shmem: move shmem_mapping out of line
[linux-2.6-block.git] / mm / shmem.c
CommitLineData
1da177e4
LT
1/*
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.
6922c0c7
HD
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
0edd73b3 11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
1da177e4
LT
12 * 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 *
853ac43a
MM
18 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
1da177e4
LT
21 * This file is released under the GPL.
22 */
23
853ac43a
MM
24#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
250297ed 28#include <linux/ramfs.h>
caefba17 29#include <linux/pagemap.h>
853ac43a 30#include <linux/file.h>
e408e695 31#include <linux/fileattr.h>
853ac43a 32#include <linux/mm.h>
46c9a946 33#include <linux/random.h>
174cd4b1 34#include <linux/sched/signal.h>
b95f1b31 35#include <linux/export.h>
5ff2121a 36#include <linux/shmem_fs.h>
853ac43a 37#include <linux/swap.h>
e2e40f2c 38#include <linux/uio.h>
749df87b 39#include <linux/hugetlb.h>
626c3920 40#include <linux/fs_parser.h>
86a2f3f2 41#include <linux/swapfile.h>
36f05cab 42#include <linux/iversion.h>
014bb1de 43#include "swap.h"
95cc09d6 44
68279f9c 45static struct vfsmount *shm_mnt __ro_after_init;
853ac43a
MM
46
47#ifdef CONFIG_SHMEM
1da177e4
LT
48/*
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
39f0247d 54#include <linux/xattr.h>
a5694255 55#include <linux/exportfs.h>
1c7c474c 56#include <linux/posix_acl.h>
feda821e 57#include <linux/posix_acl_xattr.h>
1da177e4 58#include <linux/mman.h>
1da177e4
LT
59#include <linux/string.h>
60#include <linux/slab.h>
61#include <linux/backing-dev.h>
1da177e4 62#include <linux/writeback.h>
bda97eab 63#include <linux/pagevec.h>
41ffe5d5 64#include <linux/percpu_counter.h>
83e4fa9c 65#include <linux/falloc.h>
708e3508 66#include <linux/splice.h>
1da177e4
LT
67#include <linux/security.h>
68#include <linux/swapops.h>
69#include <linux/mempolicy.h>
70#include <linux/namei.h>
b00dc3ad 71#include <linux/ctype.h>
304dbdb7 72#include <linux/migrate.h>
c1f60a5a 73#include <linux/highmem.h>
680d794b 74#include <linux/seq_file.h>
92562927 75#include <linux/magic.h>
9183df25 76#include <linux/syscalls.h>
40e041a2 77#include <linux/fcntl.h>
9183df25 78#include <uapi/linux/memfd.h>
4c27fe4c 79#include <linux/rmap.h>
2b4db796 80#include <linux/uuid.h>
e09764cf 81#include <linux/quotaops.h>
1e2f2d31 82#include <linux/rcupdate_wait.h>
304dbdb7 83
7c0f6ba6 84#include <linux/uaccess.h>
1da177e4 85
dd56b046
MG
86#include "internal.h"
87
09cbfeaf
KS
88#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
1da177e4 90
1da177e4
LT
91/* Pretend that each entry is of this size in directory's i_size */
92#define BOGO_DIRENT_SIZE 20
93
e07c469e
HD
94/* Pretend that one inode + its dentry occupy this much memory */
95#define BOGO_INODE_SIZE 1024
96
69f07ec9
HD
97/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
98#define SHORT_SYMLINK_LEN 128
99
1aac1400 100/*
f00cdc6d 101 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
9608703e 102 * inode->i_private (with i_rwsem making sure that it has only one user at
f00cdc6d 103 * a time): we would prefer not to enlarge the shmem inode just for that.
1aac1400
HD
104 */
105struct shmem_falloc {
8e205f77 106 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
1aac1400
HD
107 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
0b5071dd
AV
113struct 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;
ea3271f7 120 bool full_inums;
0b5071dd
AV
121 int huge;
122 int seen;
2c6efe9c 123 bool noswap;
e09764cf 124 unsigned short quota_types;
de4c0e7c 125 struct shmem_quota_limits qlimits;
0b5071dd
AV
126#define SHMEM_SEEN_BLOCKS 1
127#define SHMEM_SEEN_INODES 2
128#define SHMEM_SEEN_HUGE 4
ea3271f7 129#define SHMEM_SEEN_INUMS 8
2c6efe9c 130#define SHMEM_SEEN_NOSWAP 16
e09764cf 131#define SHMEM_SEEN_QUOTA 32
0b5071dd
AV
132};
133
b76db735 134#ifdef CONFIG_TMPFS
680d794b 135static unsigned long shmem_default_max_blocks(void)
136{
ca79b0c2 137 return totalram_pages() / 2;
680d794b 138}
139
140static unsigned long shmem_default_max_inodes(void)
141{
ca79b0c2
AK
142 unsigned long nr_pages = totalram_pages();
143
e07c469e
HD
144 return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
145 ULONG_MAX / BOGO_INODE_SIZE);
680d794b 146}
b76db735 147#endif
680d794b 148
da08e9b7 149static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
054a9f7c
HD
150 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
151 struct mm_struct *fault_mm, vm_fault_t *fault_type);
1da177e4 152
1da177e4
LT
153static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
154{
155 return sb->s_fs_info;
156}
157
158/*
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
163 */
164static inline int shmem_acct_size(unsigned long flags, loff_t size)
165{
0b0a0806 166 return (flags & VM_NORESERVE) ?
191c5424 167 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
1da177e4
LT
168}
169
170static inline void shmem_unacct_size(unsigned long flags, loff_t size)
171{
0b0a0806 172 if (!(flags & VM_NORESERVE))
1da177e4
LT
173 vm_unacct_memory(VM_ACCT(size));
174}
175
77142517
KK
176static inline int shmem_reacct_size(unsigned long flags,
177 loff_t oldsize, loff_t newsize)
178{
179 if (!(flags & VM_NORESERVE)) {
180 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
181 return security_vm_enough_memory_mm(current->mm,
182 VM_ACCT(newsize) - VM_ACCT(oldsize));
183 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
184 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
185 }
186 return 0;
187}
188
1da177e4
LT
189/*
190 * ... whereas tmpfs objects are accounted incrementally as
75edd345 191 * pages are allocated, in order to allow large sparse files.
4199f51a 192 * shmem_get_folio reports shmem_acct_blocks failure as -ENOSPC not -ENOMEM,
1da177e4
LT
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
194 */
4199f51a 195static inline int shmem_acct_blocks(unsigned long flags, long pages)
1da177e4 196{
800d8c63
KS
197 if (!(flags & VM_NORESERVE))
198 return 0;
199
200 return security_vm_enough_memory_mm(current->mm,
201 pages * VM_ACCT(PAGE_SIZE));
1da177e4
LT
202}
203
204static inline void shmem_unacct_blocks(unsigned long flags, long pages)
205{
0b0a0806 206 if (flags & VM_NORESERVE)
09cbfeaf 207 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
1da177e4
LT
208}
209
4199f51a 210static int shmem_inode_acct_blocks(struct inode *inode, long pages)
0f079694
MR
211{
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
c7e263ab 214 int err = -ENOSPC;
0f079694 215
4199f51a 216 if (shmem_acct_blocks(info->flags, pages))
c7e263ab 217 return err;
0f079694 218
3c1b7528 219 might_sleep(); /* when quotas */
0f079694 220 if (sbinfo->max_blocks) {
beb98686
HD
221 if (!percpu_counter_limited_add(&sbinfo->used_blocks,
222 sbinfo->max_blocks, pages))
0f079694 223 goto unacct;
e09764cf
CM
224
225 err = dquot_alloc_block_nodirty(inode, pages);
beb98686
HD
226 if (err) {
227 percpu_counter_sub(&sbinfo->used_blocks, pages);
e09764cf 228 goto unacct;
beb98686 229 }
e09764cf
CM
230 } else {
231 err = dquot_alloc_block_nodirty(inode, pages);
232 if (err)
233 goto unacct;
0f079694
MR
234 }
235
c7e263ab 236 return 0;
0f079694
MR
237
238unacct:
239 shmem_unacct_blocks(info->flags, pages);
c7e263ab 240 return err;
0f079694
MR
241}
242
3c1b7528 243static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
0f079694
MR
244{
245 struct shmem_inode_info *info = SHMEM_I(inode);
246 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
247
3c1b7528 248 might_sleep(); /* when quotas */
e09764cf
CM
249 dquot_free_block_nodirty(inode, pages);
250
0f079694
MR
251 if (sbinfo->max_blocks)
252 percpu_counter_sub(&sbinfo->used_blocks, pages);
253 shmem_unacct_blocks(info->flags, pages);
254}
255
759b9775 256static const struct super_operations shmem_ops;
aefacb20 257static const struct address_space_operations shmem_aops;
15ad7cdc 258static const struct file_operations shmem_file_operations;
92e1d5be
AV
259static const struct inode_operations shmem_inode_operations;
260static const struct inode_operations shmem_dir_inode_operations;
261static const struct inode_operations shmem_special_inode_operations;
f0f37e2f 262static const struct vm_operations_struct shmem_vm_ops;
d09e8ca6 263static const struct vm_operations_struct shmem_anon_vm_ops;
779750d2 264static struct file_system_type shmem_fs_type;
1da177e4 265
aefacb20
CH
266bool shmem_mapping(struct address_space *mapping)
267{
268 return mapping->a_ops == &shmem_aops;
269}
270EXPORT_SYMBOL_GPL(shmem_mapping);
271
d09e8ca6
PT
272bool vma_is_anon_shmem(struct vm_area_struct *vma)
273{
274 return vma->vm_ops == &shmem_anon_vm_ops;
275}
276
b0506e48
MR
277bool vma_is_shmem(struct vm_area_struct *vma)
278{
d09e8ca6 279 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
b0506e48
MR
280}
281
1da177e4 282static LIST_HEAD(shmem_swaplist);
cb5f7b9a 283static DEFINE_MUTEX(shmem_swaplist_mutex);
1da177e4 284
e09764cf
CM
285#ifdef CONFIG_TMPFS_QUOTA
286
287static int shmem_enable_quotas(struct super_block *sb,
288 unsigned short quota_types)
289{
290 int type, err = 0;
291
292 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
293 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
294 if (!(quota_types & (1 << type)))
295 continue;
296 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
297 DQUOT_USAGE_ENABLED |
298 DQUOT_LIMITS_ENABLED);
299 if (err)
300 goto out_err;
301 }
302 return 0;
303
304out_err:
305 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
306 type, err);
307 for (type--; type >= 0; type--)
308 dquot_quota_off(sb, type);
309 return err;
310}
311
312static void shmem_disable_quotas(struct super_block *sb)
313{
314 int type;
315
316 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
317 dquot_quota_off(sb, type);
318}
319
320static struct dquot **shmem_get_dquots(struct inode *inode)
321{
322 return SHMEM_I(inode)->i_dquot;
323}
324#endif /* CONFIG_TMPFS_QUOTA */
325
e809d5f0
CD
326/*
327 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
328 * produces a novel ino for the newly allocated inode.
329 *
330 * It may also be called when making a hard link to permit the space needed by
331 * each dentry. However, in that case, no new inode number is needed since that
332 * internally draws from another pool of inode numbers (currently global
333 * get_next_ino()). This case is indicated by passing NULL as inop.
334 */
335#define SHMEM_INO_BATCH 1024
336static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
5b04c689
PE
337{
338 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
e809d5f0
CD
339 ino_t ino;
340
341 if (!(sb->s_flags & SB_KERNMOUNT)) {
bf11b9a8 342 raw_spin_lock(&sbinfo->stat_lock);
bb3e96d6 343 if (sbinfo->max_inodes) {
e07c469e 344 if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
bf11b9a8 345 raw_spin_unlock(&sbinfo->stat_lock);
bb3e96d6
BS
346 return -ENOSPC;
347 }
e07c469e 348 sbinfo->free_ispace -= BOGO_INODE_SIZE;
5b04c689 349 }
e809d5f0
CD
350 if (inop) {
351 ino = sbinfo->next_ino++;
352 if (unlikely(is_zero_ino(ino)))
353 ino = sbinfo->next_ino++;
ea3271f7
CD
354 if (unlikely(!sbinfo->full_inums &&
355 ino > UINT_MAX)) {
e809d5f0
CD
356 /*
357 * Emulate get_next_ino uint wraparound for
358 * compatibility
359 */
ea3271f7
CD
360 if (IS_ENABLED(CONFIG_64BIT))
361 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
362 __func__, MINOR(sb->s_dev));
363 sbinfo->next_ino = 1;
364 ino = sbinfo->next_ino++;
e809d5f0
CD
365 }
366 *inop = ino;
367 }
bf11b9a8 368 raw_spin_unlock(&sbinfo->stat_lock);
e809d5f0
CD
369 } else if (inop) {
370 /*
371 * __shmem_file_setup, one of our callers, is lock-free: it
372 * doesn't hold stat_lock in shmem_reserve_inode since
373 * max_inodes is always 0, and is called from potentially
374 * unknown contexts. As such, use a per-cpu batched allocator
375 * which doesn't require the per-sb stat_lock unless we are at
376 * the batch boundary.
ea3271f7
CD
377 *
378 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
379 * shmem mounts are not exposed to userspace, so we don't need
380 * to worry about things like glibc compatibility.
e809d5f0
CD
381 */
382 ino_t *next_ino;
bf11b9a8 383
e809d5f0
CD
384 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
385 ino = *next_ino;
386 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
bf11b9a8 387 raw_spin_lock(&sbinfo->stat_lock);
e809d5f0
CD
388 ino = sbinfo->next_ino;
389 sbinfo->next_ino += SHMEM_INO_BATCH;
bf11b9a8 390 raw_spin_unlock(&sbinfo->stat_lock);
e809d5f0
CD
391 if (unlikely(is_zero_ino(ino)))
392 ino++;
393 }
394 *inop = ino;
395 *next_ino = ++ino;
396 put_cpu();
5b04c689 397 }
e809d5f0 398
5b04c689
PE
399 return 0;
400}
401
2daf18a7 402static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
5b04c689
PE
403{
404 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
405 if (sbinfo->max_inodes) {
bf11b9a8 406 raw_spin_lock(&sbinfo->stat_lock);
2daf18a7 407 sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
bf11b9a8 408 raw_spin_unlock(&sbinfo->stat_lock);
5b04c689
PE
409 }
410}
411
46711810 412/**
41ffe5d5 413 * shmem_recalc_inode - recalculate the block usage of an inode
1da177e4 414 * @inode: inode to recalc
3c1b7528
HD
415 * @alloced: the change in number of pages allocated to inode
416 * @swapped: the change in number of pages swapped from inode
1da177e4
LT
417 *
418 * We have to calculate the free blocks since the mm can drop
419 * undirtied hole pages behind our back.
420 *
421 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
422 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
1da177e4 423 */
3c1b7528 424static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
1da177e4
LT
425{
426 struct shmem_inode_info *info = SHMEM_I(inode);
427 long freed;
428
3c1b7528
HD
429 spin_lock(&info->lock);
430 info->alloced += alloced;
431 info->swapped += swapped;
432 freed = info->alloced - info->swapped -
433 READ_ONCE(inode->i_mapping->nrpages);
434 /*
435 * Special case: whereas normally shmem_recalc_inode() is called
436 * after i_mapping->nrpages has already been adjusted (up or down),
437 * shmem_writepage() has to raise swapped before nrpages is lowered -
438 * to stop a racing shmem_recalc_inode() from thinking that a page has
439 * been freed. Compensate here, to avoid the need for a followup call.
440 */
441 if (swapped > 0)
442 freed += swapped;
443 if (freed > 0)
1da177e4 444 info->alloced -= freed;
3c1b7528
HD
445 spin_unlock(&info->lock);
446
447 /* The quota case may block */
448 if (freed > 0)
0f079694 449 shmem_inode_unacct_blocks(inode, freed);
1da177e4
LT
450}
451
800d8c63
KS
452bool shmem_charge(struct inode *inode, long pages)
453{
509f0069 454 struct address_space *mapping = inode->i_mapping;
800d8c63 455
4199f51a 456 if (shmem_inode_acct_blocks(inode, pages))
800d8c63 457 return false;
b1cc94ab 458
aaa52e34 459 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
509f0069
HD
460 xa_lock_irq(&mapping->i_pages);
461 mapping->nrpages += pages;
462 xa_unlock_irq(&mapping->i_pages);
800d8c63 463
3c1b7528 464 shmem_recalc_inode(inode, pages, 0);
800d8c63
KS
465 return true;
466}
467
468void shmem_uncharge(struct inode *inode, long pages)
469{
3c1b7528 470 /* pages argument is currently unused: keep it to help debugging */
6ffcd825 471 /* nrpages adjustment done by __filemap_remove_folio() or caller */
aaa52e34 472
3c1b7528 473 shmem_recalc_inode(inode, 0, 0);
800d8c63
KS
474}
475
7a5d0fbb 476/*
62f945b6 477 * Replace item expected in xarray by a new item, while holding xa_lock.
