022-10-17 09:18:00 -07001096 index = indices[i];
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001097 break;
1098 }
Daniel Gomez6ea0d1c2024-08-12 15:42:04 +08001099 nr_swaps_freed += swaps_freed;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001100 continue;
1101 }
1102
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001103 folio_lock(folio);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001104
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001105 if (!unfalloc || !folio_test_uptodate(folio)) {
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001106 if (folio_mapping(folio) != mapping) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001107 /* Page was replaced by swap: retry */
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001108 folio_unlock(folio);
Vishal Moola (Oracle)9fb6bee2022-10-17 09:18:00 -07001109 index = indices[i];
Hugh Dickinsb1a36652014-07-23 14:00:13 -07001110 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001111 }
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001112 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1113 folio);
David Stevens55ac8bb2023-04-18 17:40:31 +09001114
1115 if (!folio_test_large(folio)) {
1116 truncate_inode_folio(mapping, folio);
1117 } else if (truncate_inode_partial_folio(folio, lstart, lend)) {
1118 /*
1119 * If we split a page, reset the loop so
1120 * that we pick up the new sub pages.
1121 * Otherwise the THP was entirely
1122 * dropped or the target range was
1123 * zeroed, so just continue the loop as
1124 * is.
1125 */
1126 if (!folio_test_large(folio)) {
1127 folio_unlock(folio);
1128 index = start;
1129 break;
1130 }
1131 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -07001132 }
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001133 folio_unlock(folio);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001134 }
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04001135 folio_batch_remove_exceptionals(&fbatch);
1136 folio_batch_release(&fbatch);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -07001137 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001138
Hugh Dickins3c1b7522023-08-03 22:46:11 -07001139 shmem_recalc_inode(inode, 0, -nr_swaps_freed);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001140}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001142void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1143{
1144 shmem_undo_range(inode, lstart, lend, false);
Jeff Laytoncf2766b2023-10-04 14:53:07 -04001145 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
Jeff Layton36f05ca2022-09-09 09:00:31 -04001146 inode_inc_iversion(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147}
Hugh Dickins94c1e622011-06-27 16:18:03 -07001148EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Christian Braunerb74d24f2023-01-13 12:49:12 +01001150static int shmem_getattr(struct mnt_idmap *idmap,
Christian Brauner549c7292021-01-21 14:19:43 +01001151 const struct path *path, struct kstat *stat,
David Howellsa528d352017-01-31 16:46:22 +00001152 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -07001153{
David Howellsa528d352017-01-31 16:46:22 +00001154 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -07001155 struct shmem_inode_info *info = SHMEM_I(inode);
1156
Hugh Dickins3c1b7522023-08-03 22:46:11 -07001157 if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1158 shmem_recalc_inode(inode, 0, 0);
1159
Theodore Ts'oe408e692022-07-14 21:59:12 -04001160 if (info->fsflags & FS_APPEND_FL)
1161 stat->attributes |= STATX_ATTR_APPEND;
1162 if (info->fsflags & FS_IMMUTABLE_FL)
1163 stat->attributes |= STATX_ATTR_IMMUTABLE;
1164 if (info->fsflags & FS_NODUMP_FL)
1165 stat->attributes |= STATX_ATTR_NODUMP;
1166 stat->attributes_mask |= (STATX_ATTR_APPEND |
1167 STATX_ATTR_IMMUTABLE |
1168 STATX_ATTR_NODUMP);
Jeff Layton0d72b9282023-08-07 15:38:33 -04001169 generic_fillattr(idmap, request_mask, inode, stat);
Yang Shi89fdcd22018-06-07 17:06:59 -07001170
Rik van Riele1e4cfd2024-09-03 11:19:28 -04001171 if (shmem_huge_global_enabled(inode, 0, 0, false, NULL, 0))
Yang Shi89fdcd22018-06-07 17:06:59 -07001172 stat->blksize = HPAGE_PMD_SIZE;
1173
Xavier Rochef7cd16a2022-03-22 14:39:55 -07001174 if (request_mask & STATX_BTIME) {
1175 stat->result_mask |= STATX_BTIME;
1176 stat->btime.tv_sec = info->i_crtime.tv_sec;
1177 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1178 }
1179
Yu Zhao44a30222015-09-08 15:03:33 -07001180 return 0;
1181}
1182
Christian Braunerc1632a02023-01-13 12:49:11 +01001183static int shmem_setattr(struct mnt_idmap *idmap,
Christian Brauner549c7292021-01-21 14:19:43 +01001184 struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185{
David Howells75c3cfa2015-03-17 22:26:12 +00001186 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -07001187 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 int error;
Jeff Layton36f05ca2022-09-09 09:00:31 -04001189 bool update_mtime = false;
1190 bool update_ctime = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01001192 error = setattr_prepare(idmap, dentry, attr);
Christoph Hellwigdb78b8772010-06-04 11:30:03 +02001193 if (error)
1194 return error;
1195
Daniel Verkamp6fd73532022-12-15 00:12:01 +00001196 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1197 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1198 return -EPERM;
1199 }
1200 }
1201
Hugh Dickins94c1e622011-06-27 16:18:03 -07001202 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1203 loff_t oldsize = inode->i_size;
1204 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001205
Jan Kara96087032021-04-12 15:50:21 +02001206 /* protected by i_rwsem */
David Herrmann40e041a2014-08-08 14:25:27 -07001207 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1208 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1209 return -EPERM;
1210
Hugh Dickins94c1e622011-06-27 16:18:03 -07001211 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -07001212 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1213 oldsize, newsize);
1214 if (error)
1215 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001216 i_size_write(inode, newsize);
Jeff Layton36f05ca2022-09-09 09:00:31 -04001217 update_mtime = true;
1218 } else {
1219 update_ctime = false;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001220 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001221 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001222 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001223 if (oldsize > holebegin)
1224 unmap_mapping_range(inode->i_mapping,
1225 holebegin, 0, 1);
1226 if (info->alloced)
1227 shmem_truncate_range(inode,
1228 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001229 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001230 if (oldsize > holebegin)
1231 unmap_mapping_range(inode->i_mapping,
1232 holebegin, 0, 1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001233 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
1235
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02001236 if (is_quota_modification(idmap, inode, attr)) {
1237 error = dquot_initialize(inode);
1238 if (error)
1239 return error;
1240 }
1241
1242 /* Transfer quota accounting */
1243 if (i_uid_needs_update(idmap, attr, inode) ||
1244 i_gid_needs_update(idmap, attr, inode)) {
1245 error = dquot_transfer(idmap, inode, attr);
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02001246 if (error)
1247 return error;
1248 }
1249
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01001250 setattr_copy(idmap, inode, attr);
Christoph Hellwigdb78b8772010-06-04 11:30:03 +02001251 if (attr->ia_valid & ATTR_MODE)
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01001252 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
Jeff Layton36f05ca2022-09-09 09:00:31 -04001253 if (!error && update_ctime) {
Jeff Layton65287332023-07-05 15:01:52 -04001254 inode_set_ctime_current(inode);
Jeff Layton36f05ca2022-09-09 09:00:31 -04001255 if (update_mtime)
Jeff Laytoncf2766b2023-10-04 14:53:07 -04001256 inode_set_mtime_to_ts(inode, inode_get_ctime(inode));
Jeff Layton36f05ca2022-09-09 09:00:31 -04001257 inode_inc_iversion(inode);
1258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return error;
1260}
1261
Al Viro1f895f72010-06-05 19:10:41 -04001262static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001265 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins2daf18a2023-08-08 21:33:56 -07001266 size_t freed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Hui Su30e6a512020-12-14 19:06:14 -08001268 if (shmem_mapping(inode->i_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 shmem_unacct_size(info->flags, inode->i_size);
1270 inode->i_size = 0;
Hugh Dickinsbc786392022-03-22 14:39:58 -07001271 mapping_set_exiting(inode->i_mapping);
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001272 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001273 if (!list_empty(&info->shrinklist)) {
1274 spin_lock(&sbinfo->shrinklist_lock);
1275 if (!list_empty(&info->shrinklist)) {
1276 list_del_init(&info->shrinklist);
1277 sbinfo->shrinklist_len--;
1278 }
1279 spin_unlock(&sbinfo->shrinklist_lock);
1280 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001281 while (!list_empty(&info->swaplist)) {
1282 /* Wait while shmem_unuse() is scanning this inode... */
1283 wait_var_event(&info->stop_eviction,
1284 !atomic_read(&info->stop_eviction));
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001285 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001286 /* ...but beware of the race if we peeked too early */
1287 if (!atomic_read(&info->stop_eviction))
1288 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001289 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 }
Al Viro3ed47db2016-01-22 18:08:52 -05001291 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001292
Hugh Dickins2daf18a2023-08-08 21:33:56 -07001293 simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL);
1294 shmem_free_inode(inode->i_sb, freed);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001295 WARN_ON(inode->i_blocks);
Jan Karadbd57682012-05-03 14:48:02 +02001296 clear_inode(inode);
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02001297#ifdef CONFIG_TMPFS_QUOTA
1298 dquot_free_inode(inode);
1299 dquot_drop(inode);
1300#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301}
1302
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001303static int shmem_find_swap_entries(struct address_space *mapping,
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001304 pgoff_t start, struct folio_batch *fbatch,
1305 pgoff_t *indices, unsigned int type)
Matthew Wilcox478922e2016-12-14 15:08:52 -08001306{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001307 XA_STATE(xas, &mapping->i_pages, start);
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001308 struct folio *folio;
Hugh Dickins87039542019-04-18 17:49:58 -07001309 swp_entry_t entry;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001310
1311 rcu_read_lock();
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001312 xas_for_each(&xas, folio, ULONG_MAX) {
1313 if (xas_retry(&xas, folio))
Mike Kravetz5b9c98f2018-06-07 17:05:53 -07001314 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001315
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001316 if (!xa_is_value(folio))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001317 continue;
1318
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001319 entry = radix_to_swp_entry(folio);
Miaohe Lin6cec2b92022-05-19 20:50:29 +08001320 /*
1321 * swapin error entries can be found in the mapping. But they're
1322 * deliberately ignored here as we've done everything we can do.
1323 */
Hugh Dickins87039542019-04-18 17:49:58 -07001324 if (swp_type(entry) != type)
1325 continue;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001326
Hugh Dickinse3842002022-05-21 19:53:04 -07001327 indices[folio_batch_count(fbatch)] = xas.xa_index;
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001328 if (!folio_batch_add(fbatch, folio))
1329 break;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001330
1331 if (need_resched()) {
1332 xas_pause(&xas);
1333 cond_resched_rcu();
1334 }
Matthew Wilcox478922e2016-12-14 15:08:52 -08001335 }
Matthew Wilcox478922e2016-12-14 15:08:52 -08001336 rcu_read_unlock();
Matthew Wilcoxe21a2952017-11-22 08:36:00 -05001337
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001338 return xas.xa_index;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001339}
1340
1341/*
1342 * Move the swapped pages for an inode to page cache. Returns the count
1343 * of pages swapped in, or the error in case of failure.
1344 */
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001345static int shmem_unuse_swap_entries(struct inode *inode,
1346 struct folio_batch *fbatch, pgoff_t *indices)
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001347{
1348 int i = 0;
1349 int ret = 0;
1350 int error = 0;
1351 struct address_space *mapping = inode->i_mapping;
1352
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001353 for (i = 0; i < folio_batch_count(fbatch); i++) {
1354 struct folio *folio = fbatch->folios[i];
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001355
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001356 if (!xa_is_value(folio))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001357 continue;
Hugh Dickins054a9f72023-09-29 20:31:27 -07001358 error = shmem_swapin_folio(inode, indices[i], &folio, SGP_CACHE,
1359 mapping_gfp_mask(mapping), NULL, NULL);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001360 if (error == 0) {
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001361 folio_unlock(folio);
1362 folio_put(folio);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001363 ret++;
1364 }
1365 if (error == -ENOMEM)
1366 break;
1367 error = 0;
1368 }
1369 return error ? error : ret;
Matthew Wilcox478922e2016-12-14 15:08:52 -08001370}
1371
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001372/*
1373 * If swap found in inode, free it and move page from swapcache to filecache.
1374 */
Christoph Hellwig10a9c492022-01-21 22:14:57 -08001375static int shmem_unuse_inode(struct inode *inode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001377 struct address_space *mapping = inode->i_mapping;
1378 pgoff_t start = 0;
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001379 struct folio_batch fbatch;
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001380 pgoff_t indices[PAGEVEC_SIZE];
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001381 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001383 do {
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001384 folio_batch_init(&fbatch);
1385 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1386 if (folio_batch_count(&fbatch) == 0) {
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001387 ret = 0;
1388 break;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001389 }
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001390
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001391 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001392 if (ret < 0)
1393 break;
1394
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07001395 start = indices[folio_batch_count(&fbatch) - 1];
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001396 } while (true);
1397
1398 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
1400
1401/*
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001402 * Read all the shared memory data that resides in the swap
1403 * device 'type' back into memory, so the swap device can be
1404 * unused.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 */
Christoph Hellwig10a9c492022-01-21 22:14:57 -08001406int shmem_unuse(unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001408 struct shmem_inode_info *info, *next;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001409 int error = 0;
1410
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001411 if (list_empty(&shmem_swaplist))
1412 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001414 mutex_lock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001415 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1416 if (!info->swapped) {
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001417 list_del_init(&info->swaplist);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001418 continue;
1419 }
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001420 /*
1421 * Drop the swaplist mutex while searching the inode for swap;
1422 * but before doing so, make sure shmem_evict_inode() will not
1423 * remove placeholder inode from swaplist, nor let it be freed
1424 * (igrab() would protect from unlink, but not from unmount).
1425 */
1426 atomic_inc(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001427 mutex_unlock(&shmem_swaplist_mutex);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001428
Christoph Hellwig10a9c492022-01-21 22:14:57 -08001429 error = shmem_unuse_inode(&info->vfs_inode, type);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001430 cond_resched();
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001431
1432 mutex_lock(&shmem_swaplist_mutex);
1433 next = list_next_entry(info, swaplist);
1434 if (!info->swapped)
1435 list_del_init(&info->swaplist);
Hugh Dickinsaf53d3e2019-04-18 17:50:13 -07001436 if (atomic_dec_and_test(&info->stop_eviction))
1437 wake_up_var(&info->stop_eviction);
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001438 if (error)
Hugh Dickins778dd892011-05-11 15:13:37 -07001439 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001441 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001442
Hugh Dickins778dd892011-05-11 15:13:37 -07001443 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444}
1445
1446/*
1447 * Move the page from the page cache to the swap cache.
1448 */
1449static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1450{
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07001451 struct folio *folio = page_folio(page);
Luis Chamberlain8ccee8c2023-03-09 15:05:41 -08001452 struct address_space *mapping = folio->mapping;
1453 struct inode *inode = mapping->host;
1454 struct shmem_inode_info *info = SHMEM_I(inode);
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08001455 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001456 swp_entry_t swap;
1457 pgoff_t index;
Baolin Wang65018072024-08-12 15:42:02 +08001458 int nr_pages;
Baolin Wang809bc8652024-08-12 15:42:10 +08001459 bool split = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Hugh Dickins1e6decf2021-09-02 14:54:43 -07001461 /*
Luis Chamberlaincf7992b2023-03-09 15:05:42 -08001462 * Our capabilities prevent regular writeback or sync from ever calling
1463 * shmem_writepage; but a stacking filesystem might use ->writepage of
1464 * its underlying filesystem, in which case tmpfs should write out to
1465 * swap only in response to memory pressure, and not for the writeback
1466 * threads or sync.
1467 */
1468 if (WARN_ON_ONCE(!wbc->for_reclaim))
1469 goto redirty;
1470
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08001471 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
Luis Chamberlain9a976f02023-03-09 15:05:43 -08001472 goto redirty;
1473
1474 if (!total_swap_pages)
1475 goto redirty;
1476
Luis Chamberlaincf7992b2023-03-09 15:05:42 -08001477 /*
Baolin Wang809bc8652024-08-12 15:42:10 +08001478 * If CONFIG_THP_SWAP is not enabled, the large folio should be
1479 * split when swapping.
1480 *
1481 * And shrinkage of pages beyond i_size does not split swap, so
1482 * swapout of a large folio crossing i_size needs to split too
1483 * (unless fallocate has been used to preallocate beyond EOF).
Hugh Dickins1e6decf2021-09-02 14:54:43 -07001484 */
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001485 if (folio_test_large(folio)) {
Baolin Wang809bc8652024-08-12 15:42:10 +08001486 index = shmem_fallocend(inode,
1487 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE));
1488 if ((index > folio->index && index < folio_next_index(folio)) ||
1489 !IS_ENABLED(CONFIG_THP_SWAP))
1490 split = true;
1491 }
1492
1493 if (split) {
1494try_split:
Hugh Dickins1e6decf2021-09-02 14:54:43 -07001495 /* Ensure the subpages are still dirty */
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001496 folio_test_set_dirty(folio);
Baolin Wang809bc8652024-08-12 15:42:10 +08001497 if (split_huge_page_to_list_to_order(page, wbc->list, 0))
Hugh Dickins1e6decf2021-09-02 14:54:43 -07001498 goto redirty;
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001499 folio = page_folio(page);
1500 folio_clear_dirty(folio);
Hugh Dickins1e6decf2021-09-02 14:54:43 -07001501 }
1502
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001503 index = folio->index;
Baolin Wang65018072024-08-12 15:42:02 +08001504 nr_pages = folio_nr_pages(folio);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001505
1506 /*
1507 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1508 * value into swapfile.c, the only way we can correctly account for a
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001509 * fallocated folio arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001510 *
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001511 * That's okay for a folio already fallocated earlier, but if we have
Hugh Dickins1aac1402012-05-29 15:06:42 -07001512 * not yet completed the fallocation, then (a) we want to keep track
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001513 * of this folio in case we have to undo it, and (b) it may not be a
Hugh Dickins1aac1402012-05-29 15:06:42 -07001514 * good idea to continue anyway, once we're pushing into swap. So
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001515 * reactivate the folio, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001516 */
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001517 if (!folio_test_uptodate(folio)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001518 if (inode->i_private) {
1519 struct shmem_falloc *shmem_falloc;
1520 spin_lock(&inode->i_lock);
1521 shmem_falloc = inode->i_private;
1522 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001523 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001524 index >= shmem_falloc->start &&
1525 index < shmem_falloc->next)
1526 shmem_falloc->nr_unswapped++;
1527 else
1528 shmem_falloc = NULL;
1529 spin_unlock(&inode->i_lock);
1530 if (shmem_falloc)
1531 goto redirty;
1532 }
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001533 folio_zero_range(folio, 0, folio_size(folio));
1534 flush_dcache_folio(folio);
1535 folio_mark_uptodate(folio);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001536 }
1537
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07001538 swap = folio_alloc_swap(folio);
Baolin Wang809bc8652024-08-12 15:42:10 +08001539 if (!swap.val) {
1540 if (nr_pages > 1)
1541 goto try_split;
1542
Hugh Dickins48f170f2011-07-25 17:12:37 -07001543 goto redirty;
Baolin Wang809bc8652024-08-12 15:42:10 +08001544 }
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001545
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001546 /*
1547 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001548 * if it's not already there. Do it now before the folio is
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001549 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001550 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001551 * we've incremented swapped, because shmem_unuse_inode() will
1552 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001553 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001554 mutex_lock(&shmem_swaplist_mutex);
1555 if (list_empty(&info->swaplist))
Vineeth Remanan Pillaib56a2d82019-03-05 15:47:03 -08001556 list_add(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001557
Matthew Wilcox (Oracle)a4c366f2022-09-02 20:46:08 +01001558 if (add_to_swap_cache(folio, swap,
Joonsoo Kim3852f672020-08-11 18:30:47 -07001559 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1560 NULL) == 0) {
Baolin Wang65018072024-08-12 15:42:02 +08001561 shmem_recalc_inode(inode, 0, nr_pages);
1562 swap_shmem_alloc(swap, nr_pages);
Matthew Wilcox (Oracle)4cd400f2022-09-02 20:46:03 +01001563 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001564
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001565 mutex_unlock(&shmem_swaplist_mutex);
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001566 BUG_ON(folio_mapped(folio));
Nhat Pham501a06f2023-12-07 11:24:06 -08001567 return swap_writepage(&folio->page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 }
1569
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001570 mutex_unlock(&shmem_swaplist_mutex);
Matthew Wilcox (Oracle)4081f742022-09-02 20:46:09 +01001571 put_swap_folio(folio, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572redirty:
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001573 folio_mark_dirty(folio);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001574 if (wbc->for_reclaim)
Matthew Wilcox (Oracle)f530ed02022-09-02 20:46:02 +01001575 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1576 folio_unlock(folio);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001577 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578}
1579
Hugh Dickins75edd342016-05-19 17:12:44 -07001580#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001581static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001582{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001583 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001584
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001585 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001586 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001587
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001588 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001589
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001590 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001591}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001592
1593static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1594{
1595 struct mempolicy *mpol = NULL;
1596 if (sbinfo->mpol) {
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07001597 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001598 mpol = sbinfo->mpol;
1599 mpol_get(mpol);
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07001600 raw_spin_unlock(&sbinfo->stat_lock);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001601 }
1602 return mpol;
1603}
Hugh Dickins75edd342016-05-19 17:12:44 -07001604#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1605static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1606{
1607}
1608static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1609{
1610 return NULL;
1611}
1612#endif /* CONFIG_NUMA && CONFIG_TMPFS */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001613
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001614static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
1615 pgoff_t index, unsigned int order, pgoff_t *ilx);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001616
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001617static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001618 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001620 struct mempolicy *mpol;
1621 pgoff_t ilx;
Matthew Wilcox (Oracle)a4575c42023-12-13 21:58:42 +00001622 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001624 mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
Matthew Wilcox (Oracle)a4575c42023-12-13 21:58:42 +00001625 folio = swap_cluster_readahead(swap, gfp, mpol, ilx);
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001626 mpol_cond_put(mpol);
Mel Gorman18a2f372012-12-05 14:01:41 -08001627
Matthew Wilcox (Oracle)a4575c42023-12-13 21:58:42 +00001628 return folio;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001629}
Mel Gorman18a2f372012-12-05 14:01:41 -08001630
Rik van Riel78cc8cd2021-02-25 17:16:22 -08001631/*
1632 * Make sure huge_gfp is always more limited than limit_gfp.
1633 * Some of the flags set permissions, while others set limitations.
1634 */
1635static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1636{
1637 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1638 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
Rik van Riel187df5d2021-02-25 17:16:29 -08001639 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1640 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1641
1642 /* Allow allocations only from the originally specified zones. */
1643 result |= zoneflags;
Rik van Riel78cc8cd2021-02-25 17:16:22 -08001644
1645 /*
1646 * Minimize the result gfp by taking the union with the deny flags,
1647 * and the intersection of the allow flags.