7a5d0fbb 478 */
62f945b6 479static int shmem_replace_entry(struct address_space *mapping,
7a5d0fbb
HD
480 pgoff_t index, void *expected, void *replacement)
481{
62f945b6 482 XA_STATE(xas, &mapping->i_pages, index);
6dbaf22c 483 void *item;
7a5d0fbb
HD
484
485 VM_BUG_ON(!expected);
6dbaf22c 486 VM_BUG_ON(!replacement);
62f945b6 487 item = xas_load(&xas);
7a5d0fbb
HD
488 if (item != expected)
489 return -ENOENT;
62f945b6 490 xas_store(&xas, replacement);
7a5d0fbb
HD
491 return 0;
492}
493
d1899228
HD
494/*
495 * Sometimes, before we decide whether to proceed or to fail, we must check
496 * that an entry was not already brought back from swap by a racing thread.
497 *
498 * Checking page is not enough: by the time a SwapCache page is locked, it
499 * might be reused, and again be SwapCache, using the same swap as before.
500 */
501static bool shmem_confirm_swap(struct address_space *mapping,
502 pgoff_t index, swp_entry_t swap)
503{
a12831bf 504 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
d1899228
HD
505}
506
5a6e75f8
KS
507/*
508 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
509 *
510 * SHMEM_HUGE_NEVER:
511 * disables huge pages for the mount;
512 * SHMEM_HUGE_ALWAYS:
513 * enables huge pages for the mount;
514 * SHMEM_HUGE_WITHIN_SIZE:
515 * only allocate huge pages if the page will be fully within i_size,
516 * also respect fadvise()/madvise() hints;
517 * SHMEM_HUGE_ADVISE:
518 * only allocate huge pages if requested with fadvise()/madvise();
519 */
520
521#define SHMEM_HUGE_NEVER 0
522#define SHMEM_HUGE_ALWAYS 1
523#define SHMEM_HUGE_WITHIN_SIZE 2
524#define SHMEM_HUGE_ADVISE 3
525
526/*
527 * Special values.
528 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
529 *
530 * SHMEM_HUGE_DENY:
531 * disables huge on shm_mnt and all mounts, for emergency use;
532 * SHMEM_HUGE_FORCE:
533 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
534 *
535 */
536#define SHMEM_HUGE_DENY (-1)
537#define SHMEM_HUGE_FORCE (-2)
538
396bcc52 539#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5a6e75f8
KS
540/* ifdef here to avoid bloating shmem.o when not necessary */
541
5e6e5a12 542static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
5a6e75f8 543
2cf13384
DS
544bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
545 struct mm_struct *mm, unsigned long vm_flags)
c852023e 546{
c852023e 547 loff_t i_size;
c852023e 548
f7cd16a5
XR
549 if (!S_ISREG(inode->i_mode))
550 return false;
2cf13384 551 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
c852023e 552 return false;
7c6c6cc4
ZK
553 if (shmem_huge == SHMEM_HUGE_DENY)
554 return false;
3de0c269
ZK
555 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
556 return true;
5e6e5a12
HD
557
558 switch (SHMEM_SB(inode->i_sb)->huge) {
c852023e
HD
559 case SHMEM_HUGE_ALWAYS:
560 return true;
561 case SHMEM_HUGE_WITHIN_SIZE:
de6ee659 562 index = round_up(index + 1, HPAGE_PMD_NR);
c852023e 563 i_size = round_up(i_size_read(inode), PAGE_SIZE);
de6ee659 564 if (i_size >> PAGE_SHIFT >= index)
c852023e
HD
565 return true;
566 fallthrough;
567 case SHMEM_HUGE_ADVISE:
2cf13384 568 if (mm && (vm_flags & VM_HUGEPAGE))
5e6e5a12
HD
569 return true;
570 fallthrough;
c852023e 571 default:
c852023e
HD
572 return false;
573 }
574}
5a6e75f8 575
e5f2249a 576#if defined(CONFIG_SYSFS)
5a6e75f8
KS
577static int shmem_parse_huge(const char *str)
578{
579 if (!strcmp(str, "never"))
580 return SHMEM_HUGE_NEVER;
581 if (!strcmp(str, "always"))
582 return SHMEM_HUGE_ALWAYS;
583 if (!strcmp(str, "within_size"))
584 return SHMEM_HUGE_WITHIN_SIZE;
585 if (!strcmp(str, "advise"))
586 return SHMEM_HUGE_ADVISE;
587 if (!strcmp(str, "deny"))
588 return SHMEM_HUGE_DENY;
589 if (!strcmp(str, "force"))
590 return SHMEM_HUGE_FORCE;
591 return -EINVAL;
592}
e5f2249a 593#endif
5a6e75f8 594
e5f2249a 595#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
5a6e75f8
KS
596static const char *shmem_format_huge(int huge)
597{
598 switch (huge) {
599 case SHMEM_HUGE_NEVER:
600 return "never";
601 case SHMEM_HUGE_ALWAYS:
602 return "always";
603 case SHMEM_HUGE_WITHIN_SIZE:
604 return "within_size";
605 case SHMEM_HUGE_ADVISE:
606 return "advise";
607 case SHMEM_HUGE_DENY:
608 return "deny";
609 case SHMEM_HUGE_FORCE:
610 return "force";
611 default:
612 VM_BUG_ON(1);
613 return "bad_val";
614 }
615}
f1f5929c 616#endif
5a6e75f8 617
779750d2
KS
618static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
619 struct shrink_control *sc, unsigned long nr_to_split)
620{
621 LIST_HEAD(list), *pos, *next;
253fd0f0 622 LIST_HEAD(to_remove);
779750d2
KS
623 struct inode *inode;
624 struct shmem_inode_info *info;
05624571 625 struct folio *folio;
779750d2 626 unsigned long batch = sc ? sc->nr_to_scan : 128;
62c9827c 627 int split = 0;
779750d2
KS
628
629 if (list_empty(&sbinfo->shrinklist))
630 return SHRINK_STOP;
631
632 spin_lock(&sbinfo->shrinklist_lock);
633 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
634 info = list_entry(pos, struct shmem_inode_info, shrinklist);
635
636 /* pin the inode */
637 inode = igrab(&info->vfs_inode);
638
639 /* inode is about to be evicted */
640 if (!inode) {
641 list_del_init(&info->shrinklist);
779750d2
KS
642 goto next;
643 }
644
645 /* Check if there's anything to gain */
646 if (round_up(inode->i_size, PAGE_SIZE) ==
647 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
253fd0f0 648 list_move(&info->shrinklist, &to_remove);
779750d2
KS
649 goto next;
650 }
651
652 list_move(&info->shrinklist, &list);
653next:
62c9827c 654 sbinfo->shrinklist_len--;
779750d2
KS
655 if (!--batch)
656 break;
657 }
658 spin_unlock(&sbinfo->shrinklist_lock);
659
253fd0f0
KS
660 list_for_each_safe(pos, next, &to_remove) {
661 info = list_entry(pos, struct shmem_inode_info, shrinklist);
662 inode = &info->vfs_inode;
663 list_del_init(&info->shrinklist);
664 iput(inode);
665 }
666
779750d2
KS
667 list_for_each_safe(pos, next, &list) {
668 int ret;
05624571 669 pgoff_t index;
779750d2
KS
670
671 info = list_entry(pos, struct shmem_inode_info, shrinklist);
672 inode = &info->vfs_inode;
673
b3cd54b2 674 if (nr_to_split && split >= nr_to_split)
62c9827c 675 goto move_back;
779750d2 676
05624571
MWO
677 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
678 folio = filemap_get_folio(inode->i_mapping, index);
66dabbb6 679 if (IS_ERR(folio))
779750d2
KS
680 goto drop;
681
b3cd54b2 682 /* No huge page at the end of the file: nothing to split */
05624571
MWO
683 if (!folio_test_large(folio)) {
684 folio_put(folio);
779750d2
KS
685 goto drop;
686 }
687
b3cd54b2 688 /*
62c9827c
GL
689 * Move the inode on the list back to shrinklist if we failed
690 * to lock the page at this time.
b3cd54b2
KS
691 *
692 * Waiting for the lock may lead to deadlock in the
693 * reclaim path.
694 */
05624571
MWO
695 if (!folio_trylock(folio)) {
696 folio_put(folio);
62c9827c 697 goto move_back;
b3cd54b2
KS
698 }
699
d788f5b3 700 ret = split_folio(folio);
05624571
MWO
701 folio_unlock(folio);
702 folio_put(folio);
779750d2 703
62c9827c 704 /* If split failed move the inode on the list back to shrinklist */
b3cd54b2 705 if (ret)
62c9827c 706 goto move_back;
779750d2
KS
707
708 split++;
709drop:
710 list_del_init(&info->shrinklist);
62c9827c
GL
711 goto put;
712move_back:
713 /*
714 * Make sure the inode is either on the global list or deleted
715 * from any local list before iput() since it could be deleted
716 * in another thread once we put the inode (then the local list
717 * is corrupted).
718 */
719 spin_lock(&sbinfo->shrinklist_lock);
720 list_move(&info->shrinklist, &sbinfo->shrinklist);
721 sbinfo->shrinklist_len++;
722 spin_unlock(&sbinfo->shrinklist_lock);
723put:
779750d2
KS
724 iput(inode);
725 }
726
779750d2
KS
727 return split;
728}
729
730static long shmem_unused_huge_scan(struct super_block *sb,
731 struct shrink_control *sc)
732{
733 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
734
735 if (!READ_ONCE(sbinfo->shrinklist_len))
736 return SHRINK_STOP;
737
738 return shmem_unused_huge_shrink(sbinfo, sc, 0);
739}
740
741static long shmem_unused_huge_count(struct super_block *sb,
742 struct shrink_control *sc)
743{
744 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
745 return READ_ONCE(sbinfo->shrinklist_len);
746}
396bcc52 747#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
5a6e75f8
KS
748
749#define shmem_huge SHMEM_HUGE_DENY
750
2cf13384
DS
751bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
752 struct mm_struct *mm, unsigned long vm_flags)
5e6e5a12
HD
753{
754 return false;
755}
756
779750d2
KS
757static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
758 struct shrink_control *sc, unsigned long nr_to_split)
759{
760 return 0;
761}
396bcc52 762#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
5a6e75f8 763
46f65ec1 764/*
9be7d5b0 765 * Somewhat like filemap_add_folio, but error if expected item has gone.
46f65ec1 766 */
b7dd44a1 767static int shmem_add_to_page_cache(struct folio *folio,
46f65ec1 768 struct address_space *mapping,
054a9f7c 769 pgoff_t index, void *expected, gfp_t gfp)
46f65ec1 770{
b7dd44a1
MWO
771 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
772 long nr = folio_nr_pages(folio);
46f65ec1 773
b7dd44a1
MWO
774 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
775 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
776 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
777 VM_BUG_ON(expected && folio_test_large(folio));
46f65ec1 778
b7dd44a1
MWO
779 folio_ref_add(folio, nr);
780 folio->mapping = mapping;
781 folio->index = index;
b065b432 782
054a9f7c 783 gfp &= GFP_RECLAIM_MASK;
b7dd44a1 784 folio_throttle_swaprate(folio, gfp);
3fea5a49 785
552446a4 786 do {
552446a4 787 xas_lock_irq(&xas);
6b24ca4a
MWO
788 if (expected != xas_find_conflict(&xas)) {
789 xas_set_err(&xas, -EEXIST);
790 goto unlock;
791 }
792 if (expected && xas_find_conflict(&xas)) {
552446a4 793 xas_set_err(&xas, -EEXIST);
552446a4 794 goto unlock;
800d8c63 795 }
b7dd44a1 796 xas_store(&xas, folio);
6b24ca4a
MWO
797 if (xas_error(&xas))
798 goto unlock;
3022fd7a 799 if (folio_test_pmd_mappable(folio))
b7dd44a1 800 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
b7dd44a1
MWO
801 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
802 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
3022fd7a 803 mapping->nrpages += nr;
552446a4
MW
804unlock:
805 xas_unlock_irq(&xas);
806 } while (xas_nomem(&xas, gfp));
807
808 if (xas_error(&xas)) {
054a9f7c
HD
809 folio->mapping = NULL;
810 folio_ref_sub(folio, nr);
811 return xas_error(&xas);
46f65ec1 812 }
552446a4
MW
813
814 return 0;
46f65ec1
HD
815}
816
6922c0c7 817/*
9be7d5b0 818 * Somewhat like filemap_remove_folio, but substitutes swap for @folio.
6922c0c7 819 */
4cd400fd 820static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
6922c0c7 821{
4cd400fd
MWO
822 struct address_space *mapping = folio->mapping;
823 long nr = folio_nr_pages(folio);
6922c0c7
HD
824 int error;
825
b93b0163 826 xa_lock_irq(&mapping->i_pages);
4cd400fd
MWO
827 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
828 folio->mapping = NULL;
829 mapping->nrpages -= nr;
830 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
831 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
b93b0163 832 xa_unlock_irq(&mapping->i_pages);
4cd400fd 833 folio_put(folio);
6922c0c7
HD
834 BUG_ON(error);
835}
836
7a5d0fbb 837/*
c121d3bb 838 * Remove swap entry from page cache, free the swap and its page cache.
7a5d0fbb
HD
839 */
840static int shmem_free_swap(struct address_space *mapping,
841 pgoff_t index, void *radswap)
842{
6dbaf22c 843 void *old;
7a5d0fbb 844
55f3f7ea 845 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
6dbaf22c
JW
846 if (old != radswap)
847 return -ENOENT;
848 free_swap_and_cache(radix_to_swp_entry(radswap));
849 return 0;
7a5d0fbb
HD
850}
851
6a15a370
VB
852/*
853 * Determine (in bytes) how many of the shmem object's pages mapped by the
48131e03 854 * given offsets are swapped out.
6a15a370 855 *
9608703e 856 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
6a15a370
VB
857 * as long as the inode doesn't go away and racy results are not a problem.
858 */
48131e03
VB
859unsigned long shmem_partial_swap_usage(struct address_space *mapping,
860 pgoff_t start, pgoff_t end)
6a15a370 861{
7ae3424f 862 XA_STATE(xas, &mapping->i_pages, start);
6a15a370 863 struct page *page;
48131e03 864 unsigned long swapped = 0;
e5548f85 865 unsigned long max = end - 1;
6a15a370
VB
866
867 rcu_read_lock();
e5548f85 868 xas_for_each(&xas, page, max) {
7ae3424f 869 if (xas_retry(&xas, page))
2cf938aa 870 continue;
3159f943 871 if (xa_is_value(page))
6a15a370 872 swapped++;
e5548f85
HD
873 if (xas.xa_index == max)
874 break;
6a15a370 875 if (need_resched()) {
7ae3424f 876 xas_pause(&xas);
6a15a370 877 cond_resched_rcu();
6a15a370
VB
878 }
879 }
6a15a370
VB
880 rcu_read_unlock();
881
882 return swapped << PAGE_SHIFT;
883}
884
48131e03
VB
885/*
886 * Determine (in bytes) how many of the shmem object's pages mapped by the
887 * given vma is swapped out.
888 *
9608703e 889 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
48131e03
VB
890 * as long as the inode doesn't go away and racy results are not a problem.
891 */
892unsigned long shmem_swap_usage(struct vm_area_struct *vma)
893{
894 struct inode *inode = file_inode(vma->vm_file);
895 struct shmem_inode_info *info = SHMEM_I(inode);
896 struct address_space *mapping = inode->i_mapping;
897 unsigned long swapped;
898
899 /* Be careful as we don't hold info->lock */
900 swapped = READ_ONCE(info->swapped);
901
902 /*
903 * The easier cases are when the shmem object has nothing in swap, or
904 * the vma maps it whole. Then we can simply use the stats that we
905 * already track.
906 */
907 if (!swapped)
908 return 0;
909
910 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
911 return swapped << PAGE_SHIFT;
912
913 /* Here comes the more involved part */
02399c88
PX
914 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
915 vma->vm_pgoff + vma_pages(vma));
48131e03
VB
916}
917
24513264
HD
918/*
919 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
920 */
921void shmem_unlock_mapping(struct address_space *mapping)
922{
105c988f 923 struct folio_batch fbatch;
24513264
HD
924 pgoff_t index = 0;
925
105c988f 926 folio_batch_init(&fbatch);
24513264
HD
927 /*
928 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
929 */
105c988f
MWO
930 while (!mapping_unevictable(mapping) &&
931 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
932 check_move_unevictable_folios(&fbatch);
933 folio_batch_release(&fbatch);
24513264
HD
934 cond_resched();
935 }
7a5d0fbb
HD
936}
937
b9a8a419 938static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
71725ed1 939{
b9a8a419 940 struct folio *folio;
71725ed1 941
b9a8a419 942 /*
a7f5862c 943 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
81914aff 944 * beyond i_size, and reports fallocated folios as holes.
b9a8a419 945 */
81914aff
HD
946 folio = filemap_get_entry(inode->i_mapping, index);
947 if (!folio)
b9a8a419 948 return folio;
81914aff
HD
949 if (!xa_is_value(folio)) {
950 folio_lock(folio);
951 if (folio->mapping == inode->i_mapping)
952 return folio;
953 /* The folio has been swapped out */
954 folio_unlock(folio);
955 folio_put(folio);
956 }
b9a8a419 957 /*
81914aff 958 * But read a folio back from swap if any of it is within i_size
b9a8a419
MWO
959 * (although in some cases this is just a waste of time).
960 */
a7f5862c
MWO
961 folio = NULL;
962 shmem_get_folio(inode, index, &folio, SGP_READ);
963 return folio;
71725ed1
HD
964}
965
7a5d0fbb 966/*
7f4446ee 967 * Remove range of pages and swap entries from page cache, and free them.
1635f6a7 968 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
7a5d0fbb 969 */
1635f6a7
HD
970static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
971 bool unfalloc)
1da177e4 972{
285b2c4f 973 struct address_space *mapping = inode->i_mapping;
1da177e4 974 struct shmem_inode_info *info = SHMEM_I(inode);
09cbfeaf
KS
975 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
0e499ed3 977 struct folio_batch fbatch;
7a5d0fbb 978 pgoff_t indices[PAGEVEC_SIZE];
b9a8a419
MWO
979 struct folio *folio;
980 bool same_folio;
7a5d0fbb 981 long nr_swaps_freed = 0;
285b2c4f 982 pgoff_t index;
bda97eab
HD
983 int i;
984
83e4fa9c
HD
985 if (lend == -1)
986 end = -1; /* unsigned, so actually very big */
bda97eab 987
d144bf62
HD
988 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
989 info->fallocend = start;
990
51dcbdac 991 folio_batch_init(&fbatch);
bda97eab 992 index = start;
3392ca12 993 while (index < end && find_lock_entries(mapping, &index, end - 1,
51dcbdac
MWO
994 &fbatch, indices)) {
995 for (i = 0; i < folio_batch_count(&fbatch); i++) {
b9a8a419 996 folio = fbatch.folios[i];
bda97eab 997
7b774aab 998 if (xa_is_value(folio)) {
1635f6a7
HD
999 if (unfalloc)
1000 continue;
7a5d0fbb 1001 nr_swaps_freed += !shmem_free_swap(mapping,
3392ca12 1002 indices[i], folio);
bda97eab 1003 continue;
7a5d0fbb
HD
1004 }
1005
7b774aab 1006 if (!unfalloc || !folio_test_uptodate(folio))
1e84a3d9 1007 truncate_inode_folio(mapping, folio);
7b774aab 1008 folio_unlock(folio);
bda97eab 1009 }
51dcbdac
MWO
1010 folio_batch_remove_exceptionals(&fbatch);
1011 folio_batch_release(&fbatch);
bda97eab 1012 cond_resched();
bda97eab 1013 }
1da177e4 1014
44bcabd7
HD
1015 /*
1016 * When undoing a failed fallocate, we want none of the partial folio
1017 * zeroing and splitting below, but shall want to truncate the whole
1018 * folio when !uptodate indicates that it was added by this fallocate,
1019 * even when [lstart, lend] covers only a part of the folio.
1020 */
1021 if (unfalloc)
1022 goto whole_folios;
1023
b9a8a419
MWO
1024 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1025 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1026 if (folio) {
1027 same_folio = lend < folio_pos(folio) + folio_size(folio);
1028 folio_mark_dirty(folio);
1029 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
87b11f86 1030 start = folio_next_index(folio);
b9a8a419
MWO
1031 if (same_folio)
1032 end = folio->index;
83e4fa9c 1033 }
b9a8a419
MWO
1034 folio_unlock(folio);
1035 folio_put(folio);
1036 folio = NULL;
83e4fa9c 1037 }
b9a8a419
MWO
1038
1039 if (!same_folio)
1040 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1041 if (folio) {
1042 folio_mark_dirty(folio);
1043 if (!truncate_inode_partial_folio(folio, lstart, lend))
1044 end = folio->index;
1045 folio_unlock(folio);
1046 folio_put(folio);
bda97eab
HD
1047 }
1048
44bcabd7
HD
1049whole_folios:
1050
bda97eab 1051 index = start;
b1a36650 1052 while (index < end) {
bda97eab 1053 cond_resched();
0cd6144a 1054
9fb6beea 1055 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
cf2039af 1056 indices)) {
b1a36650
HD
1057 /* If all gone or hole-punch or unfalloc, we're done */
1058 if (index == start || end != -1)
bda97eab 1059 break;
b1a36650 1060 /* But if truncating, restart to make sure all gone */
bda97eab
HD
1061 index = start;
1062 continue;
1063 }
0e499ed3 1064 for (i = 0; i < folio_batch_count(&fbatch); i++) {
b9a8a419 1065 folio = fbatch.folios[i];
bda97eab 1066
0e499ed3 1067 if (xa_is_value(folio)) {
1635f6a7
HD
1068 if (unfalloc)
1069 continue;
9fb6beea 1070 if (shmem_free_swap(mapping, indices[i], folio)) {
b1a36650 1071 /* Swap was replaced by page: retry */
9fb6beea 1072 index = indices[i];
b1a36650
HD
1073 break;
1074 }
1075 nr_swaps_freed++;
7a5d0fbb
HD
1076 continue;
1077 }
1078
0e499ed3 1079 folio_lock(folio);
800d8c63 1080
0e499ed3 1081 if (!unfalloc || !folio_test_uptodate(folio)) {
0e499ed3 1082 if (folio_mapping(folio) != mapping) {
b1a36650 1083 /* Page was replaced by swap: retry */
0e499ed3 1084 folio_unlock(folio);
9fb6beea 1085 index = indices[i];
b1a36650 1086 break;
1635f6a7 1087 }
0e499ed3
MWO
1088 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1089 folio);
55ac8bbe
DS
1090
1091 if (!folio_test_large(folio)) {
1092 truncate_inode_folio(mapping, folio);
1093 } else if (truncate_inode_partial_folio(folio, lstart, lend)) {
1094 /*
1095 * If we split a page, reset the loop so
1096 * that we pick up the new sub pages.
1097 * Otherwise the THP was entirely
1098 * dropped or the target range was
1099 * zeroed, so just continue the loop as
1100 * is.
1101 */
1102 if (!folio_test_large(folio)) {
1103 folio_unlock(folio);
1104 index = start;
1105 break;
1106 }
1107 }
7a5d0fbb 1108 }
0e499ed3 1109 folio_unlock(folio);
bda97eab 1110 }
0e499ed3
MWO
1111 folio_batch_remove_exceptionals(&fbatch);
1112 folio_batch_release(&fbatch);
bda97eab 1113 }
94c1e62d 1114
3c1b7528 1115 shmem_recalc_inode(inode, 0, -nr_swaps_freed);
1635f6a7 1116}
1da177e4 1117
1635f6a7
HD
1118void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1119{
1120 shmem_undo_range(inode, lstart, lend, false);
cf2766bb 1121 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
36f05cab 1122 inode_inc_iversion(inode);
1da177e4 1123}
94c1e62d 1124EXPORT_SYMBOL_GPL(shmem_truncate_range);
1da177e4 1125
b74d24f7 1126static int shmem_getattr(struct mnt_idmap *idmap,
549c7297 1127 const struct path *path, struct kstat *stat,
a528d35e 1128 u32 request_mask, unsigned int query_flags)
44a30220 1129{
a528d35e 1130 struct inode *inode = path->dentry->d_inode;
44a30220
YZ
1131 struct shmem_inode_info *info = SHMEM_I(inode);
1132
3c1b7528
HD
1133 if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1134 shmem_recalc_inode(inode, 0, 0);
1135
e408e695
TT
1136 if (info->fsflags & FS_APPEND_FL)
1137 stat->attributes |= STATX_ATTR_APPEND;
1138 if (info->fsflags & FS_IMMUTABLE_FL)
1139 stat->attributes |= STATX_ATTR_IMMUTABLE;
1140 if (info->fsflags & FS_NODUMP_FL)
1141 stat->attributes |= STATX_ATTR_NODUMP;
1142 stat->attributes_mask |= (STATX_ATTR_APPEND |
1143 STATX_ATTR_IMMUTABLE |
1144 STATX_ATTR_NODUMP);
0d72b928 1145 generic_fillattr(idmap, request_mask, inode, stat);
89fdcd26 1146
2cf13384 1147 if (shmem_is_huge(inode, 0, false, NULL, 0))
89fdcd26
YS
1148 stat->blksize = HPAGE_PMD_SIZE;
1149
f7cd16a5
XR
1150 if (request_mask & STATX_BTIME) {
1151 stat->result_mask |= STATX_BTIME;
1152 stat->btime.tv_sec = info->i_crtime.tv_sec;
1153 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1154 }
1155
44a30220
YZ
1156 return 0;
1157}
1158
c1632a0f 1159static int shmem_setattr(struct mnt_idmap *idmap,
549c7297 1160 struct dentry *dentry, struct iattr *attr)
1da177e4 1161{
75c3cfa8 1162 struct inode *inode = d_inode(dentry);
40e041a2 1163 struct shmem_inode_info *info = SHMEM_I(inode);
1da177e4 1164 int error;
36f05cab
JL
1165 bool update_mtime = false;
1166 bool update_ctime = true;
1da177e4 1167
7a80e5b8 1168 error = setattr_prepare(idmap, dentry, attr);
db78b877
CH
1169 if (error)
1170 return error;
1171
6fd73538
DV
1172 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1173 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1174 return -EPERM;
1175 }
1176 }
1177
94c1e62d
HD
1178 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1179 loff_t oldsize = inode->i_size;
1180 loff_t newsize = attr->ia_size;
3889e6e7 1181
9608703e 1182 /* protected by i_rwsem */
40e041a2
DH
1183 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1184 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1185 return -EPERM;
1186
94c1e62d 1187 if (newsize != oldsize) {
77142517
KK
1188 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1189 oldsize, newsize);
1190 if (error)
1191 return error;
94c1e62d 1192 i_size_write(inode, newsize);
36f05cab
JL
1193 update_mtime = true;
1194 } else {
1195 update_ctime = false;
94c1e62d 1196 }
afa2db2f 1197 if (newsize <= oldsize) {
94c1e62d 1198 loff_t holebegin = round_up(newsize, PAGE_SIZE);
d0424c42
HD
1199 if (oldsize > holebegin)
1200 unmap_mapping_range(inode->i_mapping,
1201 holebegin, 0, 1);
1202 if (info->alloced)
1203 shmem_truncate_range(inode,
1204 newsize, (loff_t)-1);
94c1e62d 1205 /* unmap again to remove racily COWed private pages */
d0424c42
HD
1206 if (oldsize > holebegin)
1207 unmap_mapping_range(inode->i_mapping,
1208 holebegin, 0, 1);
94c1e62d 1209 }
1da177e4
LT
1210 }
1211
e09764cf
CM
1212 if (is_quota_modification(idmap, inode, attr)) {
1213 error = dquot_initialize(inode);
1214 if (error)
1215 return error;
1216 }
1217
1218 /* Transfer quota accounting */
1219 if (i_uid_needs_update(idmap, attr, inode) ||
1220 i_gid_needs_update(idmap, attr, inode)) {
1221 error = dquot_transfer(idmap, inode, attr);
e09764cf
CM
1222 if (error)
1223 return error;
1224 }
1225
7a80e5b8 1226 setattr_copy(idmap, inode, attr);
db78b877 1227 if (attr->ia_valid & ATTR_MODE)
7a80e5b8 1228 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
36f05cab 1229 if (!error && update_ctime) {
65287334 1230 inode_set_ctime_current(inode);
36f05cab 1231 if (update_mtime)
cf2766bb 1232 inode_set_mtime_to_ts(inode, inode_get_ctime(inode));
36f05cab
JL
1233 inode_inc_iversion(inode);
1234 }
1da177e4
LT
1235 return error;
1236}
1237
1f895f75 1238static void shmem_evict_inode(struct inode *inode)
1da177e4 1239{
1da177e4 1240 struct shmem_inode_info *info = SHMEM_I(inode);
779750d2 1241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2daf18a7 1242 size_t freed = 0;
1da177e4 1243
30e6a51d 1244 if (shmem_mapping(inode->i_mapping)) {
1da177e4
LT
1245 shmem_unacct_size(info->flags, inode->i_size);
1246 inode->i_size = 0;
bc786390 1247 mapping_set_exiting(inode->i_mapping);
3889e6e7 1248 shmem_truncate_range(inode, 0, (loff_t)-1);
779750d2
KS
1249 if (!list_empty(&info->shrinklist)) {
1250 spin_lock(&sbinfo->shrinklist_lock);
1251 if (!list_empty(&info->shrinklist)) {
1252 list_del_init(&info->shrinklist);
1253 sbinfo->shrinklist_len--;
1254 }
1255 spin_unlock(&sbinfo->shrinklist_lock);
1256 }
af53d3e9
HD
1257 while (!list_empty(&info->swaplist)) {
1258 /* Wait while shmem_unuse() is scanning this inode... */
1259 wait_var_event(&info->stop_eviction,
1260 !atomic_read(&info->stop_eviction));
cb5f7b9a 1261 mutex_lock(&shmem_swaplist_mutex);
af53d3e9
HD
1262 /* ...but beware of the race if we peeked too early */
1263 if (!atomic_read(&info->stop_eviction))
1264 list_del_init(&info->swaplist);
cb5f7b9a 1265 mutex_unlock(&shmem_swaplist_mutex);
1da177e4 1266 }
3ed47db3 1267 }
b09e0fa4 1268
2daf18a7
HD
1269 simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL);
1270 shmem_free_inode(inode->i_sb, freed);
0f3c42f5 1271 WARN_ON(inode->i_blocks);
dbd5768f 1272 clear_inode(inode);
e09764cf
CM
1273#ifdef CONFIG_TMPFS_QUOTA
1274 dquot_free_inode(inode);
1275 dquot_drop(inode);
1276#endif
1da177e4
LT
1277}
1278
b56a2d8a 1279static int shmem_find_swap_entries(struct address_space *mapping,
da08e9b7
MWO
1280 pgoff_t start, struct folio_batch *fbatch,
1281 pgoff_t *indices, unsigned int type)
478922e2 1282{
b56a2d8a 1283 XA_STATE(xas, &mapping->i_pages, start);
da08e9b7 1284 struct folio *folio;
87039546 1285 swp_entry_t entry;
478922e2
MW
1286
1287 rcu_read_lock();
da08e9b7
MWO
1288 xas_for_each(&xas, folio, ULONG_MAX) {
1289 if (xas_retry(&xas, folio))
5b9c98f3 1290 continue;
b56a2d8a 1291
da08e9b7 1292 if (!xa_is_value(folio))
478922e2 1293 continue;
b56a2d8a 1294
da08e9b7 1295 entry = radix_to_swp_entry(folio);
6cec2b95
ML
1296 /*
1297 * swapin error entries can be found in the mapping. But they're
1298 * deliberately ignored here as we've done everything we can do.
1299 */
87039546
HD
1300 if (swp_type(entry) != type)
1301 continue;
b56a2d8a 1302
e384200e 1303 indices[folio_batch_count(fbatch)] = xas.xa_index;
da08e9b7
MWO
1304 if (!folio_batch_add(fbatch, folio))
1305 break;
b56a2d8a
VRP
1306
1307 if (need_resched()) {
1308 xas_pause(&xas);
1309 cond_resched_rcu();
1310 }
478922e2 1311 }
478922e2 1312 rcu_read_unlock();
e21a2955 1313
da08e9b7 1314 return xas.xa_index;
478922e2
MW
1315}
1316
46f65ec1 1317/*
b56a2d8a
VRP
1318 * Move the swapped pages for an inode to page cache. Returns the count
1319 * of pages swapped in, or the error in case of failure.