1648 */
1649 result |= (limit_gfp & denyflags);
1650 result |= (huge_gfp & limit_gfp) & allowflags;
1651
1652 return result;
1653}
1654
Baolin Wange7a2ab72024-06-11 18:11:08 +08001655#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Bang Li26c7d842024-07-05 11:23:09 +08001656unsigned long shmem_allowable_huge_orders(struct inode *inode,
Baolin Wange7a2ab72024-06-11 18:11:08 +08001657 struct vm_area_struct *vma, pgoff_t index,
Rik van Riele1e4cfd2024-09-03 11:19:28 -04001658 loff_t write_end, bool shmem_huge_force)
Baolin Wange7a2ab72024-06-11 18:11:08 +08001659{
1660 unsigned long mask = READ_ONCE(huge_shmem_orders_always);
1661 unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
Baolin Wang6beeab82024-07-22 13:43:19 +08001662 unsigned long vm_flags = vma ? vma->vm_flags : 0;
1663 bool global_huge;
Baolin Wange7a2ab72024-06-11 18:11:08 +08001664 loff_t i_size;
1665 int order;
1666
Baolin Wang6beeab82024-07-22 13:43:19 +08001667 if (vma && ((vm_flags & VM_NOHUGEPAGE) ||
1668 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
Baolin Wange7a2ab72024-06-11 18:11:08 +08001669 return 0;
1670
1671 /* If the hardware/firmware marked hugepage support disabled. */
1672 if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED))
1673 return 0;
1674
Rik van Riele1e4cfd2024-09-03 11:19:28 -04001675 global_huge = shmem_huge_global_enabled(inode, index, write_end,
1676 shmem_huge_force, vma, vm_flags);
Baolin Wang6beeab82024-07-22 13:43:19 +08001677 if (!vma || !vma_is_anon_shmem(vma)) {
1678 /*
1679 * For tmpfs, we now only support PMD sized THP if huge page
1680 * is enabled, otherwise fallback to order 0.
1681 */
1682 return global_huge ? BIT(HPAGE_PMD_ORDER) : 0;
1683 }
1684
Baolin Wange7a2ab72024-06-11 18:11:08 +08001685 /*
1686 * Following the 'deny' semantics of the top level, force the huge
1687 * option off from all mounts.
1688 */
1689 if (shmem_huge == SHMEM_HUGE_DENY)
1690 return 0;
1691
1692 /*
1693 * Only allow inherit orders if the top-level value is 'force', which
1694 * means non-PMD sized THP can not override 'huge' mount option now.
1695 */
1696 if (shmem_huge == SHMEM_HUGE_FORCE)
1697 return READ_ONCE(huge_shmem_orders_inherit);
1698
1699 /* Allow mTHP that will be fully within i_size. */
1700 order = highest_order(within_size_orders);
1701 while (within_size_orders) {
1702 index = round_up(index + 1, order);
1703 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1704 if (i_size >> PAGE_SHIFT >= index) {
1705 mask |= within_size_orders;
1706 break;
1707 }
1708
1709 order = next_order(&within_size_orders, order);
1710 }
1711
1712 if (vm_flags & VM_HUGEPAGE)
1713 mask |= READ_ONCE(huge_shmem_orders_madvise);
1714
1715 if (global_huge)
1716 mask |= READ_ONCE(huge_shmem_orders_inherit);
1717
Baolin Wangb66b1b72024-07-31 13:46:19 +08001718 return THP_ORDERS_ALL_FILE_DEFAULT & mask;
Baolin Wange7a2ab72024-06-11 18:11:08 +08001719}
1720
1721static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
1722 struct address_space *mapping, pgoff_t index,
1723 unsigned long orders)
1724{
Baolin Wang0bedf0012024-07-22 13:43:17 +08001725 struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
Baolin Wang4cbf3202024-07-31 13:46:20 +08001726 pgoff_t aligned_index;
Baolin Wange7a2ab72024-06-11 18:11:08 +08001727 unsigned long pages;
1728 int order;
1729
Baolin Wang0bedf0012024-07-22 13:43:17 +08001730 if (vma) {
1731 orders = thp_vma_suitable_orders(vma, vmf->address, orders);
1732 if (!orders)
1733 return 0;
1734 }
Baolin Wange7a2ab72024-06-11 18:11:08 +08001735
1736 /* Find the highest order that can add into the page cache */
1737 order = highest_order(orders);
1738 while (orders) {
1739 pages = 1UL << order;
Baolin Wang4cbf3202024-07-31 13:46:20 +08001740 aligned_index = round_down(index, pages);
Baolin Wang0bedf0012024-07-22 13:43:17 +08001741 /*
1742 * Check for conflict before waiting on a huge allocation.
1743 * Conflict might be that a huge page has just been allocated
1744 * and added to page cache by a racing thread, or that there
1745 * is already at least one small page in the huge extent.
1746 * Be careful to retry when appropriate, but not forever!
1747 * Elsewhere -EEXIST would be the right code, but not here.
1748 */
Baolin Wang4cbf3202024-07-31 13:46:20 +08001749 if (!xa_find(&mapping->i_pages, &aligned_index,
1750 aligned_index + pages - 1, XA_PRESENT))
Baolin Wange7a2ab72024-06-11 18:11:08 +08001751 break;
1752 order = next_order(&orders, order);
1753 }
1754
1755 return orders;
1756}
1757#else
Baolin Wange7a2ab72024-06-11 18:11:08 +08001758static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
1759 struct address_space *mapping, pgoff_t index,
1760 unsigned long orders)
1761{
1762 return 0;
1763}
1764#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1765
Kefeng Wang6f775462024-05-15 15:07:09 +08001766static struct folio *shmem_alloc_folio(gfp_t gfp, int order,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001767 struct shmem_inode_info *info, pgoff_t index)
1768{
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001769 struct mempolicy *mpol;
1770 pgoff_t ilx;
Kefeng Wang6f775462024-05-15 15:07:09 +08001771 struct folio *folio;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001772
Kefeng Wang6f775462024-05-15 15:07:09 +08001773 mpol = shmem_get_pgoff_policy(info, index, order, &ilx);
1774 folio = folio_alloc_mpol(gfp, order, mpol, ilx, numa_node_id());
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07001775 mpol_cond_put(mpol);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001776
Kefeng Wang6f775462024-05-15 15:07:09 +08001777 return folio;
Mel Gorman18a2f372012-12-05 14:01:41 -08001778}
1779
Baolin Wange7a2ab72024-06-11 18:11:08 +08001780static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
1781 gfp_t gfp, struct inode *inode, pgoff_t index,
1782 struct mm_struct *fault_mm, unsigned long orders)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001783{
Hugh Dickins3022fd72023-09-29 20:32:40 -07001784 struct address_space *mapping = inode->i_mapping;
Mike Rapoport0f079692017-09-06 16:22:59 -07001785 struct shmem_inode_info *info = SHMEM_I(inode);
Baolin Wange7a2ab72024-06-11 18:11:08 +08001786 unsigned long suitable_orders = 0;
1787 struct folio *folio = NULL;
Hugh Dickins3022fd72023-09-29 20:32:40 -07001788 long pages;
Baolin Wange7a2ab72024-06-11 18:11:08 +08001789 int error, order;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001790
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07001791 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Baolin Wange7a2ab72024-06-11 18:11:08 +08001792 orders = 0;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001793
Baolin Wange7a2ab72024-06-11 18:11:08 +08001794 if (orders > 0) {
Baolin Wang0bedf0012024-07-22 13:43:17 +08001795 suitable_orders = shmem_suitable_orders(inode, vmf,
Baolin Wange7a2ab72024-06-11 18:11:08 +08001796 mapping, index, orders);
Hugh Dickins3022fd72023-09-29 20:32:40 -07001797
Baolin Wange7a2ab72024-06-11 18:11:08 +08001798 order = highest_order(suitable_orders);
1799 while (suitable_orders) {
1800 pages = 1UL << order;
1801 index = round_down(index, pages);
1802 folio = shmem_alloc_folio(gfp, order, info, index);
1803 if (folio)
1804 goto allocated;
1805
1806 if (pages == HPAGE_PMD_NR)
1807 count_vm_event(THP_FILE_FALLBACK);
Ryan Roberts63d98662024-07-10 10:55:01 +01001808 count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK);
Baolin Wange7a2ab72024-06-11 18:11:08 +08001809 order = next_order(&suitable_orders, order);
1810 }
Hugh Dickins3022fd72023-09-29 20:32:40 -07001811 } else {
1812 pages = 1;
Kefeng Wang6f775462024-05-15 15:07:09 +08001813 folio = shmem_alloc_folio(gfp, 0, info, index);
Hugh Dickins3022fd72023-09-29 20:32:40 -07001814 }
1815 if (!folio)
1816 return ERR_PTR(-ENOMEM);
1817
Baolin Wange7a2ab72024-06-11 18:11:08 +08001818allocated:
Hugh Dickins3022fd72023-09-29 20:32:40 -07001819 __folio_set_locked(folio);
1820 __folio_set_swapbacked(folio);
1821
1822 gfp &= GFP_RECLAIM_MASK;
1823 error = mem_cgroup_charge(folio, fault_mm, gfp);
1824 if (error) {
1825 if (xa_find(&mapping->i_pages, &index,
1826 index + pages - 1, XA_PRESENT)) {
1827 error = -EEXIST;
Baolin Wang66f44582024-06-11 18:11:10 +08001828 } else if (pages > 1) {
1829 if (pages == HPAGE_PMD_NR) {
1830 count_vm_event(THP_FILE_FALLBACK);
1831 count_vm_event(THP_FILE_FALLBACK_CHARGE);
1832 }
Ryan Roberts63d98662024-07-10 10:55:01 +01001833 count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK);
1834 count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE);
Hugh Dickins3022fd72023-09-29 20:32:40 -07001835 }
1836 goto unlock;
Hugh Dickins75edd342016-05-19 17:12:44 -07001837 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001838
Hugh Dickins3022fd72023-09-29 20:32:40 -07001839 error = shmem_add_to_page_cache(folio, mapping, index, NULL, gfp);
1840 if (error)
1841 goto unlock;
1842
1843 error = shmem_inode_acct_blocks(inode, pages);
1844 if (error) {
1845 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1846 long freed;
1847 /*
1848 * Try to reclaim some space by splitting a few
1849 * large folios beyond i_size on the filesystem.
1850 */
Hugh Dickins15444052024-08-25 16:25:39 -07001851 shmem_unused_huge_shrink(sbinfo, NULL, pages);
Hugh Dickins3022fd72023-09-29 20:32:40 -07001852 /*
1853 * And do a shmem_recalc_inode() to account for freed pages:
1854 * except our folio is there in cache, so not quite balanced.
1855 */
1856 spin_lock(&info->lock);
1857 freed = pages + info->alloced - info->swapped -
1858 READ_ONCE(mapping->nrpages);
1859 if (freed > 0)
1860 info->alloced -= freed;
1861 spin_unlock(&info->lock);
1862 if (freed > 0)
1863 shmem_inode_unacct_blocks(inode, freed);
1864 error = shmem_inode_acct_blocks(inode, pages);
1865 if (error) {
1866 filemap_remove_folio(folio);
1867 goto unlock;
1868 }
1869 }
1870
1871 shmem_recalc_inode(inode, pages, 0);
1872 folio_add_lru(folio);
1873 return folio;
1874
1875unlock:
1876 folio_unlock(folio);
1877 folio_put(folio);
1878 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001882 * When a page is moved from swapcache to shmem filecache (either by the
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01001883 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001884 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1885 * ignorance of the mapping it belongs to. If that mapping has special
1886 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1887 * we may need to copy to a suitable page before moving to filecache.
1888 *
1889 * In a future release, this may well be extended to respect cpuset and
1890 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1891 * but for now it is a simple matter of zone.
1892 */
Matthew Wilcox (Oracle)069d8492022-05-12 20:23:04 -07001893static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001894{
Matthew Wilcox (Oracle)069d8492022-05-12 20:23:04 -07001895 return folio_zonenum(folio) > gfp_zone(gfp);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001896}
1897
Matthew Wilcox (Oracle)0d698e22022-09-02 20:46:14 +01001898static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
Baolin Wang736f0e02024-08-12 15:42:07 +08001899 struct shmem_inode_info *info, pgoff_t index,
1900 struct vm_area_struct *vma)
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001901{
Baolin Wang736f0e02024-08-12 15:42:07 +08001902 struct folio *new, *old = *foliop;
1903 swp_entry_t entry = old->swap;
1904 struct address_space *swap_mapping = swap_address_space(entry);
1905 pgoff_t swap_index = swap_cache_index(entry);
1906 XA_STATE(xas, &swap_mapping->i_pages, swap_index);
1907 int nr_pages = folio_nr_pages(old);
1908 int error = 0, i;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001909
1910 /*
1911 * We have arrived here because our zones are constrained, so don't
1912 * limit chance of success by further cpuset and node constraints.
1913 */
1914 gfp &= ~GFP_CONSTRAINT_MASK;
Baolin Wang736f0e02024-08-12 15:42:07 +08001915#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1916 if (nr_pages > 1) {
1917 gfp_t huge_gfp = vma_thp_gfp_mask(vma);
1918
1919 gfp = limit_gfp_mask(huge_gfp, gfp);
1920 }
1921#endif
1922
1923 new = shmem_alloc_folio(gfp, folio_order(old), info, index);
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001924 if (!new)
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001925 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001926
Baolin Wang736f0e02024-08-12 15:42:07 +08001927 folio_ref_add(new, nr_pages);
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001928 folio_copy(new, old);
1929 flush_dcache_folio(new);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001930
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001931 __folio_set_locked(new);
1932 __folio_set_swapbacked(new);
1933 folio_mark_uptodate(new);
David Hildenbrand3d2c9082023-08-21 18:08:48 +02001934 new->swap = entry;
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001935 folio_set_swapcache(new);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001936
Baolin Wang736f0e02024-08-12 15:42:07 +08001937 /* Swap cache still stores N entries instead of a high-order entry */
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001938 xa_lock_irq(&swap_mapping->i_pages);
Baolin Wang736f0e02024-08-12 15:42:07 +08001939 for (i = 0; i < nr_pages; i++) {
1940 void *item = xas_load(&xas);
1941
1942 if (item != old) {
1943 error = -ENOENT;
1944 break;
1945 }
1946
1947 xas_store(&xas, new);
1948 xas_next(&xas);
1949 }
Hugh Dickins0142ef62012-06-07 14:21:09 -07001950 if (!error) {
Baolin Wang9094b4a2024-06-13 16:21:19 +08001951 mem_cgroup_replace_folio(old, new);
Baolin Wang736f0e02024-08-12 15:42:07 +08001952 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, nr_pages);
1953 __lruvec_stat_mod_folio(new, NR_SHMEM, nr_pages);
1954 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -nr_pages);
1955 __lruvec_stat_mod_folio(old, NR_SHMEM, -nr_pages);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001956 }
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001957 xa_unlock_irq(&swap_mapping->i_pages);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001958
Hugh Dickins0142ef62012-06-07 14:21:09 -07001959 if (unlikely(error)) {
1960 /*
Matthew Wilcox (Oracle)32f51ea2024-08-21 20:34:37 +01001961 * Is this possible? I think not, now that our callers
1962 * check both the swapcache flag and folio->private
1963 * after getting the folio lock; but be defensive.
1964 * Reverse old to newpage for clear and free.
Hugh Dickins0142ef62012-06-07 14:21:09 -07001965 */
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001966 old = new;
Hugh Dickins0142ef62012-06-07 14:21:09 -07001967 } else {
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001968 folio_add_lru(new);
Matthew Wilcox (Oracle)0d698e22022-09-02 20:46:14 +01001969 *foliop = new;
Hugh Dickins0142ef62012-06-07 14:21:09 -07001970 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001971
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001972 folio_clear_swapcache(old);
1973 old->private = NULL;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001974
Matthew Wilcox (Oracle)907ea172022-09-02 20:46:04 +01001975 folio_unlock(old);
Baolin Wang736f0e02024-08-12 15:42:07 +08001976 /*
1977 * The old folio are removed from swap cache, drop the 'nr_pages'
1978 * reference, as well as one temporary reference getting from swap
1979 * cache.
1980 */
1981 folio_put_refs(old, nr_pages + 1);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001982 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001983}
1984
Miaohe Lin6cec2b92022-05-19 20:50:29 +08001985static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1986 struct folio *folio, swp_entry_t swap)
1987{
1988 struct address_space *mapping = inode->i_mapping;
Miaohe Lin6cec2b92022-05-19 20:50:29 +08001989 swp_entry_t swapin_error;
1990 void *old;
Baolin Wang40ff2d12024-08-12 15:42:06 +08001991 int nr_pages;
Miaohe Lin6cec2b92022-05-19 20:50:29 +08001992
Axel Rasmussenaf194872023-07-07 14:55:33 -07001993 swapin_error = make_poisoned_swp_entry();
Miaohe Lin6cec2b92022-05-19 20:50:29 +08001994 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1995 swp_to_radix_entry(swap),
1996 swp_to_radix_entry(swapin_error), 0);
1997 if (old != swp_to_radix_entry(swap))
1998 return;
1999
Baolin Wang40ff2d12024-08-12 15:42:06 +08002000 nr_pages = folio_nr_pages(folio);
Miaohe Lin6cec2b92022-05-19 20:50:29 +08002001 folio_wait_writeback(folio);
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +01002002 delete_from_swap_cache(folio);
Miaohe Lin6cec2b92022-05-19 20:50:29 +08002003 /*
Hugh Dickins3c1b7522023-08-03 22:46:11 -07002004 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
2005 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
2006 * in shmem_evict_inode().
Miaohe Lin6cec2b92022-05-19 20:50:29 +08002007 */
Baolin Wang40ff2d12024-08-12 15:42:06 +08002008 shmem_recalc_inode(inode, -nr_pages, -nr_pages);
2009 swap_free_nr(swap, nr_pages);
Miaohe Lin6cec2b92022-05-19 20:50:29 +08002010}
2011
Baolin Wang12885cb2024-08-12 15:42:09 +08002012static int shmem_split_large_entry(struct inode *inode, pgoff_t index,
2013 swp_entry_t swap, gfp_t gfp)
2014{
2015 struct address_space *mapping = inode->i_mapping;
2016 XA_STATE_ORDER(xas, &mapping->i_pages, index, 0);
2017 void *alloced_shadow = NULL;
2018 int alloced_order = 0, i;
2019
2020 /* Convert user data gfp flags to xarray node gfp flags */
2021 gfp &= GFP_RECLAIM_MASK;
2022
2023 for (;;) {
2024 int order = -1, split_order = 0;
2025 void *old = NULL;
2026
2027 xas_lock_irq(&xas);
2028 old = xas_load(&xas);
2029 if (!xa_is_value(old) || swp_to_radix_entry(swap) != old) {
2030 xas_set_err(&xas, -EEXIST);
2031 goto unlock;
2032 }
2033
2034 order = xas_get_order(&xas);
2035
2036 /* Swap entry may have changed before we re-acquire the lock */
2037 if (alloced_order &&
2038 (old != alloced_shadow || order != alloced_order)) {
2039 xas_destroy(&xas);
2040 alloced_order = 0;
2041 }
2042
2043 /* Try to split large swap entry in pagecache */
2044 if (order > 0) {
2045 if (!alloced_order) {
2046 split_order = order;
2047 goto unlock;
2048 }
2049 xas_split(&xas, old, order);
2050
2051 /*
2052 * Re-set the swap entry after splitting, and the swap
2053 * offset of the original large entry must be continuous.
2054 */
2055 for (i = 0; i < 1 << order; i++) {
2056 pgoff_t aligned_index = round_down(index, 1 << order);
2057 swp_entry_t tmp;
2058
2059 tmp = swp_entry(swp_type(swap), swp_offset(swap) + i);
2060 __xa_store(&mapping->i_pages, aligned_index + i,
2061 swp_to_radix_entry(tmp), 0);
2062 }
2063 }
2064
2065unlock:
2066 xas_unlock_irq(&xas);
2067
2068 /* split needed, alloc here and retry. */
2069 if (split_order) {
2070 xas_split_alloc(&xas, old, split_order, gfp);
2071 if (xas_error(&xas))
2072 goto error;
2073 alloced_shadow = old;
2074 alloced_order = split_order;
2075 xas_reset(&xas);
2076 continue;
2077 }
2078
2079 if (!xas_nomem(&xas, gfp))
2080 break;
2081 }
2082
2083error:
2084 if (xas_error(&xas))
2085 return xas_error(&xas);
2086
2087 return alloced_order;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07002088}
2089
2090/*
Miaohe Lin833de102022-05-30 19:58:41 +08002091 * Swap in the folio pointed to by *foliop.
2092 * Caller has to make sure that *foliop contains a valid swapped folio.
2093 * Returns 0 and the folio in foliop if success. On failure, returns the
2094 * error code and NULL in *foliop.
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002095 */
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002096static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
2097 struct folio **foliop, enum sgp_type sgp,
Baolin Wang736f0e02024-08-12 15:42:07 +08002098 gfp_t gfp, struct vm_area_struct *vma,
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002099 vm_fault_t *fault_type)
2100{
2101 struct address_space *mapping = inode->i_mapping;
Baolin Wang736f0e02024-08-12 15:42:07 +08002102 struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002103 struct shmem_inode_info *info = SHMEM_I(inode);
Kairui Songcbc2bd92022-12-20 02:58:40 +08002104 struct swap_info_struct *si;
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002105 struct folio *folio = NULL;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002106 swp_entry_t swap;
Baolin Wang40ff2d12024-08-12 15:42:06 +08002107 int error, nr_pages;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002108
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002109 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
2110 swap = radix_to_swp_entry(*foliop);
2111 *foliop = NULL;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002112
Axel Rasmussenaf194872023-07-07 14:55:33 -07002113 if (is_poisoned_swp_entry(swap))
Miaohe Lin6cec2b92022-05-19 20:50:29 +08002114 return -EIO;
2115
Kairui Songcbc2bd92022-12-20 02:58:40 +08002116 si = get_swap_device(swap);
2117 if (!si) {
2118 if (!shmem_confirm_swap(mapping, index, swap))
2119 return -EEXIST;
2120 else
2121 return -EINVAL;
2122 }
2123
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002124 /* Look it up and read it in.. */
Matthew Wilcox (Oracle)5739a812022-09-02 20:46:16 +01002125 folio = swap_cache_get_folio(swap, NULL, 0);
2126 if (!folio) {
Baolin Wang12885cb2024-08-12 15:42:09 +08002127 int split_order;
2128
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002129 /* Or update major stats only when swapin succeeds?? */
2130 if (fault_type) {
2131 *fault_type |= VM_FAULT_MAJOR;
2132 count_vm_event(PGMAJFAULT);
Hugh Dickins054a9f72023-09-29 20:31:27 -07002133 count_memcg_event_mm(fault_mm, PGMAJFAULT);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002134 }
Baolin Wang12885cb2024-08-12 15:42:09 +08002135
2136 /*
2137 * Now swap device can only swap in order 0 folio, then we
2138 * should split the large swap entry stored in the pagecache
2139 * if necessary.
2140 */
2141 split_order = shmem_split_large_entry(inode, index, swap, gfp);
2142 if (split_order < 0) {
2143 error = split_order;
2144 goto failed;
2145 }
2146
2147 /*
2148 * If the large swap entry has already been split, it is
2149 * necessary to recalculate the new swap entry based on
2150 * the old order alignment.
2151 */
2152 if (split_order > 0) {
2153 pgoff_t offset = index - round_down(index, 1 << split_order);
2154
2155 swap = swp_entry(swp_type(swap), swp_offset(swap) + offset);
2156 }
2157
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002158 /* Here we actually start the io */
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07002159 folio = shmem_swapin_cluster(swap, gfp, info, index);
Matthew Wilcox (Oracle)5739a812022-09-02 20:46:16 +01002160 if (!folio) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002161 error = -ENOMEM;
2162 goto failed;
2163 }
2164 }
2165
Miaohe Lin833de102022-05-30 19:58:41 +08002166 /* We have to do this with folio locked to prevent races */
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002167 folio_lock(folio);
2168 if (!folio_test_swapcache(folio) ||
David Hildenbrand3d2c9082023-08-21 18:08:48 +02002169 folio->swap.val != swap.val ||
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002170 !shmem_confirm_swap(mapping, index, swap)) {
2171 error = -EEXIST;
2172 goto unlock;
2173 }
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002174 if (!folio_test_uptodate(folio)) {
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002175 error = -EIO;
2176 goto failed;
2177 }
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002178 folio_wait_writeback(folio);
Baolin Wang40ff2d12024-08-12 15:42:06 +08002179 nr_pages = folio_nr_pages(folio);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002180
Steven Price8a848022020-05-13 16:37:49 +01002181 /*
2182 * Some architectures may have to restore extra metadata to the
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002183 * folio after reading from swap.
Steven Price8a848022020-05-13 16:37:49 +01002184 */
Barry Songf238b8c2024-03-23 00:41:36 +13002185 arch_swap_restore(folio_swap(swap, folio), folio);
Steven Price8a848022020-05-13 16:37:49 +01002186
Matthew Wilcox (Oracle)069d8492022-05-12 20:23:04 -07002187 if (shmem_should_replace_folio(folio, gfp)) {
Baolin Wang736f0e02024-08-12 15:42:07 +08002188 error = shmem_replace_folio(&folio, gfp, info, index, vma);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002189 if (error)
2190 goto failed;
2191 }
2192
Baolin Wang40ff2d12024-08-12 15:42:06 +08002193 error = shmem_add_to_page_cache(folio, mapping,
2194 round_down(index, nr_pages),
Hugh Dickins054a9f72023-09-29 20:31:27 -07002195 swp_to_radix_entry(swap), gfp);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002196 if (error)
2197 goto failed;
2198
Baolin Wang40ff2d12024-08-12 15:42:06 +08002199 shmem_recalc_inode(inode, 0, -nr_pages);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002200
2201 if (sgp == SGP_WRITE)
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002202 folio_mark_accessed(folio);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002203
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +01002204 delete_from_swap_cache(folio);
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002205 folio_mark_dirty(folio);
Baolin Wang40ff2d12024-08-12 15:42:06 +08002206 swap_free_nr(swap, nr_pages);
Kairui Songcbc2bd92022-12-20 02:58:40 +08002207 put_swap_device(si);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002208
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002209 *foliop = folio;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002210 return 0;
2211failed:
2212 if (!shmem_confirm_swap(mapping, index, swap))
2213 error = -EEXIST;
Miaohe Lin6cec2b92022-05-19 20:50:29 +08002214 if (error == -EIO)
2215 shmem_set_folio_swapin_error(inode, index, folio, swap);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002216unlock:
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002217 if (folio) {
2218 folio_unlock(folio);
2219 folio_put(folio);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002220 }
Kairui Songcbc2bd92022-12-20 02:58:40 +08002221 put_swap_device(si);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002222
2223 return error;
2224}
2225
2226/*
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002227 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 *
2229 * If we allocate a new one we do not mark it dirty. That's up to the
2230 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002231 * entry since a page cannot live in both the swap and page cache.
2232 *
Hugh Dickinse3e1a5062023-09-29 20:26:53 -07002233 * vmf and fault_type are only supplied by shmem_fault: otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 */
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002235static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
Rik van Riele1e4cfd2024-09-03 11:19:28 -04002236 loff_t write_end, struct folio **foliop, enum sgp_type sgp,
2237 gfp_t gfp, struct vm_fault *vmf, vm_fault_t *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238{
Hugh Dickinse3e1a5062023-09-29 20:26:53 -07002239 struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
Hugh Dickins054a9f72023-09-29 20:31:27 -07002240 struct mm_struct *fault_mm;
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -07002241 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 int error;
Baolin Wang6beeab82024-07-22 13:43:19 +08002243 bool alloced;
Baolin Wange7a2ab72024-06-11 18:11:08 +08002244 unsigned long orders = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
Christoph Hellwig1cd81fa2024-02-19 07:27:12 +01002246 if (WARN_ON_ONCE(!shmem_mapping(inode->i_mapping)))
2247 return -EINVAL;
2248
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002249 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251repeat:
Hugh Dickins75edd342016-05-19 17:12:44 -07002252 if (sgp <= SGP_CACHE &&
Hugh Dickins3022fd72023-09-29 20:32:40 -07002253 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode))
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002254 return -EINVAL;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002255
Hugh Dickins3022fd72023-09-29 20:32:40 -07002256 alloced = false;
Hugh Dickins054a9f72023-09-29 20:31:27 -07002257 fault_mm = vma ? vma->vm_mm : NULL;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002258
Hugh Dickins3022fd72023-09-29 20:32:40 -07002259 folio = filemap_get_entry(inode->i_mapping, index);
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002260 if (folio && vma && userfaultfd_minor(vma)) {
Christoph Hellwigaaeb94e2023-03-07 15:34:08 +01002261 if (!xa_is_value(folio))
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002262 folio_put(folio);
Axel Rasmussenc949b092021-06-30 18:49:20 -07002263 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
2264 return 0;
2265 }
2266
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002267 if (xa_is_value(folio)) {
Matthew Wilcox (Oracle)da08e9b2022-05-12 20:23:05 -07002268 error = shmem_swapin_folio(inode, index, &folio,
Baolin Wang736f0e02024-08-12 15:42:07 +08002269 sgp, gfp, vma, fault_type);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002270 if (error == -EEXIST)
2271 goto repeat;
2272
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002273 *foliop = folio;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002274 return error;
Hugh Dickins54af60422011-08-03 16:21:24 -07002275 }
2276
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002277 if (folio) {
Christoph Hellwigaaeb94e2023-03-07 15:34:08 +01002278 folio_lock(folio);
2279
2280 /* Has the folio been truncated or swapped out? */
Hugh Dickins3022fd72023-09-29 20:32:40 -07002281 if (unlikely(folio->mapping != inode->i_mapping)) {
Christoph Hellwigaaeb94e2023-03-07 15:34:08 +01002282 folio_unlock(folio);
2283 folio_put(folio);
2284 goto repeat;
2285 }
Hugh Dickinsacdd9f8e2021-09-02 14:54:34 -07002286 if (sgp == SGP_WRITE)
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002287 folio_mark_accessed(folio);
2288 if (folio_test_uptodate(folio))
Hugh Dickinsacdd9f8e2021-09-02 14:54:34 -07002289 goto out;
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002290 /* fallocated folio */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002291 if (sgp != SGP_READ)
2292 goto clear;
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002293 folio_unlock(folio);
2294 folio_put(folio);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 /*
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002298 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
2299 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
Hugh Dickinsacdd9f8e2021-09-02 14:54:34 -07002300 */
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002301 *foliop = NULL;
Hugh Dickinsacdd9f8e2021-09-02 14:54:34 -07002302 if (sgp == SGP_READ)
2303 return 0;
2304 if (sgp == SGP_NOALLOC)
2305 return -ENOENT;
2306
2307 /*
2308 * Fast cache lookup and swap lookup did not find it: allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 */
Hugh Dickins27ab7002011-07-25 17:12:36 -07002310
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002311 if (vma && userfaultfd_missing(vma)) {
2312 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2313 return 0;
2314 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
Baolin Wang6beeab82024-07-22 13:43:19 +08002316 /* Find hugepage orders that are allowed for anonymous shmem and tmpfs. */
Rik van Riele1e4cfd2024-09-03 11:19:28 -04002317 orders = shmem_allowable_huge_orders(inode, vma, index, write_end, false);
Baolin Wange7a2ab72024-06-11 18:11:08 +08002318 if (orders > 0) {
Hugh Dickins3022fd72023-09-29 20:32:40 -07002319 gfp_t huge_gfp;
Kees Cook27d80fa22020-04-06 20:07:51 -07002320
Hugh Dickins3022fd72023-09-29 20:32:40 -07002321 huge_gfp = vma_thp_gfp_mask(vma);
2322 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
Baolin Wange7a2ab72024-06-11 18:11:08 +08002323 folio = shmem_alloc_and_add_folio(vmf, huge_gfp,
2324 inode, index, fault_mm, orders);
Hugh Dickins3022fd72023-09-29 20:32:40 -07002325 if (!IS_ERR(folio)) {
Baolin Wang3d95bc212024-06-11 18:11:06 +08002326 if (folio_test_pmd_mappable(folio))
2327 count_vm_event(THP_FILE_ALLOC);
Ryan Roberts63d98662024-07-10 10:55:01 +01002328 count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC);
Hugh Dickins3022fd72023-09-29 20:32:40 -07002329 goto alloced;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002330 }
Hugh Dickins3022fd72023-09-29 20:32:40 -07002331 if (PTR_ERR(folio) == -EEXIST)
2332 goto repeat;
2333 }
2334
Baolin Wange7a2ab72024-06-11 18:11:08 +08002335 folio = shmem_alloc_and_add_folio(vmf, gfp, inode, index, fault_mm, 0);
Hugh Dickins3022fd72023-09-29 20:32:40 -07002336 if (IS_ERR(folio)) {
2337 error = PTR_ERR(folio);
2338 if (error == -EEXIST)
2339 goto repeat;
2340 folio = NULL;
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002341 goto unlock;
2342 }
2343
Hugh Dickins3022fd72023-09-29 20:32:40 -07002344alloced:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002345 alloced = true;
Baolin Wange7a2ab72024-06-11 18:11:08 +08002346 if (folio_test_large(folio) &&
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002347 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
Hugh Dickinsde5b8522024-08-25 15:42:45 -07002348 folio_next_index(folio)) {
Hugh Dickins3022fd72023-09-29 20:32:40 -07002349 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2350 struct shmem_inode_info *info = SHMEM_I(inode);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002351 /*
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002352 * Part of the large folio is beyond i_size: subject
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002353 * to shrink under memory pressure.
2354 */
2355 spin_lock(&sbinfo->shrinklist_lock);
2356 /*
2357 * _careful to defend against unlocked access to
2358 * ->shrink_list in shmem_unused_huge_shrink()
2359 */
2360 if (list_empty_careful(&info->shrinklist)) {
2361 list_add_tail(&info->shrinklist,
2362 &sbinfo->shrinklist);
2363 sbinfo->shrinklist_len++;
2364 }
2365 spin_unlock(&sbinfo->shrinklist_lock);
2366 }
2367
Hugh Dickins3022fd72023-09-29 20:32:40 -07002368 if (sgp == SGP_WRITE)
2369 folio_set_referenced(folio);
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002370 /*
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002371 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002372 */
2373 if (sgp == SGP_FALLOC)
2374 sgp = SGP_WRITE;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002375clear:
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002376 /*
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002377 * Let SGP_WRITE caller clear ends if write does not fill folio;
2378 * but SGP_FALLOC on a folio fallocated earlier must initialize
Vineeth Remanan Pillaic5bf1212019-03-05 15:46:58 -08002379 * it now, lest undo on failure cancel our earlier guarantee.
2380 */
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002381 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2382 long i, n = folio_nr_pages(folio);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002383
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002384 for (i = 0; i < n; i++)
2385 clear_highpage(folio_page(folio, i));
2386 flush_dcache_folio(folio);
2387 folio_mark_uptodate(folio);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07002388 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07002389
Hugh Dickins54af60422011-08-03 16:21:24 -07002390 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07002391 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002392 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07002393 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08002394 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07002395 }
Matthew Wilcox (Oracle)63ec1972020-10-13 16:51:38 -07002396out:
Matthew Wilcox (Oracle)fc26bab2022-09-02 20:46:17 +01002397 *foliop = folio;
Hugh Dickins54af60422011-08-03 16:21:24 -07002398 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07002399
Nick Piggind0217ac2007-07-19 01:47:03 -07002400 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002401 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 */
Hugh Dickinsd1899222012-07-11 14:02:47 -07002403unlock:
Hugh Dickins3022fd72023-09-29 20:32:40 -07002404 if (alloced)
2405 filemap_remove_folio(folio);
2406 shmem_recalc_inode(inode, 0, 0);
Matthew Wilcox (Oracle)b1d0ec3a2022-05-12 20:23:05 -07002407 if (folio) {
2408 folio_unlock(folio);
2409 folio_put(folio);
Hugh Dickins54af60422011-08-03 16:21:24 -07002410 }
Hugh Dickins54af60422011-08-03 16:21:24 -07002411 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412}
2413
Christoph Hellwigd7468602024-02-19 07:27:13 +01002414/**
2415 * shmem_get_folio - find, and lock a shmem folio.
2416 * @inode: inode to search
2417 * @index: the page index.
Rik van Riele1e4cfd2024-09-03 11:19:28 -04002418 * @write_end: end of a write, could extend inode size
Christoph Hellwigd7468602024-02-19 07:27:13 +01002419 * @foliop: pointer to the folio if found
2420 * @sgp: SGP_* flags to control behavior
2421 *
2422 * Looks up the page cache entry at @inode & @index. If a folio is
2423 * present, it is returned locked with an increased refcount.
2424 *
Christoph Hellwig9d8b3672024-02-19 07:27:15 +01002425 * If the caller modifies data in the folio, it must call folio_mark_dirty()
2426 * before unlocking the folio to ensure that the folio is not reclaimed.
2427 * There is no need to reserve space before calling folio_mark_dirty().
2428 *
Christoph Hellwigd7468602024-02-19 07:27:13 +01002429 * When no folio is found, the behavior depends on @sgp:
Akira Yokosawa8d4dd9d2024-02-27 14:06:48 +09002430 * - for SGP_READ, *@foliop is %NULL and 0 is returned
2431 * - for SGP_NOALLOC, *@foliop is %NULL and -ENOENT is returned
Christoph Hellwigd7468602024-02-19 07:27:13 +01002432 * - for all other flags a new folio is allocated, inserted into the
2433 * page cache and returned locked in @foliop.
2434 *
2435 * Context: May sleep.
2436 * Return: 0 if successful, else a negative error code.
2437 */
Rik van Riele1e4cfd2024-09-03 11:19:28 -04002438int shmem_get_folio(struct inode *inode, pgoff_t index, loff_t write_end,
2439 struct folio **foliop, enum sgp_type sgp)
Matthew Wilcox (Oracle)4e1fc792022-09-02 20:46:20 +01002440{
Rik van Riele1e4cfd2024-09-03 11:19:28 -04002441 return shmem_get_folio_gfp(inode, index, write_end, foliop, sgp,
Hugh Dickinse3e1a5062023-09-29 20:26:53 -07002442 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
Matthew Wilcox (Oracle)4e1fc792022-09-02 20:46:20 +01002443}
Christoph Hellwigd7468602024-02-19 07:27:13 +01002444EXPORT_SYMBOL_GPL(shmem_get_folio);
Matthew Wilcox (Oracle)4e1fc792022-09-02 20:46:20 +01002445
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002446/*
2447 * This is like autoremove_wake_function, but it removes the wait queue
2448 * entry unconditionally - even if something else had already woken the
2449 * target.
2450 */
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002451static int synchronous_wake_function(wait_queue_entry_t *wait,
2452 unsigned int mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002453{
2454 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02002455 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08002456 return ret;
2457}
2458
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002459/*
2460 * Trinity finds that probing a hole which tmpfs is punching can
2461 * prevent the hole-punch from ever completing: which in turn
2462 * locks writers out with its hold on i_rwsem. So refrain from
2463 * faulting pages into the hole while it's being punched. Although
2464 * shmem_undo_range() does remove the additions, it may be unable to
2465 * keep up, as each new page needs its own unmap_mapping_range() call,
2466 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2467 *
2468 * It does not matter if we sometimes reach this check just before the
2469 * hole-punch begins, so that one fault then races with the punch:
2470 * we just need to make racing faults a rare case.
2471 *
2472 * The implementation below would be much simpler if we just used a
2473 * standard mutex or completion: but we cannot take i_rwsem in fault,
2474 * and bloating every shmem inode for this unlikely case would be sad.
2475 */
2476static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode)
2477{
2478 struct shmem_falloc *shmem_falloc;
2479 struct file *fpin = NULL;
2480 vm_fault_t ret = 0;
2481
2482 spin_lock(&inode->i_lock);
2483 shmem_falloc = inode->i_private;
2484 if (shmem_falloc &&
2485 shmem_falloc->waitq &&
2486 vmf->pgoff >= shmem_falloc->start &&
2487 vmf->pgoff < shmem_falloc->next) {
2488 wait_queue_head_t *shmem_falloc_waitq;
2489 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2490
2491 ret = VM_FAULT_NOPAGE;
2492 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2493 shmem_falloc_waitq = shmem_falloc->waitq;
2494 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2495 TASK_UNINTERRUPTIBLE);
2496 spin_unlock(&inode->i_lock);
2497 schedule();
2498
2499 /*
2500 * shmem_falloc_waitq points into the shmem_fallocate()
2501 * stack of the hole-punching task: shmem_falloc_waitq
2502 * is usually invalid by the time we reach here, but
2503 * finish_wait() does not dereference it in that case;
2504 * though i_lock needed lest racing with wake_up_all().
2505 */
2506 spin_lock(&inode->i_lock);
2507 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2508 }
2509 spin_unlock(&inode->i_lock);
2510 if (fpin) {
2511 fput(fpin);
2512 ret = VM_FAULT_RETRY;
2513 }
2514 return ret;
2515}
2516
Souptick Joarder20acce62018-06-07 17:09:17 -07002517static vm_fault_t shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518{
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002519 struct inode *inode = file_inode(vmf->vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002520 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Matthew Wilcox (Oracle)68a54102022-09-02 20:46:18 +01002521 struct folio *folio = NULL;
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002522 vm_fault_t ret = 0;
Souptick Joarder20acce62018-06-07 17:09:17 -07002523 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002525 /*
2526 * Trinity finds that probing a hole which tmpfs is punching can
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002527 * prevent the hole-punch from ever completing: noted in i_private.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002528 */
2529 if (unlikely(inode->i_private)) {
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002530 ret = shmem_falloc_wait(vmf, inode);
2531 if (ret)
Hugh Dickins8e205f72014-07-23 14:00:10 -07002532 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002533 }
2534
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002535 WARN_ON_ONCE(vmf->page != NULL);
Rik van Riele1e4cfd2024-09-03 11:19:28 -04002536 err = shmem_get_folio_gfp(inode, vmf->pgoff, 0, &folio, SGP_CACHE,
Hugh Dickinse3e1a5062023-09-29 20:26:53 -07002537 gfp, vmf, &ret);
Souptick Joarder20acce62018-06-07 17:09:17 -07002538 if (err)
2539 return vmf_error(err);
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002540 if (folio) {
Matthew Wilcox (Oracle)68a54102022-09-02 20:46:18 +01002541 vmf->page = folio_file_page(folio, vmf->pgoff);
Hugh Dickinsf0a9ad12023-09-29 20:27:53 -07002542 ret |= VM_FAULT_LOCKED;
2543 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07002544 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545}
2546
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002547unsigned long shmem_get_unmapped_area(struct file *file,
2548 unsigned long uaddr, unsigned long len,
2549 unsigned long pgoff, unsigned long flags)
2550{
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002551 unsigned long addr;
2552 unsigned long offset;
2553 unsigned long inflated_len;
2554 unsigned long inflated_addr;
2555 unsigned long inflated_offset;
Baolin Wang5a9dd102024-06-11 18:11:09 +08002556 unsigned long hpage_size;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002557
2558 if (len > TASK_SIZE)
2559 return -ENOMEM;
2560
Rick Edgecombe529ce232024-03-25 19:16:44 -07002561 addr = mm_get_unmapped_area(current->mm, file, uaddr, len, pgoff,
2562 flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002563
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07002564 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002565 return addr;
2566 if (IS_ERR_VALUE(addr))
2567 return addr;
2568 if (addr & ~PAGE_MASK)
2569 return addr;
2570 if (addr > TASK_SIZE - len)
2571 return addr;
2572
2573 if (shmem_huge == SHMEM_HUGE_DENY)
2574 return addr;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002575 if (flags & MAP_FIXED)
2576 return addr;
2577 /*
2578 * Our priority is to support MAP_SHARED mapped hugely;
2579 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002580 * But if caller specified an address hint and we allocated area there
2581 * successfully, respect that as before.
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002582 */
Kirill A. Shutemov99158992020-01-13 16:29:13 -08002583 if (uaddr == addr)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002584 return addr;
2585
Baolin Wang5a9dd102024-06-11 18:11:09 +08002586 hpage_size = HPAGE_PMD_SIZE;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002587 if (shmem_huge != SHMEM_HUGE_FORCE) {
2588 struct super_block *sb;
Baolin Wang5a9dd102024-06-11 18:11:09 +08002589 unsigned long __maybe_unused hpage_orders;
2590 int order = 0;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002591
2592 if (file) {
2593 VM_BUG_ON(file->f_op != &shmem_file_operations);
2594 sb = file_inode(file)->i_sb;
2595 } else {
2596 /*
2597 * Called directly from mm/mmap.c, or drivers/char/mem.c
2598 * for "/dev/zero", to create a shared anonymous object.
2599 */
2600 if (IS_ERR(shm_mnt))
2601 return addr;
2602 sb = shm_mnt->mnt_sb;
Baolin Wang5a9dd102024-06-11 18:11:09 +08002603
2604 /*
2605 * Find the highest mTHP order used for anonymous shmem to
2606 * provide a suitable alignment address.