46f65ec1 1320 */
da08e9b7
MWO
1321static int shmem_unuse_swap_entries(struct inode *inode,
1322 struct folio_batch *fbatch, pgoff_t *indices)
1da177e4 1323{
b56a2d8a
VRP
1324 int i = 0;
1325 int ret = 0;
bde05d1c 1326 int error = 0;
b56a2d8a 1327 struct address_space *mapping = inode->i_mapping;
1da177e4 1328
da08e9b7
MWO
1329 for (i = 0; i < folio_batch_count(fbatch); i++) {
1330 struct folio *folio = fbatch->folios[i];
2e0e26c7 1331
da08e9b7 1332 if (!xa_is_value(folio))
b56a2d8a 1333 continue;
054a9f7c
HD
1334 error = shmem_swapin_folio(inode, indices[i], &folio, SGP_CACHE,
1335 mapping_gfp_mask(mapping), NULL, NULL);
b56a2d8a 1336 if (error == 0) {
da08e9b7
MWO
1337 folio_unlock(folio);
1338 folio_put(folio);
b56a2d8a
VRP
1339 ret++;
1340 }
1341 if (error == -ENOMEM)
1342 break;
1343 error = 0;
bde05d1c 1344 }
b56a2d8a
VRP
1345 return error ? error : ret;
1346}
bde05d1c 1347
b56a2d8a
VRP
1348/*
1349 * If swap found in inode, free it and move page from swapcache to filecache.
1350 */
10a9c496 1351static int shmem_unuse_inode(struct inode *inode, unsigned int type)
b56a2d8a
VRP
1352{
1353 struct address_space *mapping = inode->i_mapping;
1354 pgoff_t start = 0;
da08e9b7 1355 struct folio_batch fbatch;
b56a2d8a 1356 pgoff_t indices[PAGEVEC_SIZE];
b56a2d8a
VRP
1357 int ret = 0;
1358
b56a2d8a 1359 do {
da08e9b7
MWO
1360 folio_batch_init(&fbatch);
1361 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1362 if (folio_batch_count(&fbatch) == 0) {
b56a2d8a
VRP
1363 ret = 0;
1364 break;
46f65ec1 1365 }
b56a2d8a 1366
da08e9b7 1367 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
b56a2d8a
VRP
1368 if (ret < 0)
1369 break;
1370
da08e9b7 1371 start = indices[folio_batch_count(&fbatch) - 1];
b56a2d8a
VRP
1372 } while (true);
1373
1374 return ret;
1da177e4
LT
1375}
1376
1377/*
b56a2d8a
VRP
1378 * Read all the shared memory data that resides in the swap
1379 * device 'type' back into memory, so the swap device can be
1380 * unused.
1da177e4 1381 */
10a9c496 1382int shmem_unuse(unsigned int type)
1da177e4 1383{
b56a2d8a 1384 struct shmem_inode_info *info, *next;
bde05d1c
HD
1385 int error = 0;
1386
b56a2d8a
VRP
1387 if (list_empty(&shmem_swaplist))
1388 return 0;
1389
1390 mutex_lock(&shmem_swaplist_mutex);
b56a2d8a
VRP
1391 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1392 if (!info->swapped) {
6922c0c7 1393 list_del_init(&info->swaplist);
b56a2d8a
VRP
1394 continue;
1395 }
af53d3e9
HD
1396 /*
1397 * Drop the swaplist mutex while searching the inode for swap;
1398 * but before doing so, make sure shmem_evict_inode() will not
1399 * remove placeholder inode from swaplist, nor let it be freed
1400 * (igrab() would protect from unlink, but not from unmount).
1401 */
1402 atomic_inc(&info->stop_eviction);
b56a2d8a 1403 mutex_unlock(&shmem_swaplist_mutex);
b56a2d8a 1404
10a9c496 1405 error = shmem_unuse_inode(&info->vfs_inode, type);
cb5f7b9a 1406 cond_resched();
b56a2d8a
VRP
1407
1408 mutex_lock(&shmem_swaplist_mutex);
1409 next = list_next_entry(info, swaplist);
1410 if (!info->swapped)
1411 list_del_init(&info->swaplist);
af53d3e9
HD
1412 if (atomic_dec_and_test(&info->stop_eviction))
1413 wake_up_var(&info->stop_eviction);
b56a2d8a 1414 if (error)
778dd893 1415 break;
1da177e4 1416 }
cb5f7b9a 1417 mutex_unlock(&shmem_swaplist_mutex);
778dd893 1418
778dd893 1419 return error;
1da177e4
LT
1420}
1421
1422/*
1423 * Move the page from the page cache to the swap cache.
1424 */
1425static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1426{
e2e3fdc7 1427 struct folio *folio = page_folio(page);
8ccee8c1
LC
1428 struct address_space *mapping = folio->mapping;
1429 struct inode *inode = mapping->host;
1430 struct shmem_inode_info *info = SHMEM_I(inode);
2c6efe9c 1431 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
6922c0c7
HD
1432 swp_entry_t swap;
1433 pgoff_t index;
1da177e4 1434
cf7992bf
LC
1435 /*
1436 * Our capabilities prevent regular writeback or sync from ever calling
1437 * shmem_writepage; but a stacking filesystem might use ->writepage of
1438 * its underlying filesystem, in which case tmpfs should write out to
1439 * swap only in response to memory pressure, and not for the writeback
1440 * threads or sync.
1441 */
1442 if (WARN_ON_ONCE(!wbc->for_reclaim))
1443 goto redirty;
1444
2c6efe9c 1445 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
9a976f0c
LC
1446 goto redirty;
1447
1448 if (!total_swap_pages)
1449 goto redirty;
1450
1e6decf3
HD
1451 /*
1452 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1453 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1454 * and its shmem_writeback() needs them to be split when swapping.
1455 */
f530ed0e 1456 if (folio_test_large(folio)) {
1e6decf3 1457 /* Ensure the subpages are still dirty */
f530ed0e 1458 folio_test_set_dirty(folio);
1e6decf3
HD
1459 if (split_huge_page(page) < 0)
1460 goto redirty;
f530ed0e
MWO
1461 folio = page_folio(page);
1462 folio_clear_dirty(folio);
1e6decf3
HD
1463 }
1464
f530ed0e 1465 index = folio->index;
1635f6a7
HD
1466
1467 /*
1468 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1469 * value into swapfile.c, the only way we can correctly account for a
f530ed0e 1470 * fallocated folio arriving here is now to initialize it and write it.
1aac1400 1471 *
f530ed0e 1472 * That's okay for a folio already fallocated earlier, but if we have
1aac1400 1473 * not yet completed the fallocation, then (a) we want to keep track
f530ed0e 1474 * of this folio in case we have to undo it, and (b) it may not be a
1aac1400 1475 * good idea to continue anyway, once we're pushing into swap. So
f530ed0e 1476 * reactivate the folio, and let shmem_fallocate() quit when too many.
1635f6a7 1477 */
f530ed0e 1478 if (!folio_test_uptodate(folio)) {
1aac1400
HD
1479 if (inode->i_private) {
1480 struct shmem_falloc *shmem_falloc;
1481 spin_lock(&inode->i_lock);
1482 shmem_falloc = inode->i_private;
1483 if (shmem_falloc &&
8e205f77 1484 !shmem_falloc->waitq &&
1aac1400
HD
1485 index >= shmem_falloc->start &&
1486 index < shmem_falloc->next)
1487 shmem_falloc->nr_unswapped++;
1488 else
1489 shmem_falloc = NULL;
1490 spin_unlock(&inode->i_lock);
1491 if (shmem_falloc)
1492 goto redirty;
1493 }
f530ed0e
MWO
1494 folio_zero_range(folio, 0, folio_size(folio));
1495 flush_dcache_folio(folio);
1496 folio_mark_uptodate(folio);
1635f6a7
HD
1497 }
1498
e2e3fdc7 1499 swap = folio_alloc_swap(folio);
48f170fb
HD
1500 if (!swap.val)
1501 goto redirty;
d9fe526a 1502
b1dea800
HD
1503 /*
1504 * Add inode to shmem_unuse()'s list of swapped-out inodes,
f530ed0e 1505 * if it's not already there. Do it now before the folio is
6922c0c7 1506 * moved to swap cache, when its pagelock no longer protects
b1dea800 1507 * the inode from eviction. But don't unlock the mutex until
6922c0c7
HD
1508 * we've incremented swapped, because shmem_unuse_inode() will
1509 * prune a !swapped inode from the swaplist under this mutex.
b1dea800 1510 */
48f170fb
HD
1511 mutex_lock(&shmem_swaplist_mutex);
1512 if (list_empty(&info->swaplist))
b56a2d8a 1513 list_add(&info->swaplist, &shmem_swaplist);
b1dea800 1514
a4c366f0 1515 if (add_to_swap_cache(folio, swap,
3852f676
JK
1516 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1517 NULL) == 0) {
3c1b7528 1518 shmem_recalc_inode(inode, 0, 1);
267a4c76 1519 swap_shmem_alloc(swap);
4cd400fd 1520 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
267a4c76 1521
6922c0c7 1522 mutex_unlock(&shmem_swaplist_mutex);
f530ed0e 1523 BUG_ON(folio_mapped(folio));
501a06fe 1524 return swap_writepage(&folio->page, wbc);
1da177e4
LT
1525 }
1526
6922c0c7 1527 mutex_unlock(&shmem_swaplist_mutex);
4081f744 1528 put_swap_folio(folio, swap);
1da177e4 1529redirty:
f530ed0e 1530 folio_mark_dirty(folio);
d9fe526a 1531 if (wbc->for_reclaim)
f530ed0e
MWO
1532 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1533 folio_unlock(folio);
d9fe526a 1534 return 0;
1da177e4
LT
1535}
1536
75edd345 1537#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
71fe804b 1538static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
680d794b 1539{
095f1fc4 1540 char buffer[64];
680d794b 1541
71fe804b 1542 if (!mpol || mpol->mode == MPOL_DEFAULT)
095f1fc4 1543 return; /* show nothing */
680d794b 1544
a7a88b23 1545 mpol_to_str(buffer, sizeof(buffer), mpol);
095f1fc4
LS
1546
1547 seq_printf(seq, ",mpol=%s", buffer);
680d794b 1548}
71fe804b
LS
1549
1550static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1551{
1552 struct mempolicy *mpol = NULL;
1553 if (sbinfo->mpol) {
bf11b9a8 1554 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
71fe804b
LS
1555 mpol = sbinfo->mpol;
1556 mpol_get(mpol);
bf11b9a8 1557 raw_spin_unlock(&sbinfo->stat_lock);
71fe804b
LS
1558 }
1559 return mpol;
1560}
75edd345
HD
1561#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1562static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1563{
1564}
1565static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1566{
1567 return NULL;
1568}
1569#endif /* CONFIG_NUMA && CONFIG_TMPFS */
680d794b 1570
ddc1a5cb
HD
1571static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
1572 pgoff_t index, unsigned int order, pgoff_t *ilx);
800d8c63 1573
ddc1a5cb 1574static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp,
41ffe5d5 1575 struct shmem_inode_info *info, pgoff_t index)
1da177e4 1576{
ddc1a5cb
HD
1577 struct mempolicy *mpol;
1578 pgoff_t ilx;
a4575c41 1579 struct folio *folio;
52cd3b07 1580
ddc1a5cb 1581 mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
a4575c41 1582 folio = swap_cluster_readahead(swap, gfp, mpol, ilx);
ddc1a5cb 1583 mpol_cond_put(mpol);
18a2f371 1584
a4575c41 1585 return folio;
800d8c63
KS
1586}
1587
78cc8cdc
RR
1588/*
1589 * Make sure huge_gfp is always more limited than limit_gfp.
1590 * Some of the flags set permissions, while others set limitations.
1591 */
1592static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1593{
1594 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1595 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
187df5dd
RR
1596 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1597 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1598
1599 /* Allow allocations only from the originally specified zones. */
1600 result |= zoneflags;
78cc8cdc
RR
1601
1602 /*
1603 * Minimize the result gfp by taking the union with the deny flags,
1604 * and the intersection of the allow flags.
1605 */
1606 result |= (limit_gfp & denyflags);
1607 result |= (huge_gfp & limit_gfp) & allowflags;
1608
1609 return result;
1610}
1611
72827e5c 1612static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
800d8c63
KS
1613 struct shmem_inode_info *info, pgoff_t index)
1614{
ddc1a5cb
HD
1615 struct mempolicy *mpol;
1616 pgoff_t ilx;
1617 struct page *page;
800d8c63 1618
ddc1a5cb
HD
1619 mpol = shmem_get_pgoff_policy(info, index, HPAGE_PMD_ORDER, &ilx);
1620 page = alloc_pages_mpol(gfp, HPAGE_PMD_ORDER, mpol, ilx, numa_node_id());
1621 mpol_cond_put(mpol);
18a2f371 1622
ddc1a5cb 1623 return page_rmappable_folio(page);
1da177e4
LT
1624}
1625
0c023ef5 1626static struct folio *shmem_alloc_folio(gfp_t gfp,
3022fd7a 1627 struct shmem_inode_info *info, pgoff_t index)
1da177e4 1628{
ddc1a5cb
HD
1629 struct mempolicy *mpol;
1630 pgoff_t ilx;
1631 struct page *page;
1da177e4 1632
ddc1a5cb
HD
1633 mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
1634 page = alloc_pages_mpol(gfp, 0, mpol, ilx, numa_node_id());
1635 mpol_cond_put(mpol);
800d8c63 1636
ddc1a5cb 1637 return (struct folio *)page;
0c023ef5
MWO
1638}
1639
3022fd7a
HD
1640static struct folio *shmem_alloc_and_add_folio(gfp_t gfp,
1641 struct inode *inode, pgoff_t index,
1642 struct mm_struct *fault_mm, bool huge)
800d8c63 1643{
3022fd7a 1644 struct address_space *mapping = inode->i_mapping;
0f079694 1645 struct shmem_inode_info *info = SHMEM_I(inode);
72827e5c 1646 struct folio *folio;
3022fd7a
HD
1647 long pages;
1648 int error;
52cd3b07 1649
396bcc52 1650 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
800d8c63 1651 huge = false;
800d8c63 1652
3022fd7a
HD
1653 if (huge) {
1654 pages = HPAGE_PMD_NR;
1655 index = round_down(index, HPAGE_PMD_NR);
1656
1657 /*
1658 * Check for conflict before waiting on a huge allocation.
1659 * Conflict might be that a huge page has just been allocated
1660 * and added to page cache by a racing thread, or that there
1661 * is already at least one small page in the huge extent.
1662 * Be careful to retry when appropriate, but not forever!
1663 * Elsewhere -EEXIST would be the right code, but not here.
1664 */
1665 if (xa_find(&mapping->i_pages, &index,
1666 index + HPAGE_PMD_NR - 1, XA_PRESENT))
1667 return ERR_PTR(-E2BIG);
800d8c63 1668
72827e5c 1669 folio = shmem_alloc_hugefolio(gfp, info, index);
3022fd7a
HD
1670 if (!folio)
1671 count_vm_event(THP_FILE_FALLBACK);
1672 } else {
1673 pages = 1;
72827e5c 1674 folio = shmem_alloc_folio(gfp, info, index);
75edd345 1675 }
3022fd7a
HD
1676 if (!folio)
1677 return ERR_PTR(-ENOMEM);
18a2f371 1678
3022fd7a
HD
1679 __folio_set_locked(folio);
1680 __folio_set_swapbacked(folio);
1681
1682 gfp &= GFP_RECLAIM_MASK;
1683 error = mem_cgroup_charge(folio, fault_mm, gfp);
1684 if (error) {
1685 if (xa_find(&mapping->i_pages, &index,
1686 index + pages - 1, XA_PRESENT)) {
1687 error = -EEXIST;
1688 } else if (huge) {
1689 count_vm_event(THP_FILE_FALLBACK);
1690 count_vm_event(THP_FILE_FALLBACK_CHARGE);
1691 }
1692 goto unlock;
1693 }
1694
1695 error = shmem_add_to_page_cache(folio, mapping, index, NULL, gfp);
1696 if (error)
1697 goto unlock;
1698
1699 error = shmem_inode_acct_blocks(inode, pages);
1700 if (error) {
1701 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1702 long freed;
1703 /*
1704 * Try to reclaim some space by splitting a few
1705 * large folios beyond i_size on the filesystem.
1706 */
1707 shmem_unused_huge_shrink(sbinfo, NULL, 2);
1708 /*
1709 * And do a shmem_recalc_inode() to account for freed pages:
1710 * except our folio is there in cache, so not quite balanced.
1711 */
1712 spin_lock(&info->lock);
1713 freed = pages + info->alloced - info->swapped -
1714 READ_ONCE(mapping->nrpages);
1715 if (freed > 0)
1716 info->alloced -= freed;
1717 spin_unlock(&info->lock);
1718 if (freed > 0)
1719 shmem_inode_unacct_blocks(inode, freed);
1720 error = shmem_inode_acct_blocks(inode, pages);
1721 if (error) {
1722 filemap_remove_folio(folio);
1723 goto unlock;
1724 }
1725 }
1726
1727 shmem_recalc_inode(inode, pages, 0);
1728 folio_add_lru(folio);
1729 return folio;
1730
1731unlock:
1732 folio_unlock(folio);
1733 folio_put(folio);
1734 return ERR_PTR(error);
1da177e4 1735}
71fe804b 1736
bde05d1c
HD
1737/*
1738 * When a page is moved from swapcache to shmem filecache (either by the
fc26babb 1739 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
bde05d1c
HD
1740 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1741 * ignorance of the mapping it belongs to. If that mapping has special
1742 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1743 * we may need to copy to a suitable page before moving to filecache.