2607 */
2608#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2609 hpage_orders = READ_ONCE(huge_shmem_orders_always);
2610 hpage_orders |= READ_ONCE(huge_shmem_orders_within_size);
2611 hpage_orders |= READ_ONCE(huge_shmem_orders_madvise);
2612 if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER)
2613 hpage_orders |= READ_ONCE(huge_shmem_orders_inherit);
2614
2615 if (hpage_orders > 0) {
2616 order = highest_order(hpage_orders);
2617 hpage_size = PAGE_SIZE << order;
2618 }
2619#endif
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002620 }
Baolin Wang5a9dd102024-06-11 18:11:09 +08002621 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER && !order)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002622 return addr;
2623 }
2624
Baolin Wang5a9dd102024-06-11 18:11:09 +08002625 if (len < hpage_size)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002626 return addr;
2627
Baolin Wang5a9dd102024-06-11 18:11:09 +08002628 offset = (pgoff << PAGE_SHIFT) & (hpage_size - 1);
2629 if (offset && offset + len < 2 * hpage_size)
2630 return addr;
2631 if ((addr & (hpage_size - 1)) == offset)
2632 return addr;
2633
2634 inflated_len = len + hpage_size - PAGE_SIZE;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002635 if (inflated_len > TASK_SIZE)
2636 return addr;
2637 if (inflated_len < len)
2638 return addr;
2639
Rick Edgecombe529ce232024-03-25 19:16:44 -07002640 inflated_addr = mm_get_unmapped_area(current->mm, NULL, uaddr,
2641 inflated_len, 0, flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002642 if (IS_ERR_VALUE(inflated_addr))
2643 return addr;
2644 if (inflated_addr & ~PAGE_MASK)
2645 return addr;
2646
Baolin Wang5a9dd102024-06-11 18:11:09 +08002647 inflated_offset = inflated_addr & (hpage_size - 1);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002648 inflated_addr += offset - inflated_offset;
2649 if (inflated_offset > offset)
Baolin Wang5a9dd102024-06-11 18:11:09 +08002650 inflated_addr += hpage_size;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002651
2652 if (inflated_addr > TASK_SIZE - len)
2653 return addr;
2654 return inflated_addr;
2655}
2656
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002658static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659{
Al Viro496ad9a2013-01-23 17:07:38 -05002660 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002661 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662}
2663
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002664static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07002665 unsigned long addr, pgoff_t *ilx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666{
Al Viro496ad9a2013-01-23 17:07:38 -05002667 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002668 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07002670 /*
2671 * Bias interleave by inode number to distribute better across nodes;
2672 * but this interface is independent of which page order is used, so
2673 * supplies only that bias, letting caller apply the offset (adjusted
2674 * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()).
2675 */
2676 *ilx = inode->i_ino;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002677 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2678 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679}
Hugh Dickinsddc1a5c2023-10-19 13:39:08 -07002680
2681static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2682 pgoff_t index, unsigned int order, pgoff_t *ilx)
2683{
2684 struct mempolicy *mpol;
2685
2686 /* Bias interleave by inode number to distribute better across nodes */
2687 *ilx = info->vfs_inode.i_ino + (index >> order);
2688
2689 mpol = mpol_shared_policy_lookup(&info->policy, index);
2690 return mpol ? mpol : get_task_policy(current);
2691}
2692#else
2693static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2694 pgoff_t index, unsigned int order, pgoff_t *ilx)
2695{
2696 *ilx = 0;
2697 return NULL;
2698}
2699#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
Alexey Gladkovd7c9e992021-04-22 14:27:14 +02002701int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702{
Al Viro496ad9a2013-01-23 17:07:38 -05002703 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 struct shmem_inode_info *info = SHMEM_I(inode);
2705 int retval = -ENOMEM;
2706
Hugh Dickinsea0dfeb2020-04-20 18:14:14 -07002707 /*
2708 * What serializes the accesses to info->flags?
2709 * ipc_lock_object() when called from shmctl_do_lock(),
2710 * no serialization needed when called from shm_destroy().
2711 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 if (lock && !(info->flags & VM_LOCKED)) {
Alexey Gladkovd7c9e992021-04-22 14:27:14 +02002713 if (!user_shm_lock(inode->i_size, ucounts))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 goto out_nomem;
2715 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002716 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 }
Alexey Gladkovd7c9e992021-04-22 14:27:14 +02002718 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2719 user_shm_unlock(inode->i_size, ucounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002721 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 }
2723 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002724
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725out_nomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 return retval;
2727}
2728
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002729static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730{
Pasha Tatashind09e8ca2022-11-15 02:06:01 +00002731 struct inode *inode = file_inode(file);
2732 struct shmem_inode_info *info = SHMEM_I(inode);
Peter Xu22247ef2021-05-14 17:27:04 -07002733 int ret;
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002734
Lorenzo Stoakes28464bb2023-10-12 18:04:29 +01002735 ret = seal_check_write(info->seals, vma);
Peter Xu22247ef2021-05-14 17:27:04 -07002736 if (ret)
2737 return ret;
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08002738
Catalin Marinas51b0bff2019-11-29 12:45:08 +00002739 /* arm64 - allow memory tagging on RAM-based files */
Suren Baghdasaryan1c712222023-01-26 11:37:49 -08002740 vm_flags_set(vma, VM_MTE_ALLOWED);
Catalin Marinas51b0bff2019-11-29 12:45:08 +00002741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 file_accessed(file);
Pasha Tatashind09e8ca2022-11-15 02:06:01 +00002743 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2744 if (inode->i_nlink)
2745 vma->vm_ops = &shmem_vm_ops;
2746 else
2747 vma->vm_ops = &shmem_anon_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 return 0;
2749}
2750
Hugh Dickinse88e0d32023-08-10 23:27:07 -07002751static int shmem_file_open(struct inode *inode, struct file *file)
2752{
2753 file->f_mode |= FMODE_CAN_ODIRECT;
2754 return generic_file_open(inode, file);
2755}
2756
Hugh Dickinscb241332022-08-10 21:51:09 -07002757#ifdef CONFIG_TMPFS_XATTR
2758static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2759
2760/*
2761 * chattr's fsflags are unrelated to extended attributes,
2762 * but tmpfs has chosen to enable them under the same config option.
2763 */
2764static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
Theodore Ts'oe408e692022-07-14 21:59:12 -04002765{
Hugh Dickinscb241332022-08-10 21:51:09 -07002766 unsigned int i_flags = 0;
2767
2768 if (fsflags & FS_NOATIME_FL)
2769 i_flags |= S_NOATIME;
2770 if (fsflags & FS_APPEND_FL)
2771 i_flags |= S_APPEND;
2772 if (fsflags & FS_IMMUTABLE_FL)
2773 i_flags |= S_IMMUTABLE;
2774 /*
2775 * But FS_NODUMP_FL does not require any action in i_flags.
2776 */
2777 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
Theodore Ts'oe408e692022-07-14 21:59:12 -04002778}
Hugh Dickinscb241332022-08-10 21:51:09 -07002779#else
2780static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2781{
2782}
2783#define shmem_initxattrs NULL
2784#endif
Theodore Ts'oe408e692022-07-14 21:59:12 -04002785
Chuck Levera2e45952023-06-30 13:49:03 -04002786static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2787{
2788 return &SHMEM_I(inode)->dir_offsets;
2789}
2790
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02002791static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2792 struct super_block *sb,
2793 struct inode *dir, umode_t mode,
2794 dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795{
2796 struct inode *inode;
2797 struct shmem_inode_info *info;
2798 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Chris Downe809d5f2020-08-06 23:20:20 -07002799 ino_t ino;
Carlos Maiolino71480662023-07-25 16:45:05 +02002800 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801
Carlos Maiolino71480662023-07-25 16:45:05 +02002802 err = shmem_reserve_inode(sb, &ino);
2803 if (err)
2804 return ERR_PTR(err);
2805
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 inode = new_inode(sb);
Carlos Maiolino71480662023-07-25 16:45:05 +02002807 if (!inode) {
Hugh Dickins2daf18a2023-08-08 21:33:56 -07002808 shmem_free_inode(sb, 0);
Carlos Maiolino71480662023-07-25 16:45:05 +02002809 return ERR_PTR(-ENOSPC);
2810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811
Carlos Maiolino71480662023-07-25 16:45:05 +02002812 inode->i_ino = ino;
2813 inode_init_owner(idmap, inode, dir, mode);
2814 inode->i_blocks = 0;
Jeff Laytoncf2766b2023-10-04 14:53:07 -04002815 simple_inode_init_ts(inode);
Carlos Maiolino71480662023-07-25 16:45:05 +02002816 inode->i_generation = get_random_u32();
2817 info = SHMEM_I(inode);
2818 memset(info, 0, (char *)inode - (char *)info);
2819 spin_lock_init(&info->lock);
2820 atomic_set(&info->stop_eviction, 0);
2821 info->seals = F_SEAL_SEAL;
2822 info->flags = flags & VM_NORESERVE;
Jeff Laytoncf2766b2023-10-04 14:53:07 -04002823 info->i_crtime = inode_get_mtime(inode);
Carlos Maiolino71480662023-07-25 16:45:05 +02002824 info->fsflags = (dir == NULL) ? 0 :
2825 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2826 if (info->fsflags)
2827 shmem_set_inode_flags(inode, info->fsflags);
2828 INIT_LIST_HEAD(&info->shrinklist);
2829 INIT_LIST_HEAD(&info->swaplist);
Carlos Maiolino71480662023-07-25 16:45:05 +02002830 simple_xattrs_init(&info->xattrs);
2831 cache_no_acl(inode);
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07002832 if (sbinfo->noswap)
2833 mapping_set_unevictable(inode->i_mapping);
Carlos Maiolino71480662023-07-25 16:45:05 +02002834 mapping_set_large_folios(inode->i_mapping);
Joel Fernandes (Google)b45d71f2018-09-20 12:22:39 -07002835
Carlos Maiolino71480662023-07-25 16:45:05 +02002836 switch (mode & S_IFMT) {
2837 default:
2838 inode->i_op = &shmem_special_inode_operations;
2839 init_special_inode(inode, mode, dev);
2840 break;
2841 case S_IFREG:
2842 inode->i_mapping->a_ops = &shmem_aops;
2843 inode->i_op = &shmem_inode_operations;
2844 inode->i_fop = &shmem_file_operations;
2845 mpol_shared_policy_init(&info->policy,
2846 shmem_get_sbmpol(sbinfo));
2847 break;
2848 case S_IFDIR:
2849 inc_nlink(inode);
2850 /* Some things misbehave if size == 0 on a directory */
2851 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2852 inode->i_op = &shmem_dir_inode_operations;
Chuck Levera2e45952023-06-30 13:49:03 -04002853 inode->i_fop = &simple_offset_dir_operations;
2854 simple_offset_init(shmem_get_offset_ctx(inode));
Carlos Maiolino71480662023-07-25 16:45:05 +02002855 break;
2856 case S_IFLNK:
2857 /*
2858 * Must not load anything in the rbtree,
2859 * mpol_free_shared_policy will not be called.
2860 */
2861 mpol_shared_policy_init(&info->policy, NULL);
2862 break;
2863 }
2864
2865 lockdep_annotate_inode_mutex_key(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 return inode;
2867}
2868
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02002869#ifdef CONFIG_TMPFS_QUOTA
2870static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2871 struct super_block *sb, struct inode *dir,
2872 umode_t mode, dev_t dev, unsigned long flags)
2873{
2874 int err;
2875 struct inode *inode;
2876
2877 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2878 if (IS_ERR(inode))
2879 return inode;
2880
2881 err = dquot_initialize(inode);
2882 if (err)
2883 goto errout;
2884
2885 err = dquot_alloc_inode(inode);
2886 if (err) {
2887 dquot_drop(inode);
2888 goto errout;
2889 }
2890 return inode;
2891
2892errout:
2893 inode->i_flags |= S_NOQUOTA;
2894 iput(inode);
2895 return ERR_PTR(err);
2896}
2897#else
2898static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2899 struct super_block *sb, struct inode *dir,
2900 umode_t mode, dev_t dev, unsigned long flags)
2901{
2902 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2903}
2904#endif /* CONFIG_TMPFS_QUOTA */
2905
Axel Rasmussen3460f6e2021-06-30 18:49:17 -07002906#ifdef CONFIG_USERFAULTFD
Axel Rasmussen61c50042023-03-14 15:12:48 -07002907int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
Axel Rasmussen3460f6e2021-06-30 18:49:17 -07002908 struct vm_area_struct *dst_vma,
2909 unsigned long dst_addr,
2910 unsigned long src_addr,
Axel Rasmussend9712932023-03-14 15:12:49 -07002911 uffd_flags_t flags,
ZhangPengd7be6d7e2023-04-10 21:39:32 +08002912 struct folio **foliop)
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002913{
2914 struct inode *inode = file_inode(dst_vma->vm_file);
2915 struct shmem_inode_info *info = SHMEM_I(inode);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002916 struct address_space *mapping = inode->i_mapping;
2917 gfp_t gfp = mapping_gfp_mask(mapping);
2918 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002919 void *page_kaddr;
Matthew Wilcox (Oracle)b7dd44a2022-05-12 20:23:04 -07002920 struct folio *folio;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002921 int ret;
Axel Rasmussen3460f6e2021-06-30 18:49:17 -07002922 pgoff_t max_off;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002923
Hugh Dickins4199f512023-09-29 20:30:03 -07002924 if (shmem_inode_acct_blocks(inode, 1)) {
Axel Rasmussen7ed9d232021-05-14 17:27:19 -07002925 /*
2926 * We may have got a page, returned -ENOENT triggering a retry,
2927 * and now we find ourselves with -ENOMEM. Release the page, to
2928 * avoid a BUG_ON in our caller.
2929 */
ZhangPengd7be6d7e2023-04-10 21:39:32 +08002930 if (unlikely(*foliop)) {
2931 folio_put(*foliop);
2932 *foliop = NULL;
Axel Rasmussen7ed9d232021-05-14 17:27:19 -07002933 }
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07002934 return -ENOMEM;
Axel Rasmussen7ed9d232021-05-14 17:27:19 -07002935 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002936
ZhangPengd7be6d7e2023-04-10 21:39:32 +08002937 if (!*foliop) {
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07002938 ret = -ENOMEM;
Kefeng Wang6f775462024-05-15 15:07:09 +08002939 folio = shmem_alloc_folio(gfp, 0, info, pgoff);
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002940 if (!folio)
Mike Rapoport0f079692017-09-06 16:22:59 -07002941 goto out_unacct_blocks;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002942
Axel Rasmussend9712932023-03-14 15:12:49 -07002943 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002944 page_kaddr = kmap_local_folio(folio, 0);
Ira Weiny5dc21f02022-10-25 15:01:08 -07002945 /*
2946 * The read mmap_lock is held here. Despite the
2947 * mmap_lock being read recursive a deadlock is still
2948 * possible if a writer has taken a lock. For example:
2949 *
2950 * process A thread 1 takes read lock on own mmap_lock
2951 * process A thread 2 calls mmap, blocks taking write lock
2952 * process B thread 1 takes page fault, read lock on own mmap lock
2953 * process B thread 2 calls mmap, blocks taking write lock
2954 * process A thread 1 blocks taking read lock on process B
2955 * process B thread 1 blocks taking read lock on process A
2956 *
2957 * Disable page faults to prevent potential deadlock
2958 * and retry the copy outside the mmap_lock.
2959 */
2960 pagefault_disable();
Mike Rapoport8d103962017-09-06 16:23:02 -07002961 ret = copy_from_user(page_kaddr,
2962 (const void __user *)src_addr,
2963 PAGE_SIZE);
Ira Weiny5dc21f02022-10-25 15:01:08 -07002964 pagefault_enable();
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002965 kunmap_local(page_kaddr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002966
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002967 /* fallback to copy_from_user outside mmap_lock */
Mike Rapoport8d103962017-09-06 16:23:02 -07002968 if (unlikely(ret)) {
ZhangPengd7be6d7e2023-04-10 21:39:32 +08002969 *foliop = folio;
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07002970 ret = -ENOENT;
Mike Rapoport8d103962017-09-06 16:23:02 -07002971 /* don't free the page */
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07002972 goto out_unacct_blocks;
Mike Rapoport8d103962017-09-06 16:23:02 -07002973 }
Muchun Song19b482c2022-03-22 14:42:05 -07002974
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002975 flush_dcache_folio(folio);
Axel Rasmussen3460f6e2021-06-30 18:49:17 -07002976 } else { /* ZEROPAGE */
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002977 clear_user_highpage(&folio->page, dst_addr);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002978 }
2979 } else {
ZhangPengd7be6d7e2023-04-10 21:39:32 +08002980 folio = *foliop;
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002981 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
ZhangPengd7be6d7e2023-04-10 21:39:32 +08002982 *foliop = NULL;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002983 }
2984
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01002985 VM_BUG_ON(folio_test_locked(folio));
2986 VM_BUG_ON(folio_test_swapbacked(folio));
2987 __folio_set_locked(folio);
2988 __folio_set_swapbacked(folio);
2989 __folio_mark_uptodate(folio);
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002990
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002991 ret = -EFAULT;
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002992 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
Axel Rasmussen3460f6e2021-06-30 18:49:17 -07002993 if (unlikely(pgoff >= max_off))
Andrea Arcangelie2a50c12018-11-30 14:09:37 -08002994 goto out_release;
2995
Hugh Dickins054a9f72023-09-29 20:31:27 -07002996 ret = mem_cgroup_charge(folio, dst_vma->vm_mm, gfp);
2997 if (ret)
2998 goto out_release;
2999 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08003000 if (ret)
3001 goto out_release;
3002
Axel Rasmussen61c50042023-03-14 15:12:48 -07003003 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
Axel Rasmussend9712932023-03-14 15:12:49 -07003004 &folio->page, true, flags);
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07003005 if (ret)
3006 goto out_delete_from_cache;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08003007
Hugh Dickins3c1b7522023-08-03 22:46:11 -07003008 shmem_recalc_inode(inode, 1, 0);
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01003009 folio_unlock(folio);
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07003010 return 0;
3011out_delete_from_cache:
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01003012 filemap_remove_folio(folio);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08003013out_release:
Matthew Wilcox (Oracle)7a7256d2022-09-02 20:46:13 +01003014 folio_unlock(folio);
3015 folio_put(folio);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08003016out_unacct_blocks:
Mike Rapoport0f079692017-09-06 16:22:59 -07003017 shmem_inode_unacct_blocks(inode, 1);
Axel Rasmussen7d64ae32021-06-30 18:49:31 -07003018 return ret;
Mike Rapoport4c27fe42017-02-22 15:43:25 -08003019}
Axel Rasmussen3460f6e2021-06-30 18:49:17 -07003020#endif /* CONFIG_USERFAULTFD */
Mike Rapoport8d103962017-09-06 16:23:02 -07003021
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003023static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003024static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026static int
Nick Piggin800d15a2007-10-16 01:25:03 -07003027shmem_write_begin(struct file *file, struct address_space *mapping,
Matthew Wilcox (Oracle)9d6b0cd2022-02-22 14:31:43 -05003028 loff_t pos, unsigned len,
Matthew Wilcox (Oracle)1da86612024-07-15 14:24:01 -04003029 struct folio **foliop, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030{
Nick Piggin800d15a2007-10-16 01:25:03 -07003031 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07003032 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003033 pgoff_t index = pos >> PAGE_SHIFT;
Matthew Wilcox (Oracle)eff1f902022-09-02 20:46:22 +01003034 struct folio *folio;
Yang Shia7605422022-01-14 14:05:19 -08003035 int ret = 0;
David Herrmann40e041a2014-08-08 14:25:27 -07003036
Jan Kara96087032021-04-12 15:50:21 +02003037 /* i_rwsem is held by caller */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08003038 if (unlikely(info->seals & (F_SEAL_GROW |
3039 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
3040 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
David Herrmann40e041a2014-08-08 14:25:27 -07003041 return -EPERM;
3042 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
3043 return -EPERM;
3044 }
3045
Rik van Riele1e4cfd2024-09-03 11:19:28 -04003046 ret = shmem_get_folio(inode, index, pos + len, &folio, SGP_WRITE);
Yang Shia7605422022-01-14 14:05:19 -08003047 if (ret)
3048 return ret;
3049
Matthew Wilcox (Oracle)1da86612024-07-15 14:24:01 -04003050 if (folio_test_hwpoison(folio) ||
3051 (folio_test_large(folio) && folio_test_has_hwpoisoned(folio))) {
Matthew Wilcox (Oracle)eff1f902022-09-02 20:46:22 +01003052 folio_unlock(folio);
3053 folio_put(folio);
Yang Shia7605422022-01-14 14:05:19 -08003054 return -EIO;
3055 }
3056
Matthew Wilcox (Oracle)1da86612024-07-15 14:24:01 -04003057 *foliop = folio;
Yang Shia7605422022-01-14 14:05:19 -08003058 return 0;
Nick Piggin800d15a2007-10-16 01:25:03 -07003059}
3060
3061static int
3062shmem_write_end(struct file *file, struct address_space *mapping,
3063 loff_t pos, unsigned len, unsigned copied,
Matthew Wilcox (Oracle)a2258002024-07-10 15:45:32 -04003064 struct folio *folio, void *fsdata)
Nick Piggin800d15a2007-10-16 01:25:03 -07003065{
3066 struct inode *inode = mapping->host;
3067
Hugh Dickinsd3602442008-02-04 22:28:44 -08003068 if (pos + copied > inode->i_size)
3069 i_size_write(inode, pos + copied);
3070
Matthew Wilcox (Oracle)69bbb872023-01-12 13:10:31 +00003071 if (!folio_test_uptodate(folio)) {
3072 if (copied < folio_size(folio)) {
3073 size_t from = offset_in_folio(folio, pos);
3074 folio_zero_segments(folio, 0, from,
3075 from + copied, folio_size(folio));
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07003076 }
Matthew Wilcox (Oracle)69bbb872023-01-12 13:10:31 +00003077 folio_mark_uptodate(folio);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003078 }
Matthew Wilcox (Oracle)69bbb872023-01-12 13:10:31 +00003079 folio_mark_dirty(folio);
3080 folio_unlock(folio);
3081 folio_put(folio);
Nick Piggin800d15a2007-10-16 01:25:03 -07003082
Nick Piggin800d15a2007-10-16 01:25:03 -07003083 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084}
3085
Al Viro2ba5bbe2014-04-02 20:00:02 -04003086static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087{
Al Viro6e58e792014-02-03 17:07:03 -05003088 struct file *file = iocb->ki_filp;
3089 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003091 pgoff_t index;
3092 unsigned long offset;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02003093 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05003094 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05003095 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08003096
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003097 index = *ppos >> PAGE_SHIFT;
3098 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099
3100 for (;;) {
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003101 struct folio *folio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003103 pgoff_t end_index;
3104 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 loff_t i_size = i_size_read(inode);
3106
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003107 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 if (index > end_index)
3109 break;
3110 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003111 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 if (nr <= offset)
3113 break;
3114 }
3115
Rik van Riele1e4cfd2024-09-03 11:19:28 -04003116 error = shmem_get_folio(inode, index, 0, &folio, SGP_READ);
Al Viro6e58e792014-02-03 17:07:03 -05003117 if (error) {
3118 if (error == -EINVAL)
3119 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 break;
3121 }
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003122 if (folio) {
3123 folio_unlock(folio);
Yang Shia7605422022-01-14 14:05:19 -08003124
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003125 page = folio_file_page(folio, index);
Yang Shia7605422022-01-14 14:05:19 -08003126 if (PageHWPoison(page)) {
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003127 folio_put(folio);
Yang Shia7605422022-01-14 14:05:19 -08003128 error = -EIO;
3129 break;
3130 }
Hugh Dickins75edd342016-05-19 17:12:44 -07003131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132
3133 /*
3134 * We must evaluate after, since reads (unlike writes)
Jan Kara96087032021-04-12 15:50:21 +02003135 * are called without i_rwsem protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003137 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003139 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003141 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 if (nr <= offset) {
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003143 if (folio)
3144 folio_put(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 break;
3146 }
3147 }
3148 nr -= offset;
3149
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003150 if (folio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 /*
3152 * If users can be writing to this page using arbitrary
3153 * virtual addresses, take care about potential aliasing
3154 * before reading the page on the kernel side.
3155 */
3156 if (mapping_writably_mapped(mapping))
3157 flush_dcache_page(page);
3158 /*
3159 * Mark the page accessed if we read the beginning.
3160 */
3161 if (!offset)
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003162 folio_mark_accessed(folio);
Hugh Dickins1bdec44b2022-04-14 19:13:27 -07003163 /*
3164 * Ok, we have the page, and it's up-to-date, so
3165 * now we can copy it to user space...
3166 */
3167 ret = copy_page_to_iter(page, offset, nr, to);
Matthew Wilcox (Oracle)4601e2f2022-09-02 20:46:23 +01003168 folio_put(folio);
Hugh Dickins1bdec44b2022-04-14 19:13:27 -07003169
Al Virofcb14cb2022-05-22 14:59:25 -04003170 } else if (user_backed_iter(to)) {
Hugh Dickins1bdec44b2022-04-14 19:13:27 -07003171 /*
3172 * Copy to user tends to be so well optimized, but
3173 * clear_user() not so much, that it is noticeably
3174 * faster to copy the zero page instead of clearing.
3175 */
3176 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
Nick Pigginb5810032005-10-29 18:16:12 -07003177 } else {
Hugh Dickins1bdec44b2022-04-14 19:13:27 -07003178 /*
3179 * But submitting the same page twice in a row to
3180 * splice() - or others? - can result in confusion:
3181 * so don't attempt that optimization on pipes etc.
3182 */
3183 ret = iov_iter_zero(nr, to);
Nick Pigginb5810032005-10-29 18:16:12 -07003184 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185
Al Viro6e58e792014-02-03 17:07:03 -05003186 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003188 index += offset >> PAGE_SHIFT;
3189 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190
Al Viro2ba5bbe2014-04-02 20:00:02 -04003191 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 break;
Al Viro6e58e792014-02-03 17:07:03 -05003193 if (ret < nr) {
3194 error = -EFAULT;
3195 break;
3196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 cond_resched();
3198 }
3199
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003200 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05003201 file_accessed(file);
3202 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203}
3204
Hugh Dickinse88e0d32023-08-10 23:27:07 -07003205static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3206{
3207 struct file *file = iocb->ki_filp;
3208 struct inode *inode = file->f_mapping->host;
3209 ssize_t ret;
3210
3211 inode_lock(inode);
3212 ret = generic_write_checks(iocb, from);
3213 if (ret <= 0)
3214 goto unlock;
3215 ret = file_remove_privs(file);
3216 if (ret)
3217 goto unlock;
3218 ret = file_update_time(file);
3219 if (ret)
3220 goto unlock;
3221 ret = generic_perform_write(iocb, from);
3222unlock:
3223 inode_unlock(inode);
3224 return ret;
3225}
3226
David Howellsbd194b12023-05-22 14:49:56 +01003227static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
3228 struct pipe_buffer *buf)
3229{
3230 return true;
3231}
3232
3233static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
3234 struct pipe_buffer *buf)
3235{
3236}
3237
3238static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
3239 struct pipe_buffer *buf)
3240{
3241 return false;
3242}
3243
3244static const struct pipe_buf_operations zero_pipe_buf_ops = {
3245 .release = zero_pipe_buf_release,
3246 .try_steal = zero_pipe_buf_try_steal,
3247 .get = zero_pipe_buf_get,
3248};
3249
3250static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
3251 loff_t fpos, size_t size)
3252{
3253 size_t offset = fpos & ~PAGE_MASK;
3254
3255 size = min_t(size_t, size, PAGE_SIZE - offset);
3256
3257 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
3258 struct pipe_buffer *buf = pipe_head_buf(pipe);
3259
3260 *buf = (struct pipe_buffer) {
3261 .ops = &zero_pipe_buf_ops,
3262 .page = ZERO_PAGE(0),
3263 .offset = offset,
3264 .len = size,
3265 };
3266 pipe->head++;
3267 }
3268
3269 return size;
3270}
3271
3272static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
3273 struct pipe_inode_info *pipe,
3274 size_t len, unsigned int flags)
3275{
3276 struct inode *inode = file_inode(in);
3277 struct address_space *mapping = inode->i_mapping;
3278 struct folio *folio = NULL;
3279 size_t total_spliced = 0, used, npages, n, part;
3280 loff_t isize;
3281 int error = 0;
3282
3283 /* Work out how much data we can actually add into the pipe */
3284 used = pipe_occupancy(pipe->head, pipe->tail);
3285 npages = max_t(ssize_t, pipe->max_usage - used, 0);
3286 len = min_t(size_t, len, npages * PAGE_SIZE);
3287
3288 do {
3289 if (*ppos >= i_size_read(inode))
3290 break;
3291
Rik van Riele1e4cfd2024-09-03 11:19:28 -04003292 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, 0, &folio,
Hugh Dickinsfa598952023-07-23 14:05:54 -07003293 SGP_READ);
David Howellsbd194b12023-05-22 14:49:56 +01003294 if (error) {
3295 if (error == -EINVAL)
3296 error = 0;
3297 break;
3298 }
3299 if (folio) {
3300 folio_unlock(folio);
3301
Hugh Dickinsfa598952023-07-23 14:05:54 -07003302 if (folio_test_hwpoison(folio) ||
3303 (folio_test_large(folio) &&
3304 folio_test_has_hwpoisoned(folio))) {
David Howellsbd194b12023-05-22 14:49:56 +01003305 error = -EIO;
3306 break;
3307 }
3308 }
3309
3310 /*
3311 * i_size must be checked after we know the pages are Uptodate.
3312 *
3313 * Checking i_size after the check allows us to calculate
3314 * the correct value for "nr", which means the zero-filled
3315 * part of the page is not copied back to userspace (unless
3316 * another truncate extends the file - this is desired though).
3317 */
3318 isize = i_size_read(inode);
3319 if (unlikely(*ppos >= isize))
3320 break;
3321 part = min_t(loff_t, isize - *ppos, len);
3322
3323 if (folio) {
3324 /*
3325 * If users can be writing to this page using arbitrary
3326 * virtual addresses, take care about potential aliasing
3327 * before reading the page on the kernel side.
3328 */
3329 if (mapping_writably_mapped(mapping))
3330 flush_dcache_folio(folio);
3331 folio_mark_accessed(folio);
3332 /*
3333 * Ok, we have the page, and it's up-to-date, so we can
3334 * now splice it into the pipe.
3335 */
3336 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
3337 folio_put(folio);
3338 folio = NULL;
3339 } else {
Hugh Dickinsfa598952023-07-23 14:05:54 -07003340 n = splice_zeropage_into_pipe(pipe, *ppos, part);
David Howellsbd194b12023-05-22 14:49:56 +01003341 }
3342
3343 if (!n)
3344 break;
3345 len -= n;
3346 total_spliced += n;
3347 *ppos += n;
3348 in->f_ra.prev_pos = *ppos;
3349 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
3350 break;
3351
3352 cond_resched();
3353 } while (len);
3354
3355 if (folio)
3356 folio_put(folio);
3357
3358 file_accessed(in);
3359 return total_spliced ? total_spliced : error;
3360}
3361
Andrew Morton965c8e52012-12-17 15:59:39 -08003362static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003363{
3364 struct address_space *mapping = file->f_mapping;
3365 struct inode *inode = mapping->host;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003366
Andrew Morton965c8e52012-12-17 15:59:39 -08003367 if (whence != SEEK_DATA && whence != SEEK_HOLE)
3368 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003369 MAX_LFS_FILESIZE, i_size_read(inode));
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003370 if (offset < 0)
3371 return -ENXIO;
3372
Al Viro59551022016-01-22 15:40:57 -05003373 inode_lock(inode);
Jan Kara96087032021-04-12 15:50:21 +02003374 /* We're holding i_rwsem so we can access i_size directly */
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003375 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
Hugh Dickins387aae62013-08-04 11:30:25 -07003376 if (offset >= 0)
3377 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05003378 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003379 return offset;
3380}
3381
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003382static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3383 loff_t len)
3384{
Al Viro496ad9a2013-01-23 17:07:38 -05003385 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07003387 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07003388 struct shmem_falloc shmem_falloc;
Hugh Dickinsd144bf62021-09-02 14:54:21 -07003389 pgoff_t start, index, end, undo_fallocend;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003390 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003391
Hugh Dickins13ace4d2014-06-23 13:22:03 -07003392 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3393 return -EOPNOTSUPP;
3394
Al Viro59551022016-01-22 15:40:57 -05003395 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003396
3397 if (mode & FALLOC_FL_PUNCH_HOLE) {
3398 struct address_space *mapping = file->f_mapping;
3399 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3400 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07003401 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003402
Jan Kara96087032021-04-12 15:50:21 +02003403 /* protected by i_rwsem */
Joel Fernandes (Google)ab3948f2019-03-05 15:47:54 -08003404 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
David Herrmann40e041a2014-08-08 14:25:27 -07003405 error = -EPERM;
3406 goto out;
3407 }
3408
Hugh Dickins8e205f72014-07-23 14:00:10 -07003409 shmem_falloc.waitq = &shmem_falloc_waitq;
Chen Junaa71ecd2019-11-30 17:58:11 -08003410 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07003411 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3412 spin_lock(&inode->i_lock);
3413 inode->i_private = &shmem_falloc;
3414 spin_unlock(&inode->i_lock);
3415
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003416 if ((u64)unmap_end > (u64)unmap_start)
3417 unmap_mapping_range(mapping, unmap_start,
3418 1 + unmap_end - unmap_start, 0);
3419 shmem_truncate_range(inode, offset, offset + len - 1);
3420 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07003421
3422 spin_lock(&inode->i_lock);
3423 inode->i_private = NULL;
3424 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02003425 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07003426 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003427 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07003428 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003429 }
3430
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003431 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3432 error = inode_newsize_ok(inode, offset + len);
3433 if (error)
3434 goto out;
3435
David Herrmann40e041a2014-08-08 14:25:27 -07003436 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3437 error = -EPERM;
3438 goto out;
3439 }
3440
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003441 start = offset >> PAGE_SHIFT;
3442 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003443 /* Try to avoid a swapstorm if len is impossible to satisfy */
3444 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3445 error = -ENOSPC;
3446 goto out;
3447 }
3448
Hugh Dickins8e205f72014-07-23 14:00:10 -07003449 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07003450 shmem_falloc.start = start;
3451 shmem_falloc.next = start;
3452 shmem_falloc.nr_falloced = 0;
3453 shmem_falloc.nr_unswapped = 0;
3454 spin_lock(&inode->i_lock);
3455 inode->i_private = &shmem_falloc;
3456 spin_unlock(&inode->i_lock);
3457
Hugh Dickinsd144bf62021-09-02 14:54:21 -07003458 /*
3459 * info->fallocend is only relevant when huge pages might be
3460 * involved: to prevent split_huge_page() freeing fallocated
3461 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3462 */
3463 undo_fallocend = info->fallocend;
3464 if (info->fallocend < end)
3465 info->fallocend = end;
3466
Hugh Dickins050dcb52021-09-02 14:54:18 -07003467 for (index = start; index < end; ) {
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003468 struct folio *folio;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003469
3470 /*
Mikulas Patockaca86a5d2024-05-16 00:10:44 +02003471 * Check for fatal signal so that we abort early in OOM
3472 * situations. We don't want to abort in case of non-fatal
3473 * signals as large fallocate can take noticeable time and
3474 * e.g. periodic timers may result in fallocate constantly
3475 * restarting.
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003476 */
Mikulas Patockaca86a5d2024-05-16 00:10:44 +02003477 if (fatal_signal_pending(current))
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003478 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07003479 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3480 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003481 else
Rik van Riele1e4cfd2024-09-03 11:19:28 -04003482 error = shmem_get_folio(inode, index, offset + len,
3483 &folio, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003484 if (error) {
Hugh Dickinsd144bf62021-09-02 14:54:21 -07003485 info->fallocend = undo_fallocend;
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003486 /* Remove the !uptodate folios we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07003487 if (index > start) {
3488 shmem_undo_range(inode,
3489 (loff_t)start << PAGE_SHIFT,
3490 ((loff_t)index << PAGE_SHIFT) - 1, true);
3491 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07003492 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003493 }
3494
Hugh Dickins050dcb52021-09-02 14:54:18 -07003495 /*
3496 * Here is a more important optimization than it appears:
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003497 * a second SGP_FALLOC on the same large folio will clear it,
3498 * making it uptodate and un-undoable if we fail later.
Hugh Dickins050dcb52021-09-02 14:54:18 -07003499 */
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003500 index = folio_next_index(folio);
3501 /* Beware 32-bit wraparound */
3502 if (!index)
3503 index--;
Hugh Dickins050dcb52021-09-02 14:54:18 -07003504
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003505 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07003506 * Inform shmem_writepage() how far we have reached.
3507 * No need for lock or barrier: we have the page lock.
3508 */
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003509 if (!folio_test_uptodate(folio))
Hugh Dickins050dcb52021-09-02 14:54:18 -07003510 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3511 shmem_falloc.next = index;
Hugh Dickins1aac1402012-05-29 15:06:42 -07003512
3513 /*
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003514 * If !uptodate, leave it that way so that freeable folios
Hugh Dickins1635f6a2012-05-29 15:06:42 -07003515 * can be recognized if we need to rollback on error later.
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003516 * But mark it dirty so that memory pressure will swap rather
3517 * than free the folios we are allocating (and SGP_CACHE folios
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003518 * might still be clean: we now need to mark those dirty too).
3519 */
Matthew Wilcox (Oracle)b0802b22022-09-02 20:46:24 +01003520 folio_mark_dirty(folio);
3521 folio_unlock(folio);
3522 folio_put(folio);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003523 cond_resched();
3524 }
3525
3526 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3527 i_size_write(inode, offset + len);
Hugh Dickins1aac1402012-05-29 15:06:42 -07003528undone:
3529 spin_lock(&inode->i_lock);
3530 inode->i_private = NULL;
3531 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07003532out:
Hugh Dickins15f242b2022-08-10 21:55:36 -07003533 if (!error)
3534 file_modified(file);
Al Viro59551022016-01-22 15:40:57 -05003535 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003536 return error;
3537}
3538
David Howells726c3342006-06-23 02:02:58 -07003539static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540{
David Howells726c3342006-06-23 02:02:58 -07003541 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542
3543 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003544 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003546 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003548 buf->f_bavail =
3549 buf->f_bfree = sbinfo->max_blocks -
3550 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003551 }
3552 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553 buf->f_files = sbinfo->max_inodes;
Hugh Dickinse07c4692023-08-08 21:32:21 -07003554 buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 }
3556 /* else leave those fields 0 like simple_statfs */
Amir Goldstein59cda492021-03-22 19:39:44 +02003557
3558 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3559
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 return 0;
3561}
3562
3563/*
3564 * File creation. Allocate an inode, and we're done..
3565 */
3566static int
Christian Brauner5ebb29b2023-01-13 12:49:16 +01003567shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
Christian Brauner549c7292021-01-21 14:19:43 +01003568 struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00003570 struct inode *inode;
Carlos Maiolino71480662023-07-25 16:45:05 +02003571 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003573 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
Carlos Maiolino71480662023-07-25 16:45:05 +02003574 if (IS_ERR(inode))
3575 return PTR_ERR(inode);
Mimi Zohar37ec43c2013-04-14 09:21:47 -04003576
Carlos Maiolino71480662023-07-25 16:45:05 +02003577 error = simple_acl_create(dir, inode);
3578 if (error)
3579 goto out_iput;
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003580 error = security_inode_init_security(inode, dir, &dentry->d_name,
Carlos Maiolino71480662023-07-25 16:45:05 +02003581 shmem_initxattrs, NULL);
3582 if (error && error != -EOPNOTSUPP)
3583 goto out_iput;
3584
Chuck Levera2e45952023-06-30 13:49:03 -04003585 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3586 if (error)
3587 goto out_iput;
3588
Carlos Maiolino71480662023-07-25 16:45:05 +02003589 dir->i_size += BOGO_DIRENT_SIZE;
Jeff Laytoncf2766b2023-10-04 14:53:07 -04003590 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
Carlos Maiolino71480662023-07-25 16:45:05 +02003591 inode_inc_iversion(dir);
3592 d_instantiate(dentry, inode);
3593 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 return error;
Carlos Maiolino71480662023-07-25 16:45:05 +02003595
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003596out_iput:
3597 iput(inode);
3598 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599}
3600
Al Viro60545d02013-06-07 01:20:27 -04003601static int
Christian Brauner011e2b72023-01-13 12:49:18 +01003602shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
Miklos Szeredi863f1442022-09-24 07:00:00 +02003603 struct file *file, umode_t mode)
Al Viro60545d02013-06-07 01:20:27 -04003604{
3605 struct inode *inode;
Carlos Maiolino71480662023-07-25 16:45:05 +02003606 int error;
Al Viro60545d02013-06-07 01:20:27 -04003607
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003608 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
Carlos Maiolino71480662023-07-25 16:45:05 +02003609 if (IS_ERR(inode)) {
3610 error = PTR_ERR(inode);
3611 goto err_out;
Al Viro60545d02013-06-07 01:20:27 -04003612 }
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003613 error = security_inode_init_security(inode, dir, NULL,
Carlos Maiolino71480662023-07-25 16:45:05 +02003614 shmem_initxattrs, NULL);
3615 if (error && error != -EOPNOTSUPP)
3616 goto out_iput;
3617 error = simple_acl_create(dir, inode);
3618 if (error)
3619 goto out_iput;
3620 d_tmpfile(file, inode);
3621
3622err_out:
Miklos Szeredi863f1442022-09-24 07:00:00 +02003623 return finish_open_simple(file, error);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003624out_iput:
3625 iput(inode);
3626 return error;
Al Viro60545d02013-06-07 01:20:27 -04003627}
3628
Christian Braunerc54bd912023-01-13 12:49:15 +01003629static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
Christian Brauner549c7292021-01-21 14:19:43 +01003630 struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631{
3632 int error;
3633
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003634 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3635 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07003637 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 return 0;
3639}
3640
Christian Brauner6c960e62023-01-13 12:49:13 +01003641static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
Christian Brauner549c7292021-01-21 14:19:43 +01003642 struct dentry *dentry, umode_t mode, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643{
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003644 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645}
3646
3647/*
3648 * Link a file..
3649 */
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003650static int shmem_link(struct dentry *old_dentry, struct inode *dir,
3651 struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652{
David Howells75c3cfa2015-03-17 22:26:12 +00003653 struct inode *inode = d_inode(old_dentry);
Darrick J. Wong29b00e62019-02-22 22:35:32 -08003654 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655
3656 /*
3657 * No ordinary (disk based) filesystem counts links as inodes;
3658 * but each new link needs a new dentry, pinning lowmem, and
3659 * tmpfs dentries cannot be pruned until they are unlinked.
Darrick J. Wong1062af92019-02-21 08:48:09 -08003660 * But if an O_TMPFILE file is linked into the tmpfs, the
3661 * first link must skip that, to get the accounting right.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 */
Darrick J. Wong1062af92019-02-21 08:48:09 -08003663 if (inode->i_nlink) {
Chris Downe809d5f2020-08-06 23:20:20 -07003664 ret = shmem_reserve_inode(inode->i_sb, NULL);
Darrick J. Wong1062af92019-02-21 08:48:09 -08003665 if (ret)
3666 goto out;
3667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668
Chuck Levera2e45952023-06-30 13:49:03 -04003669 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3670 if (ret) {
3671 if (inode->i_nlink)
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003672 shmem_free_inode(inode->i_sb, 0);
Chuck Levera2e45952023-06-30 13:49:03 -04003673 goto out;
3674 }
3675
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 dir->i_size += BOGO_DIRENT_SIZE;
Jeff Laytoncf2766b2023-10-04 14:53:07 -04003677 inode_set_mtime_to_ts(dir,
3678 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
Jeff Layton36f05ca2022-09-09 09:00:31 -04003679 inode_inc_iversion(dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003680 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04003681 ihold(inode); /* New dentry reference */
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003682 dget(dentry); /* Extra pinning count for the created dentry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003684out:
3685 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686}
3687
3688static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3689{
David Howells75c3cfa2015-03-17 22:26:12 +00003690 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003692 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003693 shmem_free_inode(inode->i_sb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694
Chuck Levera2e45952023-06-30 13:49:03 -04003695 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
3697 dir->i_size -= BOGO_DIRENT_SIZE;
Jeff Laytoncf2766b2023-10-04 14:53:07 -04003698 inode_set_mtime_to_ts(dir,
3699 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
Jeff Layton36f05ca2022-09-09 09:00:31 -04003700 inode_inc_iversion(dir);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003701 drop_nlink(inode);
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003702 dput(dentry); /* Undo the count from "create" - does all the work */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 return 0;
3704}
3705
3706static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3707{
Chuck Leverecba88a2024-02-17 15:23:54 -05003708 if (!simple_offset_empty(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 return -ENOTEMPTY;
3710
David Howells75c3cfa2015-03-17 22:26:12 +00003711 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003712 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 return shmem_unlink(dir, dentry);
3714}
3715
Christian Braunere18275a2023-01-13 12:49:17 +01003716static int shmem_whiteout(struct mnt_idmap *idmap,
Christian Brauner549c7292021-01-21 14:19:43 +01003717 struct inode *old_dir, struct dentry *old_dentry)
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003718{
3719 struct dentry *whiteout;
3720 int error;
3721
3722 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3723 if (!whiteout)
3724 return -ENOMEM;
3725
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003726 error = shmem_mknod(idmap, old_dir, whiteout,
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003727 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3728 dput(whiteout);
3729 if (error)
3730 return error;
3731
3732 /*
3733 * Cheat and hash the whiteout while the old dentry is still in
3734 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3735 *
3736 * d_lookup() will consistently find one of them at this point,
3737 * not sure which one, but that isn't even important.
3738 */
3739 d_rehash(whiteout);
3740 return 0;
3741}
3742
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743/*
3744 * The VFS layer already does all the dentry stuff for rename,
3745 * we just have to decrement the usage count for the target if
3746 * it exists so that the VFS layer correctly free's it when it
3747 * gets overwritten.