1744 *
1745 * In a future release, this may well be extended to respect cpuset and
1746 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1747 * but for now it is a simple matter of zone.
1748 */
069d849c 1749static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
bde05d1c 1750{
069d849c 1751 return folio_zonenum(folio) > gfp_zone(gfp);
bde05d1c
HD
1752}
1753
0d698e25 1754static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
bde05d1c
HD
1755 struct shmem_inode_info *info, pgoff_t index)
1756{
d21bba2b 1757 struct folio *old, *new;
bde05d1c 1758 struct address_space *swap_mapping;
c1cb20d4 1759 swp_entry_t entry;
bde05d1c
HD
1760 pgoff_t swap_index;
1761 int error;
1762
0d698e25 1763 old = *foliop;
3d2c9087 1764 entry = old->swap;
c1cb20d4 1765 swap_index = swp_offset(entry);
907ea17e 1766 swap_mapping = swap_address_space(entry);
bde05d1c
HD
1767
1768 /*
1769 * We have arrived here because our zones are constrained, so don't
1770 * limit chance of success by further cpuset and node constraints.
1771 */
1772 gfp &= ~GFP_CONSTRAINT_MASK;
907ea17e
MWO
1773 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1774 new = shmem_alloc_folio(gfp, info, index);
1775 if (!new)
bde05d1c 1776 return -ENOMEM;
bde05d1c 1777
907ea17e
MWO
1778 folio_get(new);
1779 folio_copy(new, old);
1780 flush_dcache_folio(new);
bde05d1c 1781
907ea17e
MWO
1782 __folio_set_locked(new);
1783 __folio_set_swapbacked(new);
1784 folio_mark_uptodate(new);
3d2c9087 1785 new->swap = entry;
907ea17e 1786 folio_set_swapcache(new);
bde05d1c
HD
1787
1788 /*
1789 * Our caller will very soon move newpage out of swapcache, but it's
1790 * a nice clean interface for us to replace oldpage by newpage there.
1791 */
b93b0163 1792 xa_lock_irq(&swap_mapping->i_pages);
907ea17e 1793 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
0142ef6c 1794 if (!error) {
d21bba2b 1795 mem_cgroup_migrate(old, new);
907ea17e
MWO
1796 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1797 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1798 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1799 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
0142ef6c 1800 }
b93b0163 1801 xa_unlock_irq(&swap_mapping->i_pages);
bde05d1c 1802
0142ef6c
HD
1803 if (unlikely(error)) {
1804 /*
1805 * Is this possible? I think not, now that our callers check
1806 * both PageSwapCache and page_private after getting page lock;
1807 * but be defensive. Reverse old to newpage for clear and free.
1808 */
907ea17e 1809 old = new;
0142ef6c 1810 } else {
907ea17e 1811 folio_add_lru(new);
0d698e25 1812 *foliop = new;
0142ef6c 1813 }
bde05d1c 1814
907ea17e
MWO
1815 folio_clear_swapcache(old);
1816 old->private = NULL;
bde05d1c 1817
907ea17e
MWO
1818 folio_unlock(old);
1819 folio_put_refs(old, 2);
0142ef6c 1820 return error;
bde05d1c
HD
1821}
1822
6cec2b95
ML
1823static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1824 struct folio *folio, swp_entry_t swap)
1825{
1826 struct address_space *mapping = inode->i_mapping;
6cec2b95
ML
1827 swp_entry_t swapin_error;
1828 void *old;
1829
af19487f 1830 swapin_error = make_poisoned_swp_entry();
6cec2b95
ML
1831 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1832 swp_to_radix_entry(swap),
1833 swp_to_radix_entry(swapin_error), 0);
1834 if (old != swp_to_radix_entry(swap))
1835 return;
1836
1837 folio_wait_writeback(folio);
75fa68a5 1838 delete_from_swap_cache(folio);
6cec2b95 1839 /*
3c1b7528
HD
1840 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
1841 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
1842 * in shmem_evict_inode().
6cec2b95 1843 */
3c1b7528 1844 shmem_recalc_inode(inode, -1, -1);
6cec2b95
ML
1845 swap_free(swap);
1846}
1847
c5bf121e 1848/*
833de10f
ML
1849 * Swap in the folio pointed to by *foliop.
1850 * Caller has to make sure that *foliop contains a valid swapped folio.
1851 * Returns 0 and the folio in foliop if success. On failure, returns the
1852 * error code and NULL in *foliop.
c5bf121e 1853 */
da08e9b7
MWO
1854static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1855 struct folio **foliop, enum sgp_type sgp,
054a9f7c 1856 gfp_t gfp, struct mm_struct *fault_mm,
c5bf121e
VRP
1857 vm_fault_t *fault_type)
1858{
1859 struct address_space *mapping = inode->i_mapping;
1860 struct shmem_inode_info *info = SHMEM_I(inode);
cbc2bd98 1861 struct swap_info_struct *si;
da08e9b7 1862 struct folio *folio = NULL;
c5bf121e
VRP
1863 swp_entry_t swap;
1864 int error;
1865
da08e9b7
MWO
1866 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1867 swap = radix_to_swp_entry(*foliop);
1868 *foliop = NULL;
c5bf121e 1869
af19487f 1870 if (is_poisoned_swp_entry(swap))
6cec2b95
ML
1871 return -EIO;
1872
cbc2bd98
KS
1873 si = get_swap_device(swap);
1874 if (!si) {
1875 if (!shmem_confirm_swap(mapping, index, swap))
1876 return -EEXIST;
1877 else
1878 return -EINVAL;
1879 }
1880
c5bf121e 1881 /* Look it up and read it in.. */
5739a81c
MWO
1882 folio = swap_cache_get_folio(swap, NULL, 0);
1883 if (!folio) {
c5bf121e
VRP
1884 /* Or update major stats only when swapin succeeds?? */
1885 if (fault_type) {
1886 *fault_type |= VM_FAULT_MAJOR;
1887 count_vm_event(PGMAJFAULT);
054a9f7c 1888 count_memcg_event_mm(fault_mm, PGMAJFAULT);
c5bf121e
VRP
1889 }
1890 /* Here we actually start the io */
ddc1a5cb 1891 folio = shmem_swapin_cluster(swap, gfp, info, index);
5739a81c 1892 if (!folio) {
c5bf121e
VRP
1893 error = -ENOMEM;
1894 goto failed;
1895 }
1896 }
1897
833de10f 1898 /* We have to do this with folio locked to prevent races */
da08e9b7
MWO
1899 folio_lock(folio);
1900 if (!folio_test_swapcache(folio) ||
3d2c9087 1901 folio->swap.val != swap.val ||
c5bf121e
VRP
1902 !shmem_confirm_swap(mapping, index, swap)) {
1903 error = -EEXIST;
1904 goto unlock;
1905 }
da08e9b7 1906 if (!folio_test_uptodate(folio)) {
c5bf121e
VRP
1907 error = -EIO;
1908 goto failed;
1909 }
da08e9b7 1910 folio_wait_writeback(folio);
c5bf121e 1911
8a84802e
SP
1912 /*
1913 * Some architectures may have to restore extra metadata to the
da08e9b7 1914 * folio after reading from swap.
8a84802e 1915 */
da08e9b7 1916 arch_swap_restore(swap, folio);
8a84802e 1917
069d849c 1918 if (shmem_should_replace_folio(folio, gfp)) {
0d698e25 1919 error = shmem_replace_folio(&folio, gfp, info, index);
c5bf121e
VRP
1920 if (error)
1921 goto failed;
1922 }
1923
b7dd44a1 1924 error = shmem_add_to_page_cache(folio, mapping, index,
054a9f7c 1925 swp_to_radix_entry(swap), gfp);
3fea5a49 1926 if (error)
14235ab3 1927 goto failed;
c5bf121e 1928
3c1b7528 1929 shmem_recalc_inode(inode, 0, -1);
c5bf121e
VRP
1930
1931 if (sgp == SGP_WRITE)
da08e9b7 1932 folio_mark_accessed(folio);
c5bf121e 1933
75fa68a5 1934 delete_from_swap_cache(folio);
da08e9b7 1935 folio_mark_dirty(folio);
c5bf121e 1936 swap_free(swap);
cbc2bd98 1937 put_swap_device(si);
c5bf121e 1938
da08e9b7 1939 *foliop = folio;
c5bf121e
VRP
1940 return 0;
1941failed:
1942 if (!shmem_confirm_swap(mapping, index, swap))
1943 error = -EEXIST;
6cec2b95
ML
1944 if (error == -EIO)
1945 shmem_set_folio_swapin_error(inode, index, folio, swap);
c5bf121e 1946unlock:
da08e9b7
MWO
1947 if (folio) {
1948 folio_unlock(folio);
1949 folio_put(folio);
c5bf121e 1950 }
cbc2bd98 1951 put_swap_device(si);
c5bf121e
VRP
1952
1953 return error;
1954}
1955
1da177e4 1956/*
fc26babb 1957 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1da177e4
LT
1958 *
1959 * If we allocate a new one we do not mark it dirty. That's up to the
1960 * vm. If we swap it in we mark it dirty since we also free the swap
9e18eb29
ALC
1961 * entry since a page cannot live in both the swap and page cache.
1962 *
e3e1a506 1963 * vmf and fault_type are only supplied by shmem_fault: otherwise they are NULL.
1da177e4 1964 */
fc26babb
MWO
1965static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1966 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
e3e1a506 1967 struct vm_fault *vmf, vm_fault_t *fault_type)
1da177e4 1968{
e3e1a506 1969 struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
054a9f7c 1970 struct mm_struct *fault_mm;
b7dd44a1 1971 struct folio *folio;
1da177e4 1972 int error;
3022fd7a 1973 bool alloced;
1da177e4 1974
09cbfeaf 1975 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1da177e4 1976 return -EFBIG;
1da177e4 1977repeat:
c5bf121e 1978 if (sgp <= SGP_CACHE &&
3022fd7a 1979 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode))
c5bf121e 1980 return -EINVAL;
c5bf121e 1981
3022fd7a 1982 alloced = false;
054a9f7c 1983 fault_mm = vma ? vma->vm_mm : NULL;
c5bf121e 1984
3022fd7a 1985 folio = filemap_get_entry(inode->i_mapping, index);
b1d0ec3a 1986 if (folio && vma && userfaultfd_minor(vma)) {
aaeb94eb 1987 if (!xa_is_value(folio))
b1d0ec3a 1988 folio_put(folio);
c949b097
AR
1989 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1990 return 0;
1991 }
1992
b1d0ec3a 1993 if (xa_is_value(folio)) {
da08e9b7 1994 error = shmem_swapin_folio(inode, index, &folio,
054a9f7c 1995 sgp, gfp, fault_mm, fault_type);
c5bf121e
VRP
1996 if (error == -EEXIST)
1997 goto repeat;
54af6042 1998
fc26babb 1999 *foliop = folio;
c5bf121e 2000 return error;
54af6042
HD
2001 }
2002
b1d0ec3a 2003 if (folio) {
aaeb94eb
CH
2004 folio_lock(folio);
2005
2006 /* Has the folio been truncated or swapped out? */
3022fd7a 2007 if (unlikely(folio->mapping != inode->i_mapping)) {
aaeb94eb
CH
2008 folio_unlock(folio);
2009 folio_put(folio);
2010 goto repeat;
2011 }
acdd9f8e 2012 if (sgp == SGP_WRITE)
b1d0ec3a
MWO
2013 folio_mark_accessed(folio);
2014 if (folio_test_uptodate(folio))
acdd9f8e 2015 goto out;
fc26babb 2016 /* fallocated folio */
1635f6a7
HD
2017 if (sgp != SGP_READ)
2018 goto clear;
b1d0ec3a
MWO
2019 folio_unlock(folio);
2020 folio_put(folio);
1635f6a7 2021 }
27ab7006
HD
2022
2023 /*
fc26babb
MWO
2024 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
2025 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
acdd9f8e 2026 */
fc26babb 2027 *foliop = NULL;
acdd9f8e
HD
2028 if (sgp == SGP_READ)
2029 return 0;
2030 if (sgp == SGP_NOALLOC)
2031 return -ENOENT;
2032
2033 /*
2034 * Fast cache lookup and swap lookup did not find it: allocate.
27ab7006 2035 */
54af6042 2036
c5bf121e
VRP
2037 if (vma && userfaultfd_missing(vma)) {
2038 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2039 return 0;
2040 }
cfda0526 2041
3022fd7a
HD
2042 if (shmem_is_huge(inode, index, false, fault_mm,
2043 vma ? vma->vm_flags : 0)) {
2044 gfp_t huge_gfp;
1da177e4 2045
3022fd7a
HD
2046 huge_gfp = vma_thp_gfp_mask(vma);
2047 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2048 folio = shmem_alloc_and_add_folio(huge_gfp,
2049 inode, index, fault_mm, true);
2050 if (!IS_ERR(folio)) {
2051 count_vm_event(THP_FILE_ALLOC);
2052 goto alloced;
2053 }
2054 if (PTR_ERR(folio) == -EEXIST)
2055 goto repeat;
c5bf121e 2056 }
800d8c63 2057
3022fd7a
HD
2058 folio = shmem_alloc_and_add_folio(gfp, inode, index, fault_mm, false);
2059 if (IS_ERR(folio)) {
b1d0ec3a 2060 error = PTR_ERR(folio);
3022fd7a
HD
2061 if (error == -EEXIST)
2062 goto repeat;
b1d0ec3a 2063 folio = NULL;
c5bf121e
VRP
2064 goto unlock;
2065 }
54af6042 2066
3022fd7a 2067alloced:
c5bf121e 2068 alloced = true;
b1d0ec3a 2069 if (folio_test_pmd_mappable(folio) &&
c5bf121e 2070 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
fc26babb 2071 folio_next_index(folio) - 1) {
3022fd7a
HD
2072 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2073 struct shmem_inode_info *info = SHMEM_I(inode);
ec9516fb 2074 /*
fc26babb 2075 * Part of the large folio is beyond i_size: subject
c5bf121e 2076 * to shrink under memory pressure.
1635f6a7 2077 */
c5bf121e 2078 spin_lock(&sbinfo->shrinklist_lock);
1635f6a7 2079 /*
c5bf121e
VRP
2080 * _careful to defend against unlocked access to
2081 * ->shrink_list in shmem_unused_huge_shrink()
ec9516fb 2082 */
c5bf121e
VRP
2083 if (list_empty_careful(&info->shrinklist)) {
2084 list_add_tail(&info->shrinklist,
2085 &sbinfo->shrinklist);
2086 sbinfo->shrinklist_len++;
2087 }
2088 spin_unlock(&sbinfo->shrinklist_lock);
2089 }
800d8c63 2090
3022fd7a
HD
2091 if (sgp == SGP_WRITE)
2092 folio_set_referenced(folio);
c5bf121e 2093 /*
fc26babb 2094 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
c5bf121e
VRP
2095 */
2096 if (sgp == SGP_FALLOC)
2097 sgp = SGP_WRITE;
2098clear:
2099 /*
fc26babb
MWO
2100 * Let SGP_WRITE caller clear ends if write does not fill folio;
2101 * but SGP_FALLOC on a folio fallocated earlier must initialize
c5bf121e
VRP
2102 * it now, lest undo on failure cancel our earlier guarantee.
2103 */
b1d0ec3a
MWO
2104 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2105 long i, n = folio_nr_pages(folio);
c5bf121e 2106
b1d0ec3a
MWO
2107 for (i = 0; i < n; i++)
2108 clear_highpage(folio_page(folio, i));
2109 flush_dcache_folio(folio);
2110 folio_mark_uptodate(folio);
1da177e4 2111 }
bde05d1c 2112
54af6042 2113 /* Perhaps the file has been truncated since we checked */
75edd345 2114 if (sgp <= SGP_CACHE &&
09cbfeaf 2115 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
54af6042 2116 error = -EINVAL;
267a4c76 2117 goto unlock;
e83c32e8 2118 }
63ec1973 2119out:
fc26babb 2120 *foliop = folio;
54af6042 2121 return 0;
1da177e4 2122
59a16ead 2123 /*
54af6042 2124 * Error recovery.
59a16ead 2125 */
d1899228 2126unlock:
3022fd7a
HD
2127 if (alloced)
2128 filemap_remove_folio(folio);
2129 shmem_recalc_inode(inode, 0, 0);
b1d0ec3a
MWO
2130 if (folio) {
2131 folio_unlock(folio);
2132 folio_put(folio);
54af6042 2133 }
54af6042 2134 return error;
1da177e4
LT
2135}
2136
4e1fc793
MWO
2137int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2138 enum sgp_type sgp)
2139{
2140 return shmem_get_folio_gfp(inode, index, foliop, sgp,
e3e1a506 2141 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
4e1fc793
MWO
2142}
2143
10d20bd2
LT
2144/*
2145 * This is like autoremove_wake_function, but it removes the wait queue
2146 * entry unconditionally - even if something else had already woken the
2147 * target.
2148 */
f0a9ad1d
HD
2149static int synchronous_wake_function(wait_queue_entry_t *wait,
2150 unsigned int mode, int sync, void *key)
10d20bd2
LT
2151{
2152 int ret = default_wake_function(wait, mode, sync, key);
2055da97 2153 list_del_init(&wait->entry);
10d20bd2
LT
2154 return ret;
2155}
2156
f0a9ad1d
HD
2157/*
2158 * Trinity finds that probing a hole which tmpfs is punching can
2159 * prevent the hole-punch from ever completing: which in turn
2160 * locks writers out with its hold on i_rwsem. So refrain from
2161 * faulting pages into the hole while it's being punched. Although
2162 * shmem_undo_range() does remove the additions, it may be unable to
2163 * keep up, as each new page needs its own unmap_mapping_range() call,
2164 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2165 *
2166 * It does not matter if we sometimes reach this check just before the
2167 * hole-punch begins, so that one fault then races with the punch:
2168 * we just need to make racing faults a rare case.
2169 *
2170 * The implementation below would be much simpler if we just used a
2171 * standard mutex or completion: but we cannot take i_rwsem in fault,
2172 * and bloating every shmem inode for this unlikely case would be sad.
2173 */
2174static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode)
2175{
2176 struct shmem_falloc *shmem_falloc;
2177 struct file *fpin = NULL;
2178 vm_fault_t ret = 0;
2179
2180 spin_lock(&inode->i_lock);
2181 shmem_falloc = inode->i_private;
2182 if (shmem_falloc &&
2183 shmem_falloc->waitq &&
2184 vmf->pgoff >= shmem_falloc->start &&
2185 vmf->pgoff < shmem_falloc->next) {
2186 wait_queue_head_t *shmem_falloc_waitq;
2187 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2188
2189 ret = VM_FAULT_NOPAGE;
2190 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2191 shmem_falloc_waitq = shmem_falloc->waitq;
2192 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2193 TASK_UNINTERRUPTIBLE);
2194 spin_unlock(&inode->i_lock);
2195 schedule();
2196
2197 /*
2198 * shmem_falloc_waitq points into the shmem_fallocate()
2199 * stack of the hole-punching task: shmem_falloc_waitq
2200 * is usually invalid by the time we reach here, but
2201 * finish_wait() does not dereference it in that case;
2202 * though i_lock needed lest racing with wake_up_all().
2203 */
2204 spin_lock(&inode->i_lock);
2205 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2206 }
2207 spin_unlock(&inode->i_lock);
2208 if (fpin) {
2209 fput(fpin);
2210 ret = VM_FAULT_RETRY;
2211 }
2212 return ret;
2213}
2214
20acce67 2215static vm_fault_t shmem_fault(struct vm_fault *vmf)
1da177e4 2216{
f0a9ad1d 2217 struct inode *inode = file_inode(vmf->vma->vm_file);
9e18eb29 2218 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
68a54100 2219 struct folio *folio = NULL;
f0a9ad1d 2220 vm_fault_t ret = 0;
20acce67 2221 int err;
1da177e4 2222
f00cdc6d
HD
2223 /*
2224 * Trinity finds that probing a hole which tmpfs is punching can
f0a9ad1d 2225 * prevent the hole-punch from ever completing: noted in i_private.
f00cdc6d
HD
2226 */
2227 if (unlikely(inode->i_private)) {
f0a9ad1d
HD
2228 ret = shmem_falloc_wait(vmf, inode);
2229 if (ret)
8e205f77 2230 return ret;
f00cdc6d
HD
2231 }
2232
f0a9ad1d 2233 WARN_ON_ONCE(vmf->page != NULL);
68a54100 2234 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
e3e1a506 2235 gfp, vmf, &ret);
20acce67
SJ
2236 if (err)
2237 return vmf_error(err);
f0a9ad1d 2238 if (folio) {
68a54100 2239 vmf->page = folio_file_page(folio, vmf->pgoff);
f0a9ad1d
HD
2240 ret |= VM_FAULT_LOCKED;
2241 }
68da9f05 2242 return ret;
1da177e4
LT
2243}
2244
c01d5b30
HD
2245unsigned long shmem_get_unmapped_area(struct file *file,
2246 unsigned long uaddr, unsigned long len,
2247 unsigned long pgoff, unsigned long flags)
2248{
2249 unsigned long (*get_area)(struct file *,
2250 unsigned long, unsigned long, unsigned long, unsigned long);
2251 unsigned long addr;
2252 unsigned long offset;
2253 unsigned long inflated_len;
2254 unsigned long inflated_addr;
2255 unsigned long inflated_offset;
2256
2257 if (len > TASK_SIZE)
2258 return -ENOMEM;
2259
2260 get_area = current->mm->get_unmapped_area;
2261 addr = get_area(file, uaddr, len, pgoff, flags);
2262
396bcc52 2263 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
c01d5b30
HD
2264 return addr;
2265 if (IS_ERR_VALUE(addr))
2266 return addr;
2267 if (addr & ~PAGE_MASK)
2268 return addr;
2269 if (addr > TASK_SIZE - len)
2270 return addr;
2271
2272 if (shmem_huge == SHMEM_HUGE_DENY)
2273 return addr;
2274 if (len < HPAGE_PMD_SIZE)
2275 return addr;
2276 if (flags & MAP_FIXED)
2277 return addr;
2278 /*
2279 * Our priority is to support MAP_SHARED mapped hugely;
2280 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
99158997
KS
2281 * But if caller specified an address hint and we allocated area there
2282 * successfully, respect that as before.
c01d5b30 2283 */
99158997 2284 if (uaddr == addr)
c01d5b30
HD
2285 return addr;
2286
2287 if (shmem_huge != SHMEM_HUGE_FORCE) {
2288 struct super_block *sb;
2289
2290 if (file) {
2291 VM_BUG_ON(file->f_op != &shmem_file_operations);
2292 sb = file_inode(file)->i_sb;
2293 } else {
2294 /*
2295 * Called directly from mm/mmap.c, or drivers/char/mem.c
2296 * for "/dev/zero", to create a shared anonymous object.
2297 */
2298 if (IS_ERR(shm_mnt))
2299 return addr;
2300 sb = shm_mnt->mnt_sb;
2301 }
3089bf61 2302 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
c01d5b30
HD
2303 return addr;
2304 }
2305
2306 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2307 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2308 return addr;
2309 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2310 return addr;
2311
2312 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2313 if (inflated_len > TASK_SIZE)
2314 return addr;
2315 if (inflated_len < len)
2316 return addr;
2317
99158997 2318 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
c01d5b30
HD
2319 if (IS_ERR_VALUE(inflated_addr))
2320 return addr;
2321 if (inflated_addr & ~PAGE_MASK)
2322 return addr;
2323
2324 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2325 inflated_addr += offset - inflated_offset;
2326 if (inflated_offset > offset)
2327 inflated_addr += HPAGE_PMD_SIZE;
2328
2329 if (inflated_addr > TASK_SIZE - len)
2330 return addr;
2331 return inflated_addr;
2332}
2333
1da177e4 2334#ifdef CONFIG_NUMA
41ffe5d5 2335static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1da177e4 2336{
496ad9aa 2337 struct inode *inode = file_inode(vma->vm_file);
41ffe5d5 2338 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1da177e4
LT
2339}
2340
d8dc74f2 2341static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
ddc1a5cb 2342 unsigned long addr, pgoff_t *ilx)
1da177e4 2343{
496ad9aa 2344 struct inode *inode = file_inode(vma->vm_file);
41ffe5d5 2345 pgoff_t index;
1da177e4 2346
ddc1a5cb
HD
2347 /*
2348 * Bias interleave by inode number to distribute better across nodes;
2349 * but this interface is independent of which page order is used, so
2350 * supplies only that bias, letting caller apply the offset (adjusted
2351 * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()).
2352 */
2353 *ilx = inode->i_ino;
41ffe5d5
HD
2354 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2355 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1da177e4 2356}
ddc1a5cb
HD
2357
2358static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2359 pgoff_t index, unsigned int order, pgoff_t *ilx)
2360{
2361 struct mempolicy *mpol;
2362
2363 /* Bias interleave by inode number to distribute better across nodes */
2364 *ilx = info->vfs_inode.i_ino + (index >> order);
2365
2366 mpol = mpol_shared_policy_lookup(&info->policy, index);
2367 return mpol ? mpol : get_task_policy(current);
2368}
2369#else
2370static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2371 pgoff_t index, unsigned int order, pgoff_t *ilx)
2372{
2373 *ilx = 0;
2374 return NULL;
2375}
2376#endif /* CONFIG_NUMA */
1da177e4 2377
d7c9e99a 2378int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
1da177e4 2379{
496ad9aa 2380 struct inode *inode = file_inode(file);
1da177e4
LT
2381 struct shmem_inode_info *info = SHMEM_I(inode);
2382 int retval = -ENOMEM;
2383
ea0dfeb4
HD
2384 /*
2385 * What serializes the accesses to info->flags?
2386 * ipc_lock_object() when called from shmctl_do_lock(),
2387 * no serialization needed when called from shm_destroy().
2388 */
1da177e4 2389 if (lock && !(info->flags & VM_LOCKED)) {
d7c9e99a 2390 if (!user_shm_lock(inode->i_size, ucounts))
1da177e4
LT
2391 goto out_nomem;
2392 info->flags |= VM_LOCKED;
89e004ea 2393 mapping_set_unevictable(file->f_mapping);
1da177e4 2394 }
d7c9e99a
AG
2395 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2396 user_shm_unlock(inode->i_size, ucounts);
1da177e4 2397 info->flags &= ~VM_LOCKED;
89e004ea 2398 mapping_clear_unevictable(file->f_mapping);
1da177e4
LT
2399 }
2400 retval = 0;
89e004ea 2401
1da177e4 2402out_nomem:
1da177e4
LT
2403 return retval;
2404}
2405
9b83a6a8 2406static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1da177e4 2407{
d09e8ca6
PT
2408 struct inode *inode = file_inode(file);
2409 struct shmem_inode_info *info = SHMEM_I(inode);
22247efd 2410 int ret;
ab3948f5 2411
28464bbb 2412 ret = seal_check_write(info->seals, vma);
22247efd
PX
2413 if (ret)
2414 return ret;
ab3948f5 2415
51b0bff2 2416 /* arm64 - allow memory tagging on RAM-based files */
1c71222e 2417 vm_flags_set(vma, VM_MTE_ALLOWED);
51b0bff2 2418
1da177e4 2419 file_accessed(file);
d09e8ca6
PT
2420 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2421 if (inode->i_nlink)
2422 vma->vm_ops = &shmem_vm_ops;
2423 else
2424 vma->vm_ops = &shmem_anon_vm_ops;
1da177e4
LT
2425 return 0;
2426}
2427
e88e0d36
HD
2428static int shmem_file_open(struct inode *inode, struct file *file)
2429{
2430 file->f_mode |= FMODE_CAN_ODIRECT;
2431 return generic_file_open(inode, file);
2432}
2433
cb241339
HD
2434#ifdef CONFIG_TMPFS_XATTR
2435static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2436
2437/*
2438 * chattr's fsflags are unrelated to extended attributes,
2439 * but tmpfs has chosen to enable them under the same config option.
2440 */
2441static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2442{
2443 unsigned int i_flags = 0;
2444
2445 if (fsflags & FS_NOATIME_FL)
2446 i_flags |= S_NOATIME;
2447 if (fsflags & FS_APPEND_FL)
2448 i_flags |= S_APPEND;
2449 if (fsflags & FS_IMMUTABLE_FL)
2450 i_flags |= S_IMMUTABLE;
2451 /*
2452 * But FS_NODUMP_FL does not require any action in i_flags.
2453 */
2454 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2455}
2456#else
2457static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
e408e695 2458{
e408e695 2459}
cb241339
HD
2460#define shmem_initxattrs NULL
2461#endif
e408e695 2462
a2e45955
CL
2463static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2464{
2465 return &SHMEM_I(inode)->dir_offsets;
2466}
2467
e09764cf
CM
2468static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2469 struct super_block *sb,
2470 struct inode *dir, umode_t mode,
2471 dev_t dev, unsigned long flags)
1da177e4
LT
2472{
2473 struct inode *inode;
2474 struct shmem_inode_info *info;
2475 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
e809d5f0 2476 ino_t ino;
71480663
CM
2477 int err;
2478
2479 err = shmem_reserve_inode(sb, &ino);
2480 if (err)
2481 return ERR_PTR(err);
1da177e4 2482
1da177e4 2483 inode = new_inode(sb);
71480663 2484 if (!inode) {
2daf18a7 2485 shmem_free_inode(sb, 0);
71480663
CM
2486 return ERR_PTR(-ENOSPC);
2487 }
b45d71fb 2488
71480663
CM
2489 inode->i_ino = ino;
2490 inode_init_owner(idmap, inode, dir, mode);
2491 inode->i_blocks = 0;
cf2766bb 2492 simple_inode_init_ts(inode);
71480663
CM
2493 inode->i_generation = get_random_u32();
2494 info = SHMEM_I(inode);
2495 memset(info, 0, (char *)inode - (char *)info);
2496 spin_lock_init(&info->lock);
2497 atomic_set(&info->stop_eviction, 0);
2498 info->seals = F_SEAL_SEAL;
2499 info->flags = flags & VM_NORESERVE;
cf2766bb 2500 info->i_crtime = inode_get_mtime(inode);
71480663
CM
2501 info->fsflags = (dir == NULL) ? 0 :
2502 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2503 if (info->fsflags)
2504 shmem_set_inode_flags(inode, info->fsflags);
2505 INIT_LIST_HEAD(&info->shrinklist);
2506 INIT_LIST_HEAD(&info->swaplist);
71480663
CM
2507 simple_xattrs_init(&info->xattrs);
2508 cache_no_acl(inode);
9be7d5b0
HD
2509 if (sbinfo->noswap)
2510 mapping_set_unevictable(inode->i_mapping);
71480663
CM
2511 mapping_set_large_folios(inode->i_mapping);
2512
2513 switch (mode & S_IFMT) {
2514 default:
2515 inode->i_op = &shmem_special_inode_operations;
2516 init_special_inode(inode, mode, dev);
2517 break;
2518 case S_IFREG:
2519 inode->i_mapping->a_ops = &shmem_aops;
2520 inode->i_op = &shmem_inode_operations;
2521 inode->i_fop = &shmem_file_operations;
2522 mpol_shared_policy_init(&info->policy,
2523 shmem_get_sbmpol(sbinfo));
2524 break;
2525 case S_IFDIR:
2526 inc_nlink(inode);
2527 /* Some things misbehave if size == 0 on a directory */
2528 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2529 inode->i_op = &shmem_dir_inode_operations;
a2e45955
CL
2530 inode->i_fop = &simple_offset_dir_operations;
2531 simple_offset_init(shmem_get_offset_ctx(inode));
71480663
CM
2532 break;
2533 case S_IFLNK:
2534 /*
2535 * Must not load anything in the rbtree,
2536 * mpol_free_shared_policy will not be called.
2537 */
2538 mpol_shared_policy_init(&info->policy, NULL);
2539 break;
2540 }
2541
2542 lockdep_annotate_inode_mutex_key(inode);
1da177e4
LT
2543 return inode;
2544}
2545
e09764cf
CM
2546#ifdef CONFIG_TMPFS_QUOTA
2547static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2548 struct super_block *sb, struct inode *dir,
2549 umode_t mode, dev_t dev, unsigned long flags)
2550{
2551 int err;
2552 struct inode *inode;
2553
2554 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2555 if (IS_ERR(inode))
2556 return inode;
2557
2558 err = dquot_initialize(inode);
2559 if (err)
2560 goto errout;
2561
2562 err = dquot_alloc_inode(inode);
2563 if (err) {
2564 dquot_drop(inode);
2565 goto errout;
2566 }
2567 return inode;
2568
2569errout:
2570 inode->i_flags |= S_NOQUOTA;
2571 iput(inode);
2572 return ERR_PTR(err);
2573}
2574#else
2575static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2576 struct super_block *sb, struct inode *dir,
2577 umode_t mode, dev_t dev, unsigned long flags)
2578{
2579 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2580}
2581#endif /* CONFIG_TMPFS_QUOTA */
2582
3460f6e5 2583#ifdef CONFIG_USERFAULTFD
61c50040 2584int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
3460f6e5
AR
2585 struct vm_area_struct *dst_vma,
2586 unsigned long dst_addr,
2587 unsigned long src_addr,
d9712937 2588 uffd_flags_t flags,
d7be6d7e 2589 struct folio **foliop)
4c27fe4c
MR
2590{
2591 struct inode *inode = file_inode(dst_vma->vm_file);
2592 struct shmem_inode_info *info = SHMEM_I(inode);
4c27fe4c
MR
2593 struct address_space *mapping = inode->i_mapping;
2594 gfp_t gfp = mapping_gfp_mask(mapping);
2595 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
4c27fe4c 2596 void *page_kaddr;
b7dd44a1 2597 struct folio *folio;
4c27fe4c 2598 int ret;
3460f6e5 2599 pgoff_t max_off;
4c27fe4c 2600
4199f51a 2601 if (shmem_inode_acct_blocks(inode, 1)) {
7ed9d238
AR
2602 /*
2603 * We may have got a page, returned -ENOENT triggering a retry,
2604 * and now we find ourselves with -ENOMEM. Release the page, to
2605 * avoid a BUG_ON in our caller.