3748 */
Christian Braunere18275a2023-01-13 12:49:17 +01003749static int shmem_rename2(struct mnt_idmap *idmap,
Christian Brauner549c7292021-01-21 14:19:43 +01003750 struct inode *old_dir, struct dentry *old_dentry,
3751 struct inode *new_dir, struct dentry *new_dentry,
3752 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753{
David Howells75c3cfa2015-03-17 22:26:12 +00003754 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 int they_are_dirs = S_ISDIR(inode->i_mode);
Chuck Levera2e45952023-06-30 13:49:03 -04003756 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003758 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003759 return -EINVAL;
3760
Miklos Szeredi37456772014-07-23 15:15:34 +02003761 if (flags & RENAME_EXCHANGE)
Chuck Levera2e45952023-06-30 13:49:03 -04003762 return simple_offset_rename_exchange(old_dir, old_dentry,
3763 new_dir, new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003764
Chuck Leverecba88a2024-02-17 15:23:54 -05003765 if (!simple_offset_empty(new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 return -ENOTEMPTY;
3767
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003768 if (flags & RENAME_WHITEOUT) {
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003769 error = shmem_whiteout(idmap, old_dir, old_dentry);
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003770 if (error)
3771 return error;
3772 }
3773
Chuck Lever5a1a25b2024-04-15 11:20:55 -04003774 error = simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry);
Chuck Levera2e45952023-06-30 13:49:03 -04003775 if (error)
3776 return error;
3777
David Howells75c3cfa2015-03-17 22:26:12 +00003778 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003780 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003781 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003782 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003785 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003786 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 }
3788
3789 old_dir->i_size -= BOGO_DIRENT_SIZE;
3790 new_dir->i_size += BOGO_DIRENT_SIZE;
Jeff Layton944d0d9de2023-07-05 15:00:36 -04003791 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
Jeff Layton36f05ca2022-09-09 09:00:31 -04003792 inode_inc_iversion(old_dir);
3793 inode_inc_iversion(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 return 0;
3795}
3796
Christian Brauner7a77db92023-01-13 12:49:14 +01003797static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
Christian Brauner549c7292021-01-21 14:19:43 +01003798 struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799{
3800 int error;
3801 int len;
3802 struct inode *inode;
Matthew Wilcox (Oracle)7ad04142022-09-02 20:46:25 +01003803 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
3805 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003806 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 return -ENAMETOOLONG;
3808
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01003809 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
Joe Perches0825a6f2018-06-14 15:27:58 -07003810 VM_NORESERVE);
Carlos Maiolino71480662023-07-25 16:45:05 +02003811 if (IS_ERR(inode))
3812 return PTR_ERR(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003814 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003815 shmem_initxattrs, NULL);
Chuck Lever23a31d872023-06-30 13:48:56 -04003816 if (error && error != -EOPNOTSUPP)
3817 goto out_iput;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003818
Chuck Levera2e45952023-06-30 13:49:03 -04003819 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3820 if (error)
3821 goto out_iput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003824 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003825 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3826 if (!inode->i_link) {
Chuck Lever23a31d872023-06-30 13:48:56 -04003827 error = -ENOMEM;
Chuck Levera2e45952023-06-30 13:49:03 -04003828 goto out_remove_offset;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003829 }
3830 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003832 inode_nohighmem(inode);
Christoph Hellwige11381d2024-02-19 07:27:11 +01003833 inode->i_mapping->a_ops = &shmem_aops;
Rik van Riele1e4cfd2024-09-03 11:19:28 -04003834 error = shmem_get_folio(inode, 0, 0, &folio, SGP_WRITE);
Chuck Lever23a31d872023-06-30 13:48:56 -04003835 if (error)
Chuck Levera2e45952023-06-30 13:49:03 -04003836 goto out_remove_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837 inode->i_op = &shmem_symlink_inode_operations;
Matthew Wilcox (Oracle)7ad04142022-09-02 20:46:25 +01003838 memcpy(folio_address(folio), symname, len);
3839 folio_mark_uptodate(folio);
3840 folio_mark_dirty(folio);
3841 folio_unlock(folio);
3842 folio_put(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844 dir->i_size += BOGO_DIRENT_SIZE;
Jeff Laytoncf2766b2023-10-04 14:53:07 -04003845 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
Jeff Layton36f05ca2022-09-09 09:00:31 -04003846 inode_inc_iversion(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847 d_instantiate(dentry, inode);
3848 dget(dentry);
3849 return 0;
Chuck Levera2e45952023-06-30 13:49:03 -04003850
3851out_remove_offset:
3852 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
Chuck Lever23a31d872023-06-30 13:48:56 -04003853out_iput:
3854 iput(inode);
3855 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856}
3857
Al Virofceef392015-12-29 15:58:39 -05003858static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859{
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003860 folio_mark_accessed(arg);
3861 folio_put(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862}
3863
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003864static const char *shmem_get_link(struct dentry *dentry, struct inode *inode,
Al Virofceef392015-12-29 15:58:39 -05003865 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866{
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003867 struct folio *folio = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003868 int error;
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003869
Al Viro6a6c9902015-11-17 10:54:32 -05003870 if (!dentry) {
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003871 folio = filemap_get_folio(inode->i_mapping, 0);
Christoph Hellwig66dabbb2023-03-07 15:34:10 +01003872 if (IS_ERR(folio))
Al Viro6a6c9902015-11-17 10:54:32 -05003873 return ERR_PTR(-ECHILD);
Matthew Wilcox (Oracle)7459c142022-09-02 20:46:27 +01003874 if (PageHWPoison(folio_page(folio, 0)) ||
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003875 !folio_test_uptodate(folio)) {
3876 folio_put(folio);
Al Viro6a6c9902015-11-17 10:54:32 -05003877 return ERR_PTR(-ECHILD);
3878 }
3879 } else {
Rik van Riele1e4cfd2024-09-03 11:19:28 -04003880 error = shmem_get_folio(inode, 0, 0, &folio, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003881 if (error)
3882 return ERR_PTR(error);
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003883 if (!folio)
Yang Shia7605422022-01-14 14:05:19 -08003884 return ERR_PTR(-ECHILD);
Matthew Wilcox (Oracle)7459c142022-09-02 20:46:27 +01003885 if (PageHWPoison(folio_page(folio, 0))) {
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003886 folio_unlock(folio);
3887 folio_put(folio);
Yang Shia7605422022-01-14 14:05:19 -08003888 return ERR_PTR(-ECHILD);
3889 }
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003890 folio_unlock(folio);
Al Viro6a6c9902015-11-17 10:54:32 -05003891 }
Matthew Wilcox (Oracle)e4b57722022-09-02 20:46:26 +01003892 set_delayed_call(done, shmem_put_link, folio);
3893 return folio_address(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894}
3895
Eric Parisb09e0fa2011-05-24 17:12:39 -07003896#ifdef CONFIG_TMPFS_XATTR
Theodore Ts'oe408e692022-07-14 21:59:12 -04003897
3898static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3899{
3900 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3901
3902 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3903
3904 return 0;
3905}
3906
Christian Brauner8782a9a2023-01-13 12:49:21 +01003907static int shmem_fileattr_set(struct mnt_idmap *idmap,
Theodore Ts'oe408e692022-07-14 21:59:12 -04003908 struct dentry *dentry, struct fileattr *fa)
3909{
3910 struct inode *inode = d_inode(dentry);
3911 struct shmem_inode_info *info = SHMEM_I(inode);
3912
3913 if (fileattr_has_fsx(fa))
3914 return -EOPNOTSUPP;
Hugh Dickinscb241332022-08-10 21:51:09 -07003915 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3916 return -EOPNOTSUPP;
Theodore Ts'oe408e692022-07-14 21:59:12 -04003917
3918 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3919 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3920
Hugh Dickinscb241332022-08-10 21:51:09 -07003921 shmem_set_inode_flags(inode, info->fsflags);
Jeff Layton65287332023-07-05 15:01:52 -04003922 inode_set_ctime_current(inode);
Jeff Layton36f05ca2022-09-09 09:00:31 -04003923 inode_inc_iversion(inode);
Theodore Ts'oe408e692022-07-14 21:59:12 -04003924 return 0;
3925}
3926
Eric Parisb09e0fa2011-05-24 17:12:39 -07003927/*
3928 * Superblocks without xattr inode operations may get some security.* xattr
3929 * support from the LSM "for free". As soon as we have any other xattrs
3930 * like ACLs, we also need to implement the security.* handlers at
3931 * filesystem level, though.
3932 */
3933
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003934/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003935 * Callback for security_inode_init_security() for acquiring xattrs.
3936 */
3937static int shmem_initxattrs(struct inode *inode,
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07003938 const struct xattr *xattr_array, void *fs_info)
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003939{
3940 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003941 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003942 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003943 struct simple_xattr *new_xattr;
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003944 size_t ispace = 0;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003945 size_t len;
3946
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003947 if (sbinfo->max_inodes) {
3948 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3949 ispace += simple_xattr_space(xattr->name,
3950 xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
3951 }
3952 if (ispace) {
3953 raw_spin_lock(&sbinfo->stat_lock);
3954 if (sbinfo->free_ispace < ispace)
3955 ispace = 0;
3956 else
3957 sbinfo->free_ispace -= ispace;
3958 raw_spin_unlock(&sbinfo->stat_lock);
3959 if (!ispace)
3960 return -ENOSPC;
3961 }
3962 }
3963
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003964 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003965 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003966 if (!new_xattr)
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003967 break;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003968
3969 len = strlen(xattr->name) + 1;
3970 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
Hugh Dickins572a3d12023-08-21 10:39:20 -07003971 GFP_KERNEL_ACCOUNT);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003972 if (!new_xattr->name) {
Chengguang Xu3bef7352020-07-23 21:15:14 -07003973 kvfree(new_xattr);
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003974 break;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003975 }
3976
3977 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3978 XATTR_SECURITY_PREFIX_LEN);
3979 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3980 xattr->name, len);
3981
Christian Brauner3b4c7bc2022-11-04 13:52:42 +01003982 simple_xattr_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003983 }
3984
Hugh Dickins2daf18a2023-08-08 21:33:56 -07003985 if (xattr->name != NULL) {
3986 if (ispace) {
3987 raw_spin_lock(&sbinfo->stat_lock);
3988 sbinfo->free_ispace += ispace;
3989 raw_spin_unlock(&sbinfo->stat_lock);
3990 }
3991 simple_xattrs_free(&info->xattrs, NULL);
3992 return -ENOMEM;
3993 }
3994
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003995 return 0;
3996}
3997
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003998static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003999 struct dentry *unused, struct inode *inode,
4000 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01004001{
Al Virob2968212016-04-10 20:48:24 -04004002 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01004003
4004 name = xattr_full_name(handler, name);
4005 return simple_xattr_get(&info->xattrs, name, buffer, size);
4006}
4007
4008static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Christian Brauner39f60c12023-01-13 12:49:23 +01004009 struct mnt_idmap *idmap,
Al Viro59301222016-05-27 10:19:30 -04004010 struct dentry *unused, struct inode *inode,
4011 const char *name, const void *value,
4012 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01004013{
Al Viro59301222016-05-27 10:19:30 -04004014 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004015 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins5de75972023-08-08 21:30:59 -07004016 struct simple_xattr *old_xattr;
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004017 size_t ispace = 0;
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01004018
4019 name = xattr_full_name(handler, name);
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004020 if (value && sbinfo->max_inodes) {
4021 ispace = simple_xattr_space(name, size);
4022 raw_spin_lock(&sbinfo->stat_lock);
4023 if (sbinfo->free_ispace < ispace)
4024 ispace = 0;
4025 else
4026 sbinfo->free_ispace -= ispace;
4027 raw_spin_unlock(&sbinfo->stat_lock);
4028 if (!ispace)
4029 return -ENOSPC;
4030 }
4031
Hugh Dickins5de75972023-08-08 21:30:59 -07004032 old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
4033 if (!IS_ERR(old_xattr)) {
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004034 ispace = 0;
4035 if (old_xattr && sbinfo->max_inodes)
4036 ispace = simple_xattr_space(old_xattr->name,
4037 old_xattr->size);
Hugh Dickins5de75972023-08-08 21:30:59 -07004038 simple_xattr_free(old_xattr);
4039 old_xattr = NULL;
Jeff Layton65287332023-07-05 15:01:52 -04004040 inode_set_ctime_current(inode);
Jeff Layton36f05ca2022-09-09 09:00:31 -04004041 inode_inc_iversion(inode);
4042 }
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004043 if (ispace) {
4044 raw_spin_lock(&sbinfo->stat_lock);
4045 sbinfo->free_ispace += ispace;
4046 raw_spin_unlock(&sbinfo->stat_lock);
4047 }
Hugh Dickins5de75972023-08-08 21:30:59 -07004048 return PTR_ERR(old_xattr);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01004049}
4050
4051static const struct xattr_handler shmem_security_xattr_handler = {
4052 .prefix = XATTR_SECURITY_PREFIX,
4053 .get = shmem_xattr_handler_get,
4054 .set = shmem_xattr_handler_set,
4055};
4056
4057static const struct xattr_handler shmem_trusted_xattr_handler = {
4058 .prefix = XATTR_TRUSTED_PREFIX,
4059 .get = shmem_xattr_handler_get,
4060 .set = shmem_xattr_handler_set,
4061};
4062
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004063static const struct xattr_handler shmem_user_xattr_handler = {
4064 .prefix = XATTR_USER_PREFIX,
4065 .get = shmem_xattr_handler_get,
4066 .set = shmem_xattr_handler_set,
4067};
4068
Wedson Almeida Filho2f502862023-09-30 02:00:32 -03004069static const struct xattr_handler * const shmem_xattr_handlers[] = {
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01004070 &shmem_security_xattr_handler,
4071 &shmem_trusted_xattr_handler,
Hugh Dickins2daf18a2023-08-08 21:33:56 -07004072 &shmem_user_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004073 NULL
4074};
4075
Eric Parisb09e0fa2011-05-24 17:12:39 -07004076static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
4077{
David Howells75c3cfa2015-03-17 22:26:12 +00004078 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01004079 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07004080}
4081#endif /* CONFIG_TMPFS_XATTR */
4082
Hugh Dickins69f07ec2011-08-03 16:21:26 -07004083static const struct inode_operations shmem_short_symlink_operations = {
Xavier Rochef7cd16a2022-03-22 14:39:55 -07004084 .getattr = shmem_getattr,
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004085 .setattr = shmem_setattr,
Al Viro6b255392015-11-17 10:20:54 -05004086 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004087#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07004088 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004089#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090};
4091
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004092static const struct inode_operations shmem_symlink_inode_operations = {
Xavier Rochef7cd16a2022-03-22 14:39:55 -07004093 .getattr = shmem_getattr,
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004094 .setattr = shmem_setattr,
Al Viro6b255392015-11-17 10:20:54 -05004095 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004096#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07004097 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004098#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07004099};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004100
David M. Grimes91828a42006-10-17 00:09:45 -07004101static struct dentry *shmem_get_parent(struct dentry *child)
4102{
4103 return ERR_PTR(-ESTALE);
4104}
4105
4106static int shmem_match(struct inode *ino, void *vfh)
4107{
4108 __u32 *fh = vfh;
4109 __u64 inum = fh[2];
4110 inum = (inum << 32) | fh[1];
4111 return ino->i_ino == inum && fh[0] == ino->i_generation;
4112}
4113
Amir Goldstein12ba7802018-06-07 17:07:15 -07004114/* Find any alias of inode, but prefer a hashed alias */
4115static struct dentry *shmem_find_alias(struct inode *inode)
4116{
4117 struct dentry *alias = d_find_alias(inode);
4118
4119 return alias ?: d_find_any_alias(inode);
4120}
4121
Christoph Hellwig480b1162007-10-21 16:42:13 -07004122static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
4123 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07004124{
David M. Grimes91828a42006-10-17 00:09:45 -07004125 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07004126 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07004127 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07004128
Christoph Hellwig480b1162007-10-21 16:42:13 -07004129 if (fh_len < 3)
4130 return NULL;
4131
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07004132 inum = fid->raw[2];
4133 inum = (inum << 32) | fid->raw[1];
4134
Christoph Hellwig480b1162007-10-21 16:42:13 -07004135 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
4136 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07004137 if (inode) {
Amir Goldstein12ba7802018-06-07 17:07:15 -07004138 dentry = shmem_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07004139 iput(inode);
4140 }
4141
Christoph Hellwig480b1162007-10-21 16:42:13 -07004142 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07004143}
4144
Al Virob0b03822012-04-02 14:34:06 -04004145static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
4146 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07004147{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05304148 if (*len < 3) {
4149 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09004150 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05304151 }
David M. Grimes91828a42006-10-17 00:09:45 -07004152
Al Viro1d3382cb2010-10-23 15:19:20 -04004153 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07004154 /* Unfortunately insert_inode_hash is not idempotent,
4155 * so as we hash inodes here rather than at creation
4156 * time, we need a lock to ensure we only try
4157 * to do it once
4158 */
4159 static DEFINE_SPINLOCK(lock);
4160 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04004161 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07004162 __insert_inode_hash(inode,
4163 inode->i_ino + inode->i_generation);
4164 spin_unlock(&lock);
4165 }
4166
4167 fh[0] = inode->i_generation;
4168 fh[1] = inode->i_ino;
4169 fh[2] = ((__u64)inode->i_ino) >> 32;
4170
4171 *len = 3;
4172 return 1;
4173}
4174
Christoph Hellwig39655162007-10-21 16:42:17 -07004175static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07004176 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07004177 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07004178 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07004179};
4180
Al Viro626c3922019-09-08 20:28:06 -04004181enum shmem_param {
4182 Opt_gid,
4183 Opt_huge,
4184 Opt_mode,
4185 Opt_mpol,
4186 Opt_nr_blocks,
4187 Opt_nr_inodes,
4188 Opt_size,
4189 Opt_uid,
Chris Downea3271f2020-08-06 23:20:25 -07004190 Opt_inode32,
4191 Opt_inode64,
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004192 Opt_noswap,
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004193 Opt_quota,
4194 Opt_usrquota,
4195 Opt_grpquota,
Lukas Czernerde4c0e72023-07-25 16:45:09 +02004196 Opt_usrquota_block_hardlimit,
4197 Opt_usrquota_inode_hardlimit,
4198 Opt_grpquota_block_hardlimit,
4199 Opt_grpquota_inode_hardlimit,
Al Viro626c3922019-09-08 20:28:06 -04004200};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201
Al Viro5eede622019-12-16 13:33:32 -05004202static const struct constant_table shmem_param_enums_huge[] = {
Al Viro2710c957a2019-09-06 22:12:08 -04004203 {"never", SHMEM_HUGE_NEVER },
4204 {"always", SHMEM_HUGE_ALWAYS },
4205 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
4206 {"advise", SHMEM_HUGE_ADVISE },
Al Viro2710c957a2019-09-06 22:12:08 -04004207 {}
4208};
4209
Al Virod7167b12019-09-07 07:23:15 -04004210const struct fs_parameter_spec shmem_fs_parameters[] = {
Eric Sandeen2ec07012024-06-27 19:38:12 -05004211 fsparam_gid ("gid", Opt_gid),
Al Viro2710c957a2019-09-06 22:12:08 -04004212 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
Al Viro626c3922019-09-08 20:28:06 -04004213 fsparam_u32oct("mode", Opt_mode),
4214 fsparam_string("mpol", Opt_mpol),
4215 fsparam_string("nr_blocks", Opt_nr_blocks),
4216 fsparam_string("nr_inodes", Opt_nr_inodes),
4217 fsparam_string("size", Opt_size),
Eric Sandeen2ec07012024-06-27 19:38:12 -05004218 fsparam_uid ("uid", Opt_uid),
Chris Downea3271f2020-08-06 23:20:25 -07004219 fsparam_flag ("inode32", Opt_inode32),
4220 fsparam_flag ("inode64", Opt_inode64),
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004221 fsparam_flag ("noswap", Opt_noswap),
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004222#ifdef CONFIG_TMPFS_QUOTA
4223 fsparam_flag ("quota", Opt_quota),
4224 fsparam_flag ("usrquota", Opt_usrquota),
4225 fsparam_flag ("grpquota", Opt_grpquota),
Lukas Czernerde4c0e72023-07-25 16:45:09 +02004226 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
4227 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
4228 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
4229 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004230#endif
Al Viro626c3922019-09-08 20:28:06 -04004231 {}
4232};
4233
David Howellsf3235622019-03-25 16:38:31 +00004234static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
Al Viro626c3922019-09-08 20:28:06 -04004235{
David Howellsf3235622019-03-25 16:38:31 +00004236 struct shmem_options *ctx = fc->fs_private;
Al Viro626c3922019-09-08 20:28:06 -04004237 struct fs_parse_result result;
4238 unsigned long long size;
4239 char *rest;
4240 int opt;
Christian Brauner02006792023-08-01 18:17:04 +02004241 kuid_t kuid;
4242 kgid_t kgid;
Al Viro626c3922019-09-08 20:28:06 -04004243
Al Virod7167b12019-09-07 07:23:15 -04004244 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
David Howellsf3235622019-03-25 16:38:31 +00004245 if (opt < 0)
Al Viro626c3922019-09-08 20:28:06 -04004246 return opt;
Al Viro626c3922019-09-08 20:28:06 -04004247
4248 switch (opt) {
4249 case Opt_size:
4250 size = memparse(param->string, &rest);
Al Viroe04dc422019-09-08 19:20:12 -04004251 if (*rest == '%') {
4252 size <<= PAGE_SHIFT;
4253 size *= totalram_pages();
4254 do_div(size, 100);
4255 rest++;
4256 }
4257 if (*rest)
Al Viro626c3922019-09-08 20:28:06 -04004258 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04004259 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
4260 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04004261 break;
4262 case Opt_nr_blocks:
4263 ctx->blocks = memparse(param->string, &rest);
Hugh Dickinse07c4692023-08-08 21:32:21 -07004264 if (*rest || ctx->blocks > LONG_MAX)
Al Viro626c3922019-09-08 20:28:06 -04004265 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04004266 ctx->seen |= SHMEM_SEEN_BLOCKS;
Al Viro626c3922019-09-08 20:28:06 -04004267 break;
4268 case Opt_nr_inodes:
4269 ctx->inodes = memparse(param->string, &rest);
Hugh Dickinse07c4692023-08-08 21:32:21 -07004270 if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
Al Viro626c3922019-09-08 20:28:06 -04004271 goto bad_value;
Al Viroe04dc422019-09-08 19:20:12 -04004272 ctx->seen |= SHMEM_SEEN_INODES;
Al Viro626c3922019-09-08 20:28:06 -04004273 break;
4274 case Opt_mode:
4275 ctx->mode = result.uint_32 & 07777;
4276 break;
4277 case Opt_uid:
Eric Sandeen2ec07012024-06-27 19:38:12 -05004278 kuid = result.uid;
Christian Brauner02006792023-08-01 18:17:04 +02004279
4280 /*
4281 * The requested uid must be representable in the
4282 * filesystem's idmapping.
4283 */
4284 if (!kuid_has_mapping(fc->user_ns, kuid))
4285 goto bad_value;
4286
4287 ctx->uid = kuid;
Al Viro626c3922019-09-08 20:28:06 -04004288 break;
4289 case Opt_gid:
Eric Sandeen2ec07012024-06-27 19:38:12 -05004290 kgid = result.gid;
Christian Brauner02006792023-08-01 18:17:04 +02004291
4292 /*
4293 * The requested gid must be representable in the
4294 * filesystem's idmapping.
4295 */
4296 if (!kgid_has_mapping(fc->user_ns, kgid))
4297 goto bad_value;
4298
4299 ctx->gid = kgid;
Al Viro626c3922019-09-08 20:28:06 -04004300 break;
4301 case Opt_huge:
4302 ctx->huge = result.uint_32;
4303 if (ctx->huge != SHMEM_HUGE_NEVER &&
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004304 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
Al Viro626c3922019-09-08 20:28:06 -04004305 has_transparent_hugepage()))
4306 goto unsupported_parameter;
Al Viroe04dc422019-09-08 19:20:12 -04004307 ctx->seen |= SHMEM_SEEN_HUGE;
Al Viro626c3922019-09-08 20:28:06 -04004308 break;
4309 case Opt_mpol:
4310 if (IS_ENABLED(CONFIG_NUMA)) {
4311 mpol_put(ctx->mpol);
4312 ctx->mpol = NULL;
4313 if (mpol_parse_str(param->string, &ctx->mpol))
4314 goto bad_value;
4315 break;
4316 }
4317 goto unsupported_parameter;
Chris Downea3271f2020-08-06 23:20:25 -07004318 case Opt_inode32:
4319 ctx->full_inums = false;
4320 ctx->seen |= SHMEM_SEEN_INUMS;
4321 break;
4322 case Opt_inode64:
4323 if (sizeof(ino_t) < 8) {
4324 return invalfc(fc,
4325 "Cannot use inode64 with <64bit inums in kernel\n");
4326 }
4327 ctx->full_inums = true;
4328 ctx->seen |= SHMEM_SEEN_INUMS;
4329 break;
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004330 case Opt_noswap:
Christian Brauner01106e12023-04-20 10:57:43 +02004331 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
4332 return invalfc(fc,
4333 "Turning off swap in unprivileged tmpfs mounts unsupported");
4334 }
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004335 ctx->noswap = true;
4336 ctx->seen |= SHMEM_SEEN_NOSWAP;
4337 break;
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004338 case Opt_quota:
4339 if (fc->user_ns != &init_user_ns)
4340 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4341 ctx->seen |= SHMEM_SEEN_QUOTA;
4342 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
4343 break;
4344 case Opt_usrquota:
4345 if (fc->user_ns != &init_user_ns)
4346 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4347 ctx->seen |= SHMEM_SEEN_QUOTA;
4348 ctx->quota_types |= QTYPE_MASK_USR;
4349 break;
4350 case Opt_grpquota:
4351 if (fc->user_ns != &init_user_ns)
4352 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4353 ctx->seen |= SHMEM_SEEN_QUOTA;
4354 ctx->quota_types |= QTYPE_MASK_GRP;
4355 break;
Lukas Czernerde4c0e72023-07-25 16:45:09 +02004356 case Opt_usrquota_block_hardlimit:
4357 size = memparse(param->string, &rest);
4358 if (*rest || !size)
4359 goto bad_value;
4360 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4361 return invalfc(fc,
4362 "User quota block hardlimit too large.");
4363 ctx->qlimits.usrquota_bhardlimit = size;
4364 break;
4365 case Opt_grpquota_block_hardlimit:
4366 size = memparse(param->string, &rest);
4367 if (*rest || !size)
4368 goto bad_value;
4369 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4370 return invalfc(fc,
4371 "Group quota block hardlimit too large.");
4372 ctx->qlimits.grpquota_bhardlimit = size;
4373 break;
4374 case Opt_usrquota_inode_hardlimit:
4375 size = memparse(param->string, &rest);
4376 if (*rest || !size)
4377 goto bad_value;
4378 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4379 return invalfc(fc,
4380 "User quota inode hardlimit too large.");
4381 ctx->qlimits.usrquota_ihardlimit = size;
4382 break;
4383 case Opt_grpquota_inode_hardlimit:
4384 size = memparse(param->string, &rest);
4385 if (*rest || !size)
4386 goto bad_value;
4387 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4388 return invalfc(fc,
4389 "Group quota inode hardlimit too large.");
4390 ctx->qlimits.grpquota_ihardlimit = size;
4391 break;
Al Viroe04dc422019-09-08 19:20:12 -04004392 }
4393 return 0;
4394
Al Viro626c3922019-09-08 20:28:06 -04004395unsupported_parameter:
Al Virof35aa2b2019-12-21 21:35:55 -05004396 return invalfc(fc, "Unsupported parameter '%s'", param->key);
Al Viro626c3922019-09-08 20:28:06 -04004397bad_value:
Al Virof35aa2b2019-12-21 21:35:55 -05004398 return invalfc(fc, "Bad value for '%s'", param->key);
Al Viroe04dc422019-09-08 19:20:12 -04004399}
4400
David Howellsf3235622019-03-25 16:38:31 +00004401static int shmem_parse_options(struct fs_context *fc, void *data)
Al Viroe04dc422019-09-08 19:20:12 -04004402{
David Howellsf3235622019-03-25 16:38:31 +00004403 char *options = data;
4404
Al Viro33f37c62019-10-09 22:48:01 -04004405 if (options) {
4406 int err = security_sb_eat_lsm_opts(options, &fc->security);
4407 if (err)
4408 return err;
4409 }
4410
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00004411 while (options != NULL) {
Al Viro626c3922019-09-08 20:28:06 -04004412 char *this_char = options;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00004413 for (;;) {
4414 /*
4415 * NUL-terminate this option: unfortunately,
4416 * mount options form a comma-separated list,
4417 * but mpol's nodelist may also contain commas.
4418 */
4419 options = strchr(options, ',');
4420 if (options == NULL)
4421 break;
4422 options++;
4423 if (!isdigit(*options)) {
4424 options[-1] = '\0';
4425 break;
4426 }
4427 }
Al Viro626c3922019-09-08 20:28:06 -04004428 if (*this_char) {
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -07004429 char *value = strchr(this_char, '=');
David Howellsf3235622019-03-25 16:38:31 +00004430 size_t len = 0;
Al Viro626c3922019-09-08 20:28:06 -04004431 int err;
4432
4433 if (value) {
4434 *value++ = '\0';
David Howellsf3235622019-03-25 16:38:31 +00004435 len = strlen(value);
Al Viro626c3922019-09-08 20:28:06 -04004436 }
David Howellsf3235622019-03-25 16:38:31 +00004437 err = vfs_parse_fs_string(fc, this_char, value, len);
4438 if (err < 0)
4439 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 }
4442 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443}
4444
David Howellsf3235622019-03-25 16:38:31 +00004445/*
4446 * Reconfigure a shmem filesystem.
David Howellsf3235622019-03-25 16:38:31 +00004447 */
4448static int shmem_reconfigure(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449{
David Howellsf3235622019-03-25 16:38:31 +00004450 struct shmem_options *ctx = fc->fs_private;
4451 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
Hugh Dickinse07c4692023-08-08 21:32:21 -07004452 unsigned long used_isp;
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07004453 struct mempolicy *mpol = NULL;
David Howellsf3235622019-03-25 16:38:31 +00004454 const char *err;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004455
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07004456 raw_spin_lock(&sbinfo->stat_lock);
Hugh Dickinse07c4692023-08-08 21:32:21 -07004457 used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
ZhaoLong Wang0c98c8e2022-06-29 20:43:24 +08004458
David Howellsf3235622019-03-25 16:38:31 +00004459 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4460 if (!sbinfo->max_blocks) {
4461 err = "Cannot retroactively limit size";
Al Viro0b5071d2019-09-08 18:49:18 -04004462 goto out;
David Howellsf3235622019-03-25 16:38:31 +00004463 }
Al Viro0b5071d2019-09-08 18:49:18 -04004464 if (percpu_counter_compare(&sbinfo->used_blocks,
David Howellsf3235622019-03-25 16:38:31 +00004465 ctx->blocks) > 0) {
4466 err = "Too small a size for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04004467 goto out;
David Howellsf3235622019-03-25 16:38:31 +00004468 }
Al Viro0b5071d2019-09-08 18:49:18 -04004469 }
David Howellsf3235622019-03-25 16:38:31 +00004470 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4471 if (!sbinfo->max_inodes) {
4472 err = "Cannot retroactively limit inodes";
Al Viro0b5071d2019-09-08 18:49:18 -04004473 goto out;
David Howellsf3235622019-03-25 16:38:31 +00004474 }
Hugh Dickinse07c4692023-08-08 21:32:21 -07004475 if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
David Howellsf3235622019-03-25 16:38:31 +00004476 err = "Too few inodes for current use";
Al Viro0b5071d2019-09-08 18:49:18 -04004477 goto out;
David Howellsf3235622019-03-25 16:38:31 +00004478 }
Al Viro0b5071d2019-09-08 18:49:18 -04004479 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004480
Chris Downea3271f2020-08-06 23:20:25 -07004481 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4482 sbinfo->next_ino > UINT_MAX) {
4483 err = "Current inum too high to switch to 32-bit inums";
4484 goto out;
4485 }
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004486 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4487 err = "Cannot disable swap on remount";
4488 goto out;
4489 }
4490 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4491 err = "Cannot enable swap on remount if it was disabled on first mount";
4492 goto out;
4493 }
Chris Downea3271f2020-08-06 23:20:25 -07004494
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004495 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4496 !sb_any_quota_loaded(fc->root->d_sb)) {
4497 err = "Cannot enable quota on remount";
4498 goto out;
4499 }
4500
Lukas Czernerde4c0e72023-07-25 16:45:09 +02004501#ifdef CONFIG_TMPFS_QUOTA
4502#define CHANGED_LIMIT(name) \
4503 (ctx->qlimits.name## hardlimit && \
4504 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4505
4506 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4507 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4508 err = "Cannot change global quota limit on remount";
4509 goto out;
4510 }
4511#endif /* CONFIG_TMPFS_QUOTA */
4512
David Howellsf3235622019-03-25 16:38:31 +00004513 if (ctx->seen & SHMEM_SEEN_HUGE)
4514 sbinfo->huge = ctx->huge;
Chris Downea3271f2020-08-06 23:20:25 -07004515 if (ctx->seen & SHMEM_SEEN_INUMS)
4516 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00004517 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4518 sbinfo->max_blocks = ctx->blocks;
4519 if (ctx->seen & SHMEM_SEEN_INODES) {
4520 sbinfo->max_inodes = ctx->inodes;
Hugh Dickinse07c4692023-08-08 21:32:21 -07004521 sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
Al Viro0b5071d2019-09-08 18:49:18 -04004522 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07004523
Greg Thelen5f001102013-02-22 16:36:01 -08004524 /*
4525 * Preserve previous mempolicy unless mpol remount option was specified.
4526 */
David Howellsf3235622019-03-25 16:38:31 +00004527 if (ctx->mpol) {
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07004528 mpol = sbinfo->mpol;
David Howellsf3235622019-03-25 16:38:31 +00004529 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4530 ctx->mpol = NULL;
Greg Thelen5f001102013-02-22 16:36:01 -08004531 }
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004532
4533 if (ctx->noswap)
4534 sbinfo->noswap = true;
4535
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07004536 raw_spin_unlock(&sbinfo->stat_lock);
4537 mpol_put(mpol);
David Howellsf3235622019-03-25 16:38:31 +00004538 return 0;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004539out:
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07004540 raw_spin_unlock(&sbinfo->stat_lock);
Al Virof35aa2b2019-12-21 21:35:55 -05004541 return invalfc(fc, "%s", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004542}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004543
Al Viro34c80b12011-12-08 21:32:45 -05004544static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004545{
Al Viro34c80b12011-12-08 21:32:45 -05004546 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
Tu Jinjiang283ebde2023-05-25 11:16:40 +08004547 struct mempolicy *mpol;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004548
4549 if (sbinfo->max_blocks != shmem_default_max_blocks())
ZhangPengb91742d2023-08-04 09:25:56 +08004550 seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004551 if (sbinfo->max_inodes != shmem_default_max_inodes())
4552 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
Joe Perches0825a6f2018-06-14 15:27:58 -07004553 if (sbinfo->mode != (0777 | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04004554 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08004555 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4556 seq_printf(seq, ",uid=%u",
4557 from_kuid_munged(&init_user_ns, sbinfo->uid));
4558 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4559 seq_printf(seq, ",gid=%u",
4560 from_kgid_munged(&init_user_ns, sbinfo->gid));
Chris Downea3271f2020-08-06 23:20:25 -07004561
4562 /*
4563 * Showing inode{64,32} might be useful even if it's the system default,
4564 * since then people don't have to resort to checking both here and
4565 * /proc/config.gz to confirm 64-bit inums were successfully applied
4566 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4567 *
4568 * We hide it when inode64 isn't the default and we are using 32-bit
4569 * inodes, since that probably just means the feature isn't even under
4570 * consideration.
4571 *
4572 * As such:
4573 *
4574 * +-----------------+-----------------+
4575 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4576 * +------------------+-----------------+-----------------+
4577 * | full_inums=true | show | show |
4578 * | full_inums=false | show | hide |
4579 * +------------------+-----------------+-----------------+
4580 *
4581 */
4582 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4583 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004584#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004585 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4586 if (sbinfo->huge)
4587 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4588#endif
Tu Jinjiang283ebde2023-05-25 11:16:40 +08004589 mpol = shmem_get_sbmpol(sbinfo);
4590 shmem_show_mpol(seq, mpol);
4591 mpol_put(mpol);
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004592 if (sbinfo->noswap)
4593 seq_printf(seq, ",noswap");
Jan Karab4d3de52024-01-29 13:01:31 +01004594#ifdef CONFIG_TMPFS_QUOTA
4595 if (sb_has_quota_active(root->d_sb, USRQUOTA))
4596 seq_printf(seq, ",usrquota");
4597 if (sb_has_quota_active(root->d_sb, GRPQUOTA))
4598 seq_printf(seq, ",grpquota");
4599 if (sbinfo->qlimits.usrquota_bhardlimit)
4600 seq_printf(seq, ",usrquota_block_hardlimit=%lld",
4601 sbinfo->qlimits.usrquota_bhardlimit);
4602 if (sbinfo->qlimits.grpquota_bhardlimit)
4603 seq_printf(seq, ",grpquota_block_hardlimit=%lld",
4604 sbinfo->qlimits.grpquota_bhardlimit);
4605 if (sbinfo->qlimits.usrquota_ihardlimit)
4606 seq_printf(seq, ",usrquota_inode_hardlimit=%lld",
4607 sbinfo->qlimits.usrquota_ihardlimit);
4608 if (sbinfo->qlimits.grpquota_ihardlimit)
4609 seq_printf(seq, ",grpquota_inode_hardlimit=%lld",
4610 sbinfo->qlimits.grpquota_ihardlimit);
4611#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004612 return 0;
4613}
David Herrmann9183df22014-08-08 14:25:29 -07004614
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004615#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004616
4617static void shmem_put_super(struct super_block *sb)
4618{
Hugh Dickins602586a2010-08-17 15:23:56 -07004619 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4620
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004621#ifdef CONFIG_TMPFS_QUOTA
4622 shmem_disable_quotas(sb);
4623#endif
Chris Downe809d5f2020-08-06 23:20:20 -07004624 free_percpu(sbinfo->ino_batch);
Hugh Dickins602586a2010-08-17 15:23:56 -07004625 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08004626 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07004627 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628 sb->s_fs_info = NULL;
4629}
4630
David Howellsf3235622019-03-25 16:38:31 +00004631static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004632{
David Howellsf3235622019-03-25 16:38:31 +00004633 struct shmem_options *ctx = fc->fs_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004634 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004635 struct shmem_sb_info *sbinfo;
Carlos Maiolino71480662023-07-25 16:45:05 +02004636 int error = -ENOMEM;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004637
4638 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07004639 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004640 L1_CACHE_BYTES), GFP_KERNEL);
4641 if (!sbinfo)
Carlos Maiolino71480662023-07-25 16:45:05 +02004642 return error;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004643
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004644 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004646#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07004647 /*
4648 * Per default we only allow half of the physical ram per
4649 * tmpfs instance, limiting inodes to one per page of lowmem;
4650 * but the internal instance is left unlimited.
4651 */
Linus Torvalds1751e8a2017-11-27 13:05:09 -08004652 if (!(sb->s_flags & SB_KERNMOUNT)) {
David Howellsf3235622019-03-25 16:38:31 +00004653 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4654 ctx->blocks = shmem_default_max_blocks();
4655 if (!(ctx->seen & SHMEM_SEEN_INODES))
4656 ctx->inodes = shmem_default_max_inodes();
Chris Downea3271f2020-08-06 23:20:25 -07004657 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4658 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
Luis Chamberlain2c6efe92023-03-09 15:05:45 -08004659 sbinfo->noswap = ctx->noswap;
Al Viroca4e0512013-08-31 12:57:10 -04004660 } else {
Linus Torvalds1751e8a2017-11-27 13:05:09 -08004661 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004662 }
David M. Grimes91828a42006-10-17 00:09:45 -07004663 sb->s_export_op = &shmem_export_ops;
Jeff Layton36f05ca2022-09-09 09:00:31 -04004664 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004665#else
Linus Torvalds1751e8a2017-11-27 13:05:09 -08004666 sb->s_flags |= SB_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004667#endif
David Howellsf3235622019-03-25 16:38:31 +00004668 sbinfo->max_blocks = ctx->blocks;
Hugh Dickinse07c4692023-08-08 21:32:21 -07004669 sbinfo->max_inodes = ctx->inodes;
4670 sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
Chris Downe809d5f2020-08-06 23:20:20 -07004671 if (sb->s_flags & SB_KERNMOUNT) {
4672 sbinfo->ino_batch = alloc_percpu(ino_t);
4673 if (!sbinfo->ino_batch)
4674 goto failed;
4675 }
David Howellsf3235622019-03-25 16:38:31 +00004676 sbinfo->uid = ctx->uid;
4677 sbinfo->gid = ctx->gid;
Chris Downea3271f2020-08-06 23:20:25 -07004678 sbinfo->full_inums = ctx->full_inums;
David Howellsf3235622019-03-25 16:38:31 +00004679 sbinfo->mode = ctx->mode;
4680 sbinfo->huge = ctx->huge;
4681 sbinfo->mpol = ctx->mpol;
4682 ctx->mpol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004683
Sebastian Andrzej Siewiorbf11b9a2021-09-02 14:54:03 -07004684 raw_spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09004685 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07004686 goto failed;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07004687 spin_lock_init(&sbinfo->shrinklist_lock);
4688 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004689
Hugh Dickins285b2c42011-08-03 16:21:20 -07004690 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004691 sb->s_blocksize = PAGE_SIZE;
4692 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004693 sb->s_magic = TMPFS_MAGIC;
4694 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01004695 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07004696#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004697 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07004698#endif
4699#ifdef CONFIG_TMPFS_POSIX_ACL
Linus Torvalds1751e8a2017-11-27 13:05:09 -08004700 sb->s_flags |= SB_POSIXACL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004701#endif
Kent Overstreeta4af51c2024-02-06 21:56:15 -05004702 uuid_t uuid;
4703 uuid_gen(&uuid);
4704 super_set_uuid(sb, uuid.b, sizeof(uuid));
Hugh Dickins0edd73b2005-06-21 17:15:04 -07004705
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004706#ifdef CONFIG_TMPFS_QUOTA
4707 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4708 sb->dq_op = &shmem_quota_operations;
4709 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4710 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4711
Lukas Czernerde4c0e72023-07-25 16:45:09 +02004712 /* Copy the default limits from ctx into sbinfo */
4713 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4714 sizeof(struct shmem_quota_limits));
4715
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004716 if (shmem_enable_quotas(sb, ctx->quota_types))
4717 goto failed;
4718 }
4719#endif /* CONFIG_TMPFS_QUOTA */
4720
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07004721 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL,
4722 S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Carlos Maiolino71480662023-07-25 16:45:05 +02004723 if (IS_ERR(inode)) {
4724 error = PTR_ERR(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004725 goto failed;
Carlos Maiolino71480662023-07-25 16:45:05 +02004726 }
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004727 inode->i_uid = sbinfo->uid;
4728 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05004729 sb->s_root = d_make_root(inode);
4730 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05004731 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004732 return 0;
4733
Linus Torvalds1da177e2005-04-16 15:20:36 -07004734failed:
4735 shmem_put_super(sb);
Carlos Maiolino71480662023-07-25 16:45:05 +02004736 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737}
4738
David Howellsf3235622019-03-25 16:38:31 +00004739static int shmem_get_tree(struct fs_context *fc)
4740{
4741 return get_tree_nodev(fc, shmem_fill_super);
4742}
4743
4744static void shmem_free_fc(struct fs_context *fc)
4745{
4746 struct shmem_options *ctx = fc->fs_private;
4747
4748 if (ctx) {
4749 mpol_put(ctx->mpol);
4750 kfree(ctx);
4751 }
4752}
4753
4754static const struct fs_context_operations shmem_fs_context_ops = {
4755 .free = shmem_free_fc,
4756 .get_tree = shmem_get_tree,
4757#ifdef CONFIG_TMPFS
4758 .parse_monolithic = shmem_parse_options,
4759 .parse_param = shmem_parse_one,
4760 .reconfigure = shmem_reconfigure,
4761#endif
4762};
4763
Alexey Dobriyan68279f92023-10-11 19:55:00 +03004764static struct kmem_cache *shmem_inode_cachep __ro_after_init;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004765
4766static struct inode *shmem_alloc_inode(struct super_block *sb)
4767{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004768 struct shmem_inode_info *info;
Muchun Songfd60b282022-03-22 14:41:03 -07004769 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004770 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004772 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004773}
4774
Al Viro74b1da52019-04-15 23:19:05 -04004775static void shmem_free_in_core_inode(struct inode *inode)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11004776{
Al Viro84e710d2016-04-15 00:58:55 -04004777 if (S_ISLNK(inode->i_mode))
4778 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11004779 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4780}
4781
Linus Torvalds1da177e2005-04-16 15:20:36 -07004782static void shmem_destroy_inode(struct inode *inode)
4783{
Al Viro09208d12011-07-26 03:15:03 -04004784 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004785 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Chuck Levera2e45952023-06-30 13:49:03 -04004786 if (S_ISDIR(inode->i_mode))
4787 simple_offset_destroy(shmem_get_offset_ctx(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004788}
4789
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004790static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004792 struct shmem_inode_info *info = foo;
4793 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004794}
4795
Alexey Dobriyan68279f92023-10-11 19:55:00 +03004796static void __init shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004797{
4798 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4799 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08004800 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004801}
4802
Alexey Dobriyan68279f92023-10-11 19:55:00 +03004803static void __init shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004804{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07004805 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004806}
4807
Yang Shia7605422022-01-14 14:05:19 -08004808/* Keep the page in page cache instead of truncating it */
Matthew Wilcox (Oracle)af7628d2023-11-17 16:14:47 +00004809static int shmem_error_remove_folio(struct address_space *mapping,
4810 struct folio *folio)
Yang Shia7605422022-01-14 14:05:19 -08004811{
4812 return 0;
4813}
4814
Christoph Hellwigaefacb22024-02-19 07:27:10 +01004815static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816 .writepage = shmem_writepage,
Matthew Wilcox (Oracle)46de8b972022-02-09 20:22:13 +00004817 .dirty_folio = noop_dirty_folio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004818#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07004819 .write_begin = shmem_write_begin,
4820 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004821#endif
Andrew Morton1c939232014-10-09 15:27:59 -07004822#ifdef CONFIG_MIGRATION
Matthew Wilcox (Oracle)54184652022-06-06 10:27:41 -04004823 .migrate_folio = migrate_folio,
Andrew Morton1c939232014-10-09 15:27:59 -07004824#endif
Matthew Wilcox (Oracle)af7628d2023-11-17 16:14:47 +00004825 .error_remove_folio = shmem_error_remove_folio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004826};
4827
Helge Deller15ad7cd2006-12-06 20:40:36 -08004828static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004829 .mmap = shmem_mmap,
Hugh Dickinse88e0d32023-08-10 23:27:07 -07004830 .open = shmem_file_open,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004831 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08004833 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04004834 .read_iter = shmem_file_read_iter,
Hugh Dickinse88e0d32023-08-10 23:27:07 -07004835 .write_iter = shmem_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02004836 .fsync = noop_fsync,
David Howellsbd194b12023-05-22 14:49:56 +01004837 .splice_read = shmem_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04004838 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07004839 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004840#endif
4841};
4842
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004843static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07004844 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07004845 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004846#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07004847 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08004848 .set_acl = simple_set_acl,
Theodore Ts'oe408e692022-07-14 21:59:12 -04004849 .fileattr_get = shmem_fileattr_get,
4850 .fileattr_set = shmem_fileattr_set,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004851#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852};
4853
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004854static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004855#ifdef CONFIG_TMPFS
Xavier Rochef7cd16a2022-03-22 14:39:55 -07004856 .getattr = shmem_getattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004857 .create = shmem_create,
4858 .lookup = simple_lookup,
4859 .link = shmem_link,
4860 .unlink = shmem_unlink,
4861 .symlink = shmem_symlink,
4862 .mkdir = shmem_mkdir,
4863 .rmdir = shmem_rmdir,
4864 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02004865 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04004866 .tmpfile = shmem_tmpfile,
Chuck Levera2e45952023-06-30 13:49:03 -04004867 .get_offset_ctx = shmem_get_offset_ctx,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004868#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07004869#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07004870 .listxattr = shmem_listxattr,
Theodore Ts'oe408e692022-07-14 21:59:12 -04004871 .fileattr_get = shmem_fileattr_get,
4872 .fileattr_set = shmem_fileattr_set,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004873#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004874#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07004875 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08004876 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004877#endif
4878};
4879
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08004880static const struct inode_operations shmem_special_inode_operations = {
Xavier Rochef7cd16a2022-03-22 14:39:55 -07004881 .getattr = shmem_getattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004882#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07004883 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07004884#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004885#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07004886 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08004887 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07004888#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004889};
4890
Hugh Dickins759b9772007-03-05 00:30:28 -08004891static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004892 .alloc_inode = shmem_alloc_inode,
Al Viro74b1da52019-04-15 23:19:05 -04004893 .free_inode = shmem_free_in_core_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004894 .destroy_inode = shmem_destroy_inode,
4895#ifdef CONFIG_TMPFS
4896 .statfs = shmem_statfs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08004897 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898#endif
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004899#ifdef CONFIG_TMPFS_QUOTA
4900 .get_dquots = shmem_get_dquots,
4901#endif
Al Viro1f895f72010-06-05 19:10:41 -04004902 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004903 .drop_inode = generic_delete_inode,
4904 .put_super = shmem_put_super,
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004905#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07004906 .nr_cached_objects = shmem_unused_huge_count,
4907 .free_cached_objects = shmem_unused_huge_scan,
4908#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004909};
4910
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04004911static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07004912 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07004913 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004914#ifdef CONFIG_NUMA
4915 .set_policy = shmem_set_policy,
4916 .get_policy = shmem_get_policy,
4917#endif
4918};
4919
Pasha Tatashind09e8ca2022-11-15 02:06:01 +00004920static const struct vm_operations_struct shmem_anon_vm_ops = {
4921 .fault = shmem_fault,
4922 .map_pages = filemap_map_pages,
4923#ifdef CONFIG_NUMA
4924 .set_policy = shmem_set_policy,
4925 .get_policy = shmem_get_policy,
4926#endif
4927};
4928
David Howellsf3235622019-03-25 16:38:31 +00004929int shmem_init_fs_context(struct fs_context *fc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930{
David Howellsf3235622019-03-25 16:38:31 +00004931 struct shmem_options *ctx;
4932
4933 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4934 if (!ctx)
4935 return -ENOMEM;
4936
4937 ctx->mode = 0777 | S_ISVTX;
4938 ctx->uid = current_fsuid();
4939 ctx->gid = current_fsgid();
4940
4941 fc->fs_private = ctx;
4942 fc->ops = &shmem_fs_context_ops;
4943 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004944}
4945
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004946static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004947 .owner = THIS_MODULE,
4948 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00004949 .init_fs_context = shmem_init_fs_context,
4950#ifdef CONFIG_TMPFS
Al Virod7167b12019-09-07 07:23:15 -04004951 .parameters = shmem_fs_parameters,
David Howellsf3235622019-03-25 16:38:31 +00004952#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004953 .kill_sb = kill_litter_super,
Christian Braunerdb58b5e2023-09-20 16:40:22 +02004954 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004955};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004956
Miaohe Lin9096bbe2022-04-28 23:15:58 -07004957void __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958{
4959 int error;
4960
weiping zhang9a8ec03e2017-11-15 17:38:18 -08004961 shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004962
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004963#ifdef CONFIG_TMPFS_QUOTA
Kemeng Shia838e5d2024-07-15 21:05:32 +08004964 register_quota_format(&shmem_quota_format);
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004965#endif
4966
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004967 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004968 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07004969 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004970 goto out2;
4971 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07004972
Al Viroca4e0512013-08-31 12:57:10 -04004973 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004974 if (IS_ERR(shm_mnt)) {
4975 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07004976 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004977 goto out1;
4978 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004979
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07004980#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07004981 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004982 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4983 else
Hugh Dickins5e6e5a12021-09-02 14:54:37 -07004984 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
Baolin Wang4b989952024-06-11 18:11:07 +08004985
4986 /*
4987 * Default to setting PMD-sized THP to inherit the global setting and
4988 * disable all other multi-size THPs.