2606 */
d7be6d7e
Z
2607 if (unlikely(*foliop)) {
2608 folio_put(*foliop);
2609 *foliop = NULL;
7ed9d238 2610 }
7d64ae3a 2611 return -ENOMEM;
7ed9d238 2612 }
4c27fe4c 2613
d7be6d7e 2614 if (!*foliop) {
7d64ae3a 2615 ret = -ENOMEM;
7a7256d5
MWO
2616 folio = shmem_alloc_folio(gfp, info, pgoff);
2617 if (!folio)
0f079694 2618 goto out_unacct_blocks;
4c27fe4c 2619
d9712937 2620 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
7a7256d5 2621 page_kaddr = kmap_local_folio(folio, 0);
5dc21f0c
IW
2622 /*
2623 * The read mmap_lock is held here. Despite the
2624 * mmap_lock being read recursive a deadlock is still
2625 * possible if a writer has taken a lock. For example:
2626 *
2627 * process A thread 1 takes read lock on own mmap_lock
2628 * process A thread 2 calls mmap, blocks taking write lock
2629 * process B thread 1 takes page fault, read lock on own mmap lock
2630 * process B thread 2 calls mmap, blocks taking write lock
2631 * process A thread 1 blocks taking read lock on process B
2632 * process B thread 1 blocks taking read lock on process A
2633 *
2634 * Disable page faults to prevent potential deadlock
2635 * and retry the copy outside the mmap_lock.
2636 */
2637 pagefault_disable();
8d103963
MR
2638 ret = copy_from_user(page_kaddr,
2639 (const void __user *)src_addr,
2640 PAGE_SIZE);
5dc21f0c 2641 pagefault_enable();
7a7256d5 2642 kunmap_local(page_kaddr);
8d103963 2643
c1e8d7c6 2644 /* fallback to copy_from_user outside mmap_lock */
8d103963 2645 if (unlikely(ret)) {
d7be6d7e 2646 *foliop = folio;
7d64ae3a 2647 ret = -ENOENT;
8d103963 2648 /* don't free the page */
7d64ae3a 2649 goto out_unacct_blocks;
8d103963 2650 }
19b482c2 2651
7a7256d5 2652 flush_dcache_folio(folio);
3460f6e5 2653 } else { /* ZEROPAGE */
7a7256d5 2654 clear_user_highpage(&folio->page, dst_addr);
4c27fe4c
MR
2655 }
2656 } else {
d7be6d7e 2657 folio = *foliop;
7a7256d5 2658 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
d7be6d7e 2659 *foliop = NULL;
4c27fe4c
MR
2660 }
2661
7a7256d5
MWO
2662 VM_BUG_ON(folio_test_locked(folio));
2663 VM_BUG_ON(folio_test_swapbacked(folio));
2664 __folio_set_locked(folio);
2665 __folio_set_swapbacked(folio);
2666 __folio_mark_uptodate(folio);
9cc90c66 2667
e2a50c1f 2668 ret = -EFAULT;
e2a50c1f 2669 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3460f6e5 2670 if (unlikely(pgoff >= max_off))
e2a50c1f
AA
2671 goto out_release;
2672
054a9f7c
HD
2673 ret = mem_cgroup_charge(folio, dst_vma->vm_mm, gfp);
2674 if (ret)
2675 goto out_release;
2676 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp);
4c27fe4c 2677 if (ret)
3fea5a49 2678 goto out_release;
4c27fe4c 2679
61c50040 2680 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
d9712937 2681 &folio->page, true, flags);
7d64ae3a
AR
2682 if (ret)
2683 goto out_delete_from_cache;
4c27fe4c 2684
3c1b7528 2685 shmem_recalc_inode(inode, 1, 0);
7a7256d5 2686 folio_unlock(folio);
7d64ae3a
AR
2687 return 0;
2688out_delete_from_cache:
7a7256d5 2689 filemap_remove_folio(folio);
4c27fe4c 2690out_release:
7a7256d5
MWO
2691 folio_unlock(folio);
2692 folio_put(folio);
4c27fe4c 2693out_unacct_blocks:
0f079694 2694 shmem_inode_unacct_blocks(inode, 1);
7d64ae3a 2695 return ret;
8d103963 2696}
3460f6e5 2697#endif /* CONFIG_USERFAULTFD */
8d103963 2698
1da177e4 2699#ifdef CONFIG_TMPFS
92e1d5be 2700static const struct inode_operations shmem_symlink_inode_operations;
69f07ec9 2701static const struct inode_operations shmem_short_symlink_operations;
1da177e4 2702
1da177e4 2703static int
800d15a5 2704shmem_write_begin(struct file *file, struct address_space *mapping,
9d6b0cd7 2705 loff_t pos, unsigned len,
800d15a5 2706 struct page **pagep, void **fsdata)
1da177e4 2707{
800d15a5 2708 struct inode *inode = mapping->host;
40e041a2 2709 struct shmem_inode_info *info = SHMEM_I(inode);
09cbfeaf 2710 pgoff_t index = pos >> PAGE_SHIFT;
eff1f906 2711 struct folio *folio;
a7605426 2712 int ret = 0;
40e041a2 2713
9608703e 2714 /* i_rwsem is held by caller */
ab3948f5
JFG
2715 if (unlikely(info->seals & (F_SEAL_GROW |
2716 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2717 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
40e041a2
DH
2718 return -EPERM;
2719 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2720 return -EPERM;
2721 }
2722
eff1f906 2723 ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
a7605426
YS
2724 if (ret)
2725 return ret;
2726
eff1f906 2727 *pagep = folio_file_page(folio, index);
a7605426 2728 if (PageHWPoison(*pagep)) {
eff1f906
MWO
2729 folio_unlock(folio);
2730 folio_put(folio);
a7605426
YS
2731 *pagep = NULL;
2732 return -EIO;
2733 }
2734
2735 return 0;
800d15a5
NP
2736}
2737
2738static int
2739shmem_write_end(struct file *file, struct address_space *mapping,
2740 loff_t pos, unsigned len, unsigned copied,
2741 struct page *page, void *fsdata)
2742{
69bbb87b 2743 struct folio *folio = page_folio(page);
800d15a5
NP
2744 struct inode *inode = mapping->host;
2745
d3602444
HD
2746 if (pos + copied > inode->i_size)
2747 i_size_write(inode, pos + copied);
2748
69bbb87b
MWO
2749 if (!folio_test_uptodate(folio)) {
2750 if (copied < folio_size(folio)) {
2751 size_t from = offset_in_folio(folio, pos);
2752 folio_zero_segments(folio, 0, from,
2753 from + copied, folio_size(folio));
ec9516fb 2754 }
69bbb87b 2755 folio_mark_uptodate(folio);
ec9516fb 2756 }
69bbb87b
MWO
2757 folio_mark_dirty(folio);
2758 folio_unlock(folio);
2759 folio_put(folio);
800d15a5 2760
800d15a5 2761 return copied;
1da177e4
LT
2762}
2763
2ba5bbed 2764static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1da177e4 2765{
6e58e79d
AV
2766 struct file *file = iocb->ki_filp;
2767 struct inode *inode = file_inode(file);
1da177e4 2768 struct address_space *mapping = inode->i_mapping;
41ffe5d5
HD
2769 pgoff_t index;
2770 unsigned long offset;
f7c1d074 2771 int error = 0;
cb66a7a1 2772 ssize_t retval = 0;
6e58e79d 2773 loff_t *ppos = &iocb->ki_pos;
a0ee5ec5 2774
09cbfeaf
KS
2775 index = *ppos >> PAGE_SHIFT;
2776 offset = *ppos & ~PAGE_MASK;
1da177e4
LT
2777
2778 for (;;) {
4601e2fc 2779 struct folio *folio = NULL;
1da177e4 2780 struct page *page = NULL;
41ffe5d5
HD
2781 pgoff_t end_index;
2782 unsigned long nr, ret;
1da177e4
LT
2783 loff_t i_size = i_size_read(inode);
2784
09cbfeaf 2785 end_index = i_size >> PAGE_SHIFT;
1da177e4
LT
2786 if (index > end_index)
2787 break;
2788 if (index == end_index) {
09cbfeaf 2789 nr = i_size & ~PAGE_MASK;
1da177e4
LT
2790 if (nr <= offset)
2791 break;
2792 }
2793
4601e2fc 2794 error = shmem_get_folio(inode, index, &folio, SGP_READ);
6e58e79d
AV
2795 if (error) {
2796 if (error == -EINVAL)
2797 error = 0;
1da177e4
LT
2798 break;
2799 }
4601e2fc
MWO
2800 if (folio) {
2801 folio_unlock(folio);
a7605426 2802
4601e2fc 2803 page = folio_file_page(folio, index);
a7605426 2804 if (PageHWPoison(page)) {
4601e2fc 2805 folio_put(folio);
a7605426
YS
2806 error = -EIO;
2807 break;
2808 }
75edd345 2809 }
1da177e4
LT
2810
2811 /*
2812 * We must evaluate after, since reads (unlike writes)
9608703e 2813 * are called without i_rwsem protection against truncate
1da177e4 2814 */
09cbfeaf 2815 nr = PAGE_SIZE;
1da177e4 2816 i_size = i_size_read(inode);
09cbfeaf 2817 end_index = i_size >> PAGE_SHIFT;
1da177e4 2818 if (index == end_index) {
09cbfeaf 2819 nr = i_size & ~PAGE_MASK;
1da177e4 2820 if (nr <= offset) {
4601e2fc
MWO
2821 if (folio)
2822 folio_put(folio);
1da177e4
LT
2823 break;
2824 }
2825 }
2826 nr -= offset;
2827
4601e2fc 2828 if (folio) {
1da177e4
LT
2829 /*
2830 * If users can be writing to this page using arbitrary
2831 * virtual addresses, take care about potential aliasing
2832 * before reading the page on the kernel side.
2833 */
2834 if (mapping_writably_mapped(mapping))
2835 flush_dcache_page(page);
2836 /*
2837 * Mark the page accessed if we read the beginning.
2838 */
2839 if (!offset)
4601e2fc 2840 folio_mark_accessed(folio);
1bdec44b
HD
2841 /*
2842 * Ok, we have the page, and it's up-to-date, so
2843 * now we can copy it to user space...
2844 */
2845 ret = copy_page_to_iter(page, offset, nr, to);
4601e2fc 2846 folio_put(folio);
1bdec44b 2847
fcb14cb1 2848 } else if (user_backed_iter(to)) {
1bdec44b
HD
2849 /*
2850 * Copy to user tends to be so well optimized, but
2851 * clear_user() not so much, that it is noticeably
2852 * faster to copy the zero page instead of clearing.
2853 */
2854 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
b5810039 2855 } else {
1bdec44b
HD
2856 /*
2857 * But submitting the same page twice in a row to
2858 * splice() - or others? - can result in confusion:
2859 * so don't attempt that optimization on pipes etc.
2860 */
2861 ret = iov_iter_zero(nr, to);
b5810039 2862 }
1da177e4 2863
6e58e79d 2864 retval += ret;
1da177e4 2865 offset += ret;
09cbfeaf
KS
2866 index += offset >> PAGE_SHIFT;
2867 offset &= ~PAGE_MASK;
1da177e4 2868
2ba5bbed 2869 if (!iov_iter_count(to))
1da177e4 2870 break;
6e58e79d
AV
2871 if (ret < nr) {
2872 error = -EFAULT;
2873 break;
2874 }
1da177e4
LT
2875 cond_resched();
2876 }
2877
09cbfeaf 2878 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
6e58e79d
AV
2879 file_accessed(file);
2880 return retval ? retval : error;
1da177e4
LT
2881}
2882
e88e0d36
HD
2883static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2884{
2885 struct file *file = iocb->ki_filp;
2886 struct inode *inode = file->f_mapping->host;
2887 ssize_t ret;
2888
2889 inode_lock(inode);
2890 ret = generic_write_checks(iocb, from);
2891 if (ret <= 0)
2892 goto unlock;
2893 ret = file_remove_privs(file);
2894 if (ret)
2895 goto unlock;
2896 ret = file_update_time(file);
2897 if (ret)
2898 goto unlock;
2899 ret = generic_perform_write(iocb, from);
2900unlock:
2901 inode_unlock(inode);
2902 return ret;
2903}
2904
bd194b18
DH
2905static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
2906 struct pipe_buffer *buf)
2907{
2908 return true;
2909}
2910
2911static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
2912 struct pipe_buffer *buf)
2913{
2914}
2915
2916static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
2917 struct pipe_buffer *buf)
2918{
2919 return false;
2920}
2921
2922static const struct pipe_buf_operations zero_pipe_buf_ops = {
2923 .release = zero_pipe_buf_release,
2924 .try_steal = zero_pipe_buf_try_steal,
2925 .get = zero_pipe_buf_get,
2926};
2927
2928static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
2929 loff_t fpos, size_t size)
2930{
2931 size_t offset = fpos & ~PAGE_MASK;
2932
2933 size = min_t(size_t, size, PAGE_SIZE - offset);
2934
2935 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2936 struct pipe_buffer *buf = pipe_head_buf(pipe);
2937
2938 *buf = (struct pipe_buffer) {
2939 .ops = &zero_pipe_buf_ops,
2940 .page = ZERO_PAGE(0),
2941 .offset = offset,
2942 .len = size,
2943 };
2944 pipe->head++;
2945 }
2946
2947 return size;
2948}
2949
2950static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2951 struct pipe_inode_info *pipe,
2952 size_t len, unsigned int flags)
2953{
2954 struct inode *inode = file_inode(in);
2955 struct address_space *mapping = inode->i_mapping;
2956 struct folio *folio = NULL;
2957 size_t total_spliced = 0, used, npages, n, part;
2958 loff_t isize;
2959 int error = 0;
2960
2961 /* Work out how much data we can actually add into the pipe */
2962 used = pipe_occupancy(pipe->head, pipe->tail);
2963 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2964 len = min_t(size_t, len, npages * PAGE_SIZE);
2965
2966 do {
2967 if (*ppos >= i_size_read(inode))
2968 break;
2969
fa598952
HD
2970 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
2971 SGP_READ);
bd194b18
DH
2972 if (error) {
2973 if (error == -EINVAL)
2974 error = 0;
2975 break;
2976 }
2977 if (folio) {
2978 folio_unlock(folio);
2979
fa598952
HD
2980 if (folio_test_hwpoison(folio) ||
2981 (folio_test_large(folio) &&
2982 folio_test_has_hwpoisoned(folio))) {
bd194b18
DH
2983 error = -EIO;
2984 break;
2985 }
2986 }
2987
2988 /*
2989 * i_size must be checked after we know the pages are Uptodate.
2990 *
2991 * Checking i_size after the check allows us to calculate
2992 * the correct value for "nr", which means the zero-filled
2993 * part of the page is not copied back to userspace (unless
2994 * another truncate extends the file - this is desired though).
2995 */
2996 isize = i_size_read(inode);
2997 if (unlikely(*ppos >= isize))
2998 break;
2999 part = min_t(loff_t, isize - *ppos, len);
3000
3001 if (folio) {
3002 /*
3003 * If users can be writing to this page using arbitrary
3004 * virtual addresses, take care about potential aliasing
3005 * before reading the page on the kernel side.
3006 */
3007 if (mapping_writably_mapped(mapping))
3008 flush_dcache_folio(folio);
3009 folio_mark_accessed(folio);
3010 /*
3011 * Ok, we have the page, and it's up-to-date, so we can
3012 * now splice it into the pipe.