4989 */
4990 huge_shmem_orders_inherit = BIT(HPAGE_PMD_ORDER);
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004991#endif
Miaohe Lin9096bbe2022-04-28 23:15:58 -07004992 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004993
4994out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004995 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004996out2:
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004997#ifdef CONFIG_TMPFS_QUOTA
4998 unregister_quota_format(&shmem_quota_format);
Carlos Maiolinoe09764c2023-07-25 16:45:08 +02004999#endif
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07005000 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005001 shm_mnt = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005002}
Matt Mackall853ac432009-01-06 14:40:20 -08005003
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07005004#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005005static ssize_t shmem_enabled_show(struct kobject *kobj,
Joe Perches79d4d382020-12-14 19:14:53 -08005006 struct kobj_attribute *attr, char *buf)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005007{
Colin Ian King26083eb62019-11-30 17:58:04 -08005008 static const int values[] = {
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005009 SHMEM_HUGE_ALWAYS,
5010 SHMEM_HUGE_WITHIN_SIZE,
5011 SHMEM_HUGE_ADVISE,
5012 SHMEM_HUGE_NEVER,
5013 SHMEM_HUGE_DENY,
5014 SHMEM_HUGE_FORCE,
5015 };
Joe Perches79d4d382020-12-14 19:14:53 -08005016 int len = 0;
5017 int i;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005018
Joe Perches79d4d382020-12-14 19:14:53 -08005019 for (i = 0; i < ARRAY_SIZE(values); i++) {
5020 len += sysfs_emit_at(buf, len,
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07005021 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
5022 i ? " " : "", shmem_format_huge(values[i]));
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005023 }
Joe Perches79d4d382020-12-14 19:14:53 -08005024 len += sysfs_emit_at(buf, len, "\n");
5025
5026 return len;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005027}
5028
5029static ssize_t shmem_enabled_store(struct kobject *kobj,
5030 struct kobj_attribute *attr, const char *buf, size_t count)
5031{
5032 char tmp[16];
5033 int huge;
5034
5035 if (count + 1 > sizeof(tmp))
5036 return -EINVAL;
5037 memcpy(tmp, buf, count);
5038 tmp[count] = '\0';
5039 if (count && tmp[count - 1] == '\n')
5040 tmp[count - 1] = '\0';
5041
5042 huge = shmem_parse_huge(tmp);
5043 if (huge == -EINVAL)
5044 return -EINVAL;
5045 if (!has_transparent_hugepage() &&
5046 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
5047 return -EINVAL;
5048
Baolin Wang4b989952024-06-11 18:11:07 +08005049 /* Do not override huge allocation policy with non-PMD sized mTHP */
5050 if (huge == SHMEM_HUGE_FORCE &&
5051 huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
5052 return -EINVAL;
5053
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005054 shmem_huge = huge;
Kirill A. Shutemov435c0b872017-08-25 15:55:33 -07005055 if (shmem_huge > SHMEM_HUGE_DENY)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07005056 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
5057 return count;
5058}
5059
Miaohe Lin4bfa8ad2022-03-22 14:40:04 -07005060struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
Baolin Wang4b989952024-06-11 18:11:07 +08005061static DEFINE_SPINLOCK(huge_shmem_orders_lock);
5062
5063static ssize_t thpsize_shmem_enabled_show(struct kobject *kobj,
5064 struct kobj_attribute *attr, char *buf)
5065{
5066 int order = to_thpsize(kobj)->order;
5067 const char *output;
5068
5069 if (test_bit(order, &huge_shmem_orders_always))
5070 output = "[always] inherit within_size advise never";
5071 else if (test_bit(order, &huge_shmem_orders_inherit))
5072 output = "always [inherit] within_size advise never";
5073 else if (test_bit(order, &huge_shmem_orders_within_size))
5074 output = "always inherit [within_size] advise never";
5075 else if (test_bit(order, &huge_shmem_orders_madvise))
5076 output = "always inherit within_size [advise] never";
5077 else
5078 output = "always inherit within_size advise [never]";
5079
5080 return sysfs_emit(buf, "%s\n", output);
5081}
5082
5083static ssize_t thpsize_shmem_enabled_store(struct kobject *kobj,
5084 struct kobj_attribute *attr,
5085 const char *buf, size_t count)
5086{
5087 int order = to_thpsize(kobj)->order;
5088 ssize_t ret = count;
5089
5090 if (sysfs_streq(buf, "always")) {
5091 spin_lock(&huge_shmem_orders_lock);
5092 clear_bit(order, &huge_shmem_orders_inherit);
5093 clear_bit(order, &huge_shmem_orders_madvise);
5094 clear_bit(order, &huge_shmem_orders_within_size);
5095 set_bit(order, &huge_shmem_orders_always);
5096 spin_unlock(&huge_shmem_orders_lock);
5097 } else if (sysfs_streq(buf, "inherit")) {
5098 /* Do not override huge allocation policy with non-PMD sized mTHP */
5099 if (shmem_huge == SHMEM_HUGE_FORCE &&
5100 order != HPAGE_PMD_ORDER)
5101 return -EINVAL;
5102
5103 spin_lock(&huge_shmem_orders_lock);
5104 clear_bit(order, &huge_shmem_orders_always);
5105 clear_bit(order, &huge_shmem_orders_madvise);
5106 clear_bit(order, &huge_shmem_orders_within_size);
5107 set_bit(order, &huge_shmem_orders_inherit);
5108 spin_unlock(&huge_shmem_orders_lock);
5109 } else if (sysfs_streq(buf, "within_size")) {
5110 spin_lock(&huge_shmem_orders_lock);
5111 clear_bit(order, &huge_shmem_orders_always);
5112 clear_bit(order, &huge_shmem_orders_inherit);
5113 clear_bit(order, &huge_shmem_orders_madvise);
5114 set_bit(order, &huge_shmem_orders_within_size);
5115 spin_unlock(&huge_shmem_orders_lock);
Bang Li843a2e22024-06-28 11:23:27 +08005116 } else if (sysfs_streq(buf, "advise")) {
Baolin Wang4b989952024-06-11 18:11:07 +08005117 spin_lock(&huge_shmem_orders_lock);
5118 clear_bit(order, &huge_shmem_orders_always);
5119 clear_bit(order, &huge_shmem_orders_inherit);
5120 clear_bit(order, &huge_shmem_orders_within_size);
5121 set_bit(order, &huge_shmem_orders_madvise);
5122 spin_unlock(&huge_shmem_orders_lock);
5123 } else if (sysfs_streq(buf, "never")) {
5124 spin_lock(&huge_shmem_orders_lock);
5125 clear_bit(order, &huge_shmem_orders_always);
5126 clear_bit(order, &huge_shmem_orders_inherit);
5127 clear_bit(order, &huge_shmem_orders_within_size);
5128 clear_bit(order, &huge_shmem_orders_madvise);
5129 spin_unlock(&huge_shmem_orders_lock);
5130 } else {
5131 ret = -EINVAL;
5132 }
5133
5134 return ret;
5135}
5136
5137struct kobj_attribute thpsize_shmem_enabled_attr =
5138 __ATTR(shmem_enabled, 0644, thpsize_shmem_enabled_show, thpsize_shmem_enabled_store);
Matthew Wilcox (Oracle)396bcc52020-04-06 20:04:35 -07005139#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07005140
Matt Mackall853ac432009-01-06 14:40:20 -08005141#else /* !CONFIG_SHMEM */
5142
5143/*
5144 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
5145 *
5146 * This is intended for small system where the benefits of the full
5147 * shmem code (swap-backed and resource-limited) are outweighed by
5148 * their complexity. On systems without swap this code should be
5149 * effectively equivalent, but much lighter weight.
5150 */
5151
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07005152static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08005153 .name = "tmpfs",
David Howellsf3235622019-03-25 16:38:31 +00005154 .init_fs_context = ramfs_init_fs_context,
Al Virod7167b12019-09-07 07:23:15 -04005155 .parameters = ramfs_fs_parameters,
Roberto Sassu36ce9d72023-06-07 18:15:23 +02005156 .kill_sb = ramfs_kill_sb,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08005157 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08005158};
5159
Miaohe Lin9096bbe2022-04-28 23:15:58 -07005160void __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08005161{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07005162 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08005163
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07005164 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08005165 BUG_ON(IS_ERR(shm_mnt));
Matt Mackall853ac432009-01-06 14:40:20 -08005166}
5167
Christoph Hellwig10a9c492022-01-21 22:14:57 -08005168int shmem_unuse(unsigned int type)
Matt Mackall853ac432009-01-06 14:40:20 -08005169{
5170 return 0;
5171}
5172
Alexey Gladkovd7c9e992021-04-22 14:27:14 +02005173int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
Hugh Dickins3f96b792009-09-21 17:03:37 -07005174{
5175 return 0;
5176}
5177
Hugh Dickins24513262012-01-20 14:34:21 -08005178void shmem_unlock_mapping(struct address_space *mapping)
5179{
5180}
5181
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07005182#ifdef CONFIG_MMU
5183unsigned long shmem_get_unmapped_area(struct file *file,
5184 unsigned long addr, unsigned long len,
5185 unsigned long pgoff, unsigned long flags)
5186{
Rick Edgecombe529ce232024-03-25 19:16:44 -07005187 return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07005188}
5189#endif
5190
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07005191void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07005192{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07005193 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07005194}
5195EXPORT_SYMBOL_GPL(shmem_truncate_range);
5196
Hugh Dickins0b0a0802009-02-24 20:51:52 +00005197#define shmem_vm_ops generic_file_vm_ops
Pasha Tatashind09e8ca2022-11-15 02:06:01 +00005198#define shmem_anon_vm_ops generic_file_vm_ops
Hugh Dickins0b0a0802009-02-24 20:51:52 +00005199#define shmem_file_operations ramfs_file_operations
Hugh Dickins0b0a0802009-02-24 20:51:52 +00005200#define shmem_acct_size(flags, size) 0
5201#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08005202
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07005203static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
5204 struct super_block *sb, struct inode *dir,
5205 umode_t mode, dev_t dev, unsigned long flags)
Carlos Maiolino71480662023-07-25 16:45:05 +02005206{
5207 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
5208 return inode ? inode : ERR_PTR(-ENOSPC);
5209}
5210
Matt Mackall853ac432009-01-06 14:40:20 -08005211#endif /* CONFIG_SHMEM */
5212
5213/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005214
Hugh Dickins9be7d5b2023-09-29 20:28:50 -07005215static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name,
5216 loff_t size, unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005217{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005218 struct inode *inode;
Al Viro93dec2d2018-07-08 23:02:03 -04005219 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005220
Matthew Auld703321b2017-10-06 23:18:13 +01005221 if (IS_ERR(mnt))
5222 return ERR_CAST(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005223
Hugh Dickins285b2c42011-08-03 16:21:20 -07005224 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005225 return ERR_PTR(-EINVAL);
5226
5227 if (shmem_acct_size(flags, size))
5228 return ERR_PTR(-ENOMEM);
5229
Giuseppe Scrivano7a80e5b2023-01-20 10:43:46 +01005230 if (is_idmapped_mnt(mnt))
5231 return ERR_PTR(-EINVAL);
5232
5233 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
5234 S_IFREG | S_IRWXUGO, 0, flags);
Carlos Maiolino71480662023-07-25 16:45:05 +02005235 if (IS_ERR(inode)) {
Al Virodac2d1f2018-06-09 09:27:41 -04005236 shmem_unacct_size(flags, size);
Carlos Maiolino71480662023-07-25 16:45:05 +02005237 return ERR_CAST(inode);
Al Virodac2d1f2018-06-09 09:27:41 -04005238 }
Eric Parisc7277092013-12-02 11:24:19 +00005239 inode->i_flags |= i_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005240 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02005241 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05005242 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
Al Viro93dec2d2018-07-08 23:02:03 -04005243 if (!IS_ERR(res))
5244 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
5245 &shmem_file_operations);
Al Viro26567cd2013-03-01 20:22:53 -05005246 if (IS_ERR(res))
Al Viro93dec2d2018-07-08 23:02:03 -04005247 iput(inode);
Al Viro6b4d0b22013-02-14 21:37:26 -05005248 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005249}
Eric Parisc7277092013-12-02 11:24:19 +00005250
5251/**
5252 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
5253 * kernel internal. There will be NO LSM permission checks against the
5254 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07005255 * higher layer. The users are the big_key and shm implementations. LSM
5256 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00005257 * @name: name for dentry (to be seen in /proc/<pid>/maps
5258 * @size: size to be set for the file
5259 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5260 */
5261struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
5262{
Matthew Auld703321b2017-10-06 23:18:13 +01005263 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
Eric Parisc7277092013-12-02 11:24:19 +00005264}
Christoph Hellwigbe9d9362024-02-19 07:27:14 +01005265EXPORT_SYMBOL_GPL(shmem_kernel_file_setup);
Eric Parisc7277092013-12-02 11:24:19 +00005266
5267/**
5268 * shmem_file_setup - get an unlinked file living in tmpfs
5269 * @name: name for dentry (to be seen in /proc/<pid>/maps
5270 * @size: size to be set for the file
5271 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5272 */
5273struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
5274{
Matthew Auld703321b2017-10-06 23:18:13 +01005275 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
Eric Parisc7277092013-12-02 11:24:19 +00005276}
Keith Packard395e0dd2008-06-20 00:08:06 -07005277EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005278
Randy Dunlap46711812008-03-19 17:00:41 -07005279/**
Matthew Auld703321b2017-10-06 23:18:13 +01005280 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
5281 * @mnt: the tmpfs mount where the file will be created
5282 * @name: name for dentry (to be seen in /proc/<pid>/maps
5283 * @size: size to be set for the file
5284 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5285 */
5286struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
5287 loff_t size, unsigned long flags)
5288{
5289 return __shmem_file_setup(mnt, name, size, flags, 0);
5290}
5291EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
5292
5293/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07005294 * shmem_zero_setup - setup a shared anonymous mapping
Peter Collingbourne45e55302020-08-06 23:23:37 -07005295 * @vma: the vma to be mmapped is prepared by do_mmap
Linus Torvalds1da177e2005-04-16 15:20:36 -07005296 */
5297int shmem_zero_setup(struct vm_area_struct *vma)
5298{
5299 struct file *file;
5300 loff_t size = vma->vm_end - vma->vm_start;
5301
Hugh Dickins66fc1302015-06-14 09:48:09 -07005302 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07005303 * Cloning a new file under mmap_lock leads to a lock ordering conflict
Hugh Dickins66fc1302015-06-14 09:48:09 -07005304 * between XFS directory reading and selinux: since this file is only
5305 * accessible to the user through its mapping, use S_PRIVATE flag to
5306 * bypass file security, in the same way as shmem_kernel_file_setup().
5307 */
Matthew Auld703321b2017-10-06 23:18:13 +01005308 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005309 if (IS_ERR(file))
5310 return PTR_ERR(file);
5311
5312 if (vma->vm_file)
5313 fput(vma->vm_file);
5314 vma->vm_file = file;
Pasha Tatashind09e8ca2022-11-15 02:06:01 +00005315 vma->vm_ops = &shmem_anon_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07005316
Linus Torvalds1da177e2005-04-16 15:20:36 -07005317 return 0;
5318}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07005319
5320/**
Matthew Wilcox (Oracle)f01b2b32023-02-06 16:25:20 +00005321 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
5322 * @mapping: the folio's address_space
5323 * @index: the folio index
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07005324 * @gfp: the page allocator flags to use if allocating
5325 *
5326 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
5327 * with any new page allocations done using the specified allocation flags.
Matthew Wilcox (Oracle)7e0a1262022-04-29 11:53:28 -04005328 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07005329 * suit tmpfs, since it may have pages in swapcache, and needs to find those
5330 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
5331 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07005332 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
5333 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07005334 */
Matthew Wilcox (Oracle)f01b2b32023-02-06 16:25:20 +00005335struct folio *shmem_read_folio_gfp(struct address_space *mapping,
5336 pgoff_t index, gfp_t gfp)
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07005337{
Hugh Dickins68da9f02011-07-25 17:12:34 -07005338#ifdef CONFIG_SHMEM
5339 struct inode *inode = mapping->host;
Matthew Wilcox (Oracle)a3a9c392022-09-02 20:46:19 +01005340 struct folio *folio;
Hugh Dickins68da9f02011-07-25 17:12:34 -07005341 int error;
5342
Rik van Riele1e4cfd2024-09-03 11:19:28 -04005343 error = shmem_get_folio_gfp(inode, index, 0, &folio, SGP_CACHE,
Hugh Dickinse3e1a5062023-09-29 20:26:53 -07005344 gfp, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07005345 if (error)
Yang Shia7605422022-01-14 14:05:19 -08005346 return ERR_PTR(error);
5347
Matthew Wilcox (Oracle)a3a9c392022-09-02 20:46:19 +01005348 folio_unlock(folio);
Matthew Wilcox (Oracle)f01b2b32023-02-06 16:25:20 +00005349 return folio;
5350#else
5351 /*
5352 * The tiny !SHMEM case uses ramfs without swap
5353 */
5354 return mapping_read_folio_gfp(mapping, index, gfp);
5355#endif
5356}
5357EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
5358
5359struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
5360 pgoff_t index, gfp_t gfp)
5361{
5362 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
5363 struct page *page;
5364
5365 if (IS_ERR(folio))
5366 return &folio->page;
5367
Matthew Wilcox (Oracle)a3a9c392022-09-02 20:46:19 +01005368 page = folio_file_page(folio, index);
Yang Shia7605422022-01-14 14:05:19 -08005369 if (PageHWPoison(page)) {
Matthew Wilcox (Oracle)a3a9c392022-09-02 20:46:19 +01005370 folio_put(folio);
Yang Shia7605422022-01-14 14:05:19 -08005371 return ERR_PTR(-EIO);
5372 }
5373
Hugh Dickins68da9f02011-07-25 17:12:34 -07005374 return page;
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07005375}
5376EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);