3013 */
3014 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
3015 folio_put(folio);
3016 folio = NULL;
3017 } else {
fa598952 3018 n = splice_zeropage_into_pipe(pipe, *ppos, part);
bd194b18
DH
3019 }
3020
3021 if (!n)
3022 break;
3023 len -= n;
3024 total_spliced += n;
3025 *ppos += n;
3026 in->f_ra.prev_pos = *ppos;
3027 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
3028 break;
3029
3030 cond_resched();
3031 } while (len);
3032
3033 if (folio)
3034 folio_put(folio);
3035
3036 file_accessed(in);
3037 return total_spliced ? total_spliced : error;
3038}
3039
965c8e59 3040static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
220f2ac9
HD
3041{
3042 struct address_space *mapping = file->f_mapping;
3043 struct inode *inode = mapping->host;
220f2ac9 3044
965c8e59
AM
3045 if (whence != SEEK_DATA && whence != SEEK_HOLE)
3046 return generic_file_llseek_size(file, offset, whence,
220f2ac9 3047 MAX_LFS_FILESIZE, i_size_read(inode));
41139aa4
MWO
3048 if (offset < 0)
3049 return -ENXIO;
3050
5955102c 3051 inode_lock(inode);
9608703e 3052 /* We're holding i_rwsem so we can access i_size directly */
41139aa4 3053 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
387aae6f
HD
3054 if (offset >= 0)
3055 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
5955102c 3056 inode_unlock(inode);
220f2ac9
HD
3057 return offset;
3058}
3059
83e4fa9c
HD
3060static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3061 loff_t len)
3062{
496ad9aa 3063 struct inode *inode = file_inode(file);
e2d12e22 3064 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
40e041a2 3065 struct shmem_inode_info *info = SHMEM_I(inode);
1aac1400 3066 struct shmem_falloc shmem_falloc;
d144bf62 3067 pgoff_t start, index, end, undo_fallocend;
e2d12e22 3068 int error;
83e4fa9c 3069
13ace4d0
HD
3070 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3071 return -EOPNOTSUPP;
3072
5955102c 3073 inode_lock(inode);
83e4fa9c
HD
3074
3075 if (mode & FALLOC_FL_PUNCH_HOLE) {
3076 struct address_space *mapping = file->f_mapping;
3077 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3078 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
8e205f77 3079 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
83e4fa9c 3080
9608703e 3081 /* protected by i_rwsem */
ab3948f5 3082 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
40e041a2
DH
3083 error = -EPERM;
3084 goto out;
3085 }
3086
8e205f77 3087 shmem_falloc.waitq = &shmem_falloc_waitq;
aa71ecd8 3088 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
f00cdc6d
HD
3089 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3090 spin_lock(&inode->i_lock);
3091 inode->i_private = &shmem_falloc;
3092 spin_unlock(&inode->i_lock);
3093
83e4fa9c
HD
3094 if ((u64)unmap_end > (u64)unmap_start)
3095 unmap_mapping_range(mapping, unmap_start,
3096 1 + unmap_end - unmap_start, 0);
3097 shmem_truncate_range(inode, offset, offset + len - 1);
3098 /* No need to unmap again: hole-punching leaves COWed pages */
8e205f77
HD
3099
3100 spin_lock(&inode->i_lock);
3101 inode->i_private = NULL;
3102 wake_up_all(&shmem_falloc_waitq);
2055da97 3103 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
8e205f77 3104 spin_unlock(&inode->i_lock);
83e4fa9c 3105 error = 0;
8e205f77 3106 goto out;
e2d12e22
HD
3107 }
3108
3109 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3110 error = inode_newsize_ok(inode, offset + len);
3111 if (error)
3112 goto out;
3113
40e041a2
DH
3114 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3115 error = -EPERM;
3116 goto out;
3117 }
3118
09cbfeaf
KS
3119 start = offset >> PAGE_SHIFT;
3120 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
e2d12e22
HD
3121 /* Try to avoid a swapstorm if len is impossible to satisfy */
3122 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3123 error = -ENOSPC;
3124 goto out;
83e4fa9c
HD
3125 }
3126
8e205f77 3127 shmem_falloc.waitq = NULL;
1aac1400
HD
3128 shmem_falloc.start = start;
3129 shmem_falloc.next = start;
3130 shmem_falloc.nr_falloced = 0;
3131 shmem_falloc.nr_unswapped = 0;
3132 spin_lock(&inode->i_lock);
3133 inode->i_private = &shmem_falloc;
3134 spin_unlock(&inode->i_lock);
3135
d144bf62
HD
3136 /*
3137 * info->fallocend is only relevant when huge pages might be
3138 * involved: to prevent split_huge_page() freeing fallocated
3139 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3140 */
3141 undo_fallocend = info->fallocend;
3142 if (info->fallocend < end)
3143 info->fallocend = end;
3144
050dcb5c 3145 for (index = start; index < end; ) {
b0802b22 3146 struct folio *folio;
e2d12e22
HD
3147
3148 /*
3149 * Good, the fallocate(2) manpage permits EINTR: we may have
3150 * been interrupted because we are using up too much memory.
3151 */
3152 if (signal_pending(current))
3153 error = -EINTR;
1aac1400
HD
3154 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3155 error = -ENOMEM;
e2d12e22 3156 else
b0802b22
MWO
3157 error = shmem_get_folio(inode, index, &folio,
3158 SGP_FALLOC);
e2d12e22 3159 if (error) {
d144bf62 3160 info->fallocend = undo_fallocend;
b0802b22 3161 /* Remove the !uptodate folios we added */
7f556567
HD
3162 if (index > start) {
3163 shmem_undo_range(inode,
3164 (loff_t)start << PAGE_SHIFT,
3165 ((loff_t)index << PAGE_SHIFT) - 1, true);
3166 }
1aac1400 3167 goto undone;
e2d12e22
HD
3168 }
3169
050dcb5c
HD
3170 /*
3171 * Here is a more important optimization than it appears:
b0802b22
MWO
3172 * a second SGP_FALLOC on the same large folio will clear it,
3173 * making it uptodate and un-undoable if we fail later.
050dcb5c 3174 */
b0802b22
MWO
3175 index = folio_next_index(folio);
3176 /* Beware 32-bit wraparound */
3177 if (!index)
3178 index--;
050dcb5c 3179
1aac1400
HD
3180 /*
3181 * Inform shmem_writepage() how far we have reached.
3182 * No need for lock or barrier: we have the page lock.
3183 */
b0802b22 3184 if (!folio_test_uptodate(folio))
050dcb5c
HD
3185 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3186 shmem_falloc.next = index;
1aac1400 3187
e2d12e22 3188 /*
b0802b22 3189 * If !uptodate, leave it that way so that freeable folios
1635f6a7 3190 * can be recognized if we need to rollback on error later.
b0802b22
MWO
3191 * But mark it dirty so that memory pressure will swap rather
3192 * than free the folios we are allocating (and SGP_CACHE folios
e2d12e22
HD
3193 * might still be clean: we now need to mark those dirty too).
3194 */
b0802b22
MWO
3195 folio_mark_dirty(folio);
3196 folio_unlock(folio);
3197 folio_put(folio);
e2d12e22
HD
3198 cond_resched();
3199 }
3200
3201 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3202 i_size_write(inode, offset + len);
1aac1400
HD
3203undone:
3204 spin_lock(&inode->i_lock);
3205 inode->i_private = NULL;
3206 spin_unlock(&inode->i_lock);
e2d12e22 3207out:
15f242bb
HD
3208 if (!error)
3209 file_modified(file);
5955102c 3210 inode_unlock(inode);
83e4fa9c
HD
3211 return error;
3212}
3213
726c3342 3214static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1da177e4 3215{
726c3342 3216 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1da177e4
LT
3217
3218 buf->f_type = TMPFS_MAGIC;
09cbfeaf 3219 buf->f_bsize = PAGE_SIZE;
1da177e4 3220 buf->f_namelen = NAME_MAX;
0edd73b3 3221 if (sbinfo->max_blocks) {
1da177e4 3222 buf->f_blocks = sbinfo->max_blocks;
41ffe5d5
HD
3223 buf->f_bavail =
3224 buf->f_bfree = sbinfo->max_blocks -
3225 percpu_counter_sum(&sbinfo->used_blocks);
0edd73b3
HD
3226 }
3227 if (sbinfo->max_inodes) {
1da177e4 3228 buf->f_files = sbinfo->max_inodes;
e07c469e 3229 buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
1da177e4
LT
3230 }
3231 /* else leave those fields 0 like simple_statfs */
59cda49e
AG
3232
3233 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3234
1da177e4
LT
3235 return 0;
3236}
3237
3238/*
3239 * File creation. Allocate an inode, and we're done..
3240 */
3241static int
5ebb29be 3242shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
549c7297 3243 struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 3244{
0b0a0806 3245 struct inode *inode;
71480663 3246 int error;
1da177e4 3247
7a80e5b8 3248 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
71480663
CM
3249 if (IS_ERR(inode))
3250 return PTR_ERR(inode);
37ec43cd 3251
71480663
CM
3252 error = simple_acl_create(dir, inode);
3253 if (error)
3254 goto out_iput;
9be7d5b0 3255 error = security_inode_init_security(inode, dir, &dentry->d_name,
71480663
CM
3256 shmem_initxattrs, NULL);
3257 if (error && error != -EOPNOTSUPP)
3258 goto out_iput;
3259
a2e45955
CL
3260 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3261 if (error)
3262 goto out_iput;
3263
71480663 3264 dir->i_size += BOGO_DIRENT_SIZE;
cf2766bb 3265 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
71480663
CM
3266 inode_inc_iversion(dir);
3267 d_instantiate(dentry, inode);
3268 dget(dentry); /* Extra count - pin the dentry in core */
1da177e4 3269 return error;
71480663 3270
feda821e
CH
3271out_iput:
3272 iput(inode);
3273 return error;
1da177e4
LT
3274}
3275
60545d0d 3276static int
011e2b71 3277shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
863f144f 3278 struct file *file, umode_t mode)
60545d0d
AV
3279{
3280 struct inode *inode;
71480663 3281 int error;
60545d0d 3282
7a80e5b8 3283 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
71480663
CM
3284 if (IS_ERR(inode)) {
3285 error = PTR_ERR(inode);
3286 goto err_out;
60545d0d 3287 }
9be7d5b0 3288 error = security_inode_init_security(inode, dir, NULL,
71480663
CM
3289 shmem_initxattrs, NULL);
3290 if (error && error != -EOPNOTSUPP)
3291 goto out_iput;
3292 error = simple_acl_create(dir, inode);
3293 if (error)
3294 goto out_iput;
3295 d_tmpfile(file, inode);
3296
3297err_out:
863f144f 3298 return finish_open_simple(file, error);
feda821e
CH
3299out_iput:
3300 iput(inode);
3301 return error;
60545d0d
AV
3302}
3303
c54bd91e 3304static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
549c7297 3305 struct dentry *dentry, umode_t mode)
1da177e4
LT
3306{
3307 int error;
3308
7a80e5b8
GS
3309 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3310 if (error)
1da177e4 3311 return error;
d8c76e6f 3312 inc_nlink(dir);
1da177e4
LT
3313 return 0;
3314}
3315
6c960e68 3316static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
549c7297 3317 struct dentry *dentry, umode_t mode, bool excl)
1da177e4 3318{
7a80e5b8 3319 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
1da177e4
LT
3320}
3321
3322/*
3323 * Link a file..
3324 */
9be7d5b0
HD
3325static int shmem_link(struct dentry *old_dentry, struct inode *dir,
3326 struct dentry *dentry)
1da177e4 3327{
75c3cfa8 3328 struct inode *inode = d_inode(old_dentry);
29b00e60 3329 int ret = 0;
1da177e4
LT
3330
3331 /*
3332 * No ordinary (disk based) filesystem counts links as inodes;
3333 * but each new link needs a new dentry, pinning lowmem, and
3334 * tmpfs dentries cannot be pruned until they are unlinked.
1062af92
DW
3335 * But if an O_TMPFILE file is linked into the tmpfs, the
3336 * first link must skip that, to get the accounting right.
1da177e4 3337 */
1062af92 3338 if (inode->i_nlink) {
e809d5f0 3339 ret = shmem_reserve_inode(inode->i_sb, NULL);
1062af92
DW
3340 if (ret)
3341 goto out;
3342 }
1da177e4 3343
a2e45955
CL
3344 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3345 if (ret) {
3346 if (inode->i_nlink)
2daf18a7 3347 shmem_free_inode(inode->i_sb, 0);
a2e45955
CL
3348 goto out;
3349 }
3350
1da177e4 3351 dir->i_size += BOGO_DIRENT_SIZE;
cf2766bb
JL
3352 inode_set_mtime_to_ts(dir,
3353 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
36f05cab 3354 inode_inc_iversion(dir);
d8c76e6f 3355 inc_nlink(inode);
7de9c6ee 3356 ihold(inode); /* New dentry reference */
9be7d5b0 3357 dget(dentry); /* Extra pinning count for the created dentry */
1da177e4 3358 d_instantiate(dentry, inode);
5b04c689
PE
3359out:
3360 return ret;
1da177e4
LT
3361}
3362
3363static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3364{
75c3cfa8 3365 struct inode *inode = d_inode(dentry);
1da177e4 3366
5b04c689 3367 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2daf18a7 3368 shmem_free_inode(inode->i_sb, 0);
1da177e4 3369
a2e45955 3370 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
1da177e4
LT
3371
3372 dir->i_size -= BOGO_DIRENT_SIZE;
cf2766bb
JL
3373 inode_set_mtime_to_ts(dir,
3374 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
36f05cab 3375 inode_inc_iversion(dir);
9a53c3a7 3376 drop_nlink(inode);
9be7d5b0 3377 dput(dentry); /* Undo the count from "create" - does all the work */
1da177e4
LT
3378 return 0;
3379}
3380
3381static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3382{
3383 if (!simple_empty(dentry))
3384 return -ENOTEMPTY;
3385
75c3cfa8 3386 drop_nlink(d_inode(dentry));
9a53c3a7 3387 drop_nlink(dir);
1da177e4
LT
3388 return shmem_unlink(dir, dentry);
3389}
3390
e18275ae 3391static int shmem_whiteout(struct mnt_idmap *idmap,
549c7297 3392 struct inode *old_dir, struct dentry *old_dentry)
46fdb794
MS
3393{
3394 struct dentry *whiteout;
3395 int error;
3396
3397 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3398 if (!whiteout)
3399 return -ENOMEM;
3400
7a80e5b8 3401 error = shmem_mknod(idmap, old_dir, whiteout,
46fdb794
MS
3402 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3403 dput(whiteout);
3404 if (error)
3405 return error;
3406
3407 /*
3408 * Cheat and hash the whiteout while the old dentry is still in
3409 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3410 *
3411 * d_lookup() will consistently find one of them at this point,
3412 * not sure which one, but that isn't even important.
3413 */
3414 d_rehash(whiteout);
3415 return 0;
3416}
3417
1da177e4
LT
3418/*
3419 * The VFS layer already does all the dentry stuff for rename,
3420 * we just have to decrement the usage count for the target if
3421 * it exists so that the VFS layer correctly free's it when it
3422 * gets overwritten.
3423 */
e18275ae 3424static int shmem_rename2(struct mnt_idmap *idmap,
549c7297
CB
3425 struct inode *old_dir, struct dentry *old_dentry,
3426 struct inode *new_dir, struct dentry *new_dentry,
3427 unsigned int flags)
1da177e4 3428{
75c3cfa8 3429 struct inode *inode = d_inode(old_dentry);
1da177e4 3430 int they_are_dirs = S_ISDIR(inode->i_mode);
a2e45955 3431 int error;
1da177e4 3432
46fdb794 3433 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3b69ff51
MS
3434 return -EINVAL;
3435
37456771 3436 if (flags & RENAME_EXCHANGE)
a2e45955
CL
3437 return simple_offset_rename_exchange(old_dir, old_dentry,
3438 new_dir, new_dentry);
37456771 3439
1da177e4
LT
3440 if (!simple_empty(new_dentry))
3441 return -ENOTEMPTY;
3442
46fdb794 3443 if (flags & RENAME_WHITEOUT) {
7a80e5b8 3444 error = shmem_whiteout(idmap, old_dir, old_dentry);
46fdb794
MS
3445 if (error)
3446 return error;
3447 }
3448
a2e45955
CL
3449 simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry);
3450 error = simple_offset_add(shmem_get_offset_ctx(new_dir), old_dentry);
3451 if (error)
3452 return error;
3453
75c3cfa8 3454 if (d_really_is_positive(new_dentry)) {
1da177e4 3455 (void) shmem_unlink(new_dir, new_dentry);
b928095b 3456 if (they_are_dirs) {
75c3cfa8 3457 drop_nlink(d_inode(new_dentry));
9a53c3a7 3458 drop_nlink(old_dir);
b928095b 3459 }
1da177e4 3460 } else if (they_are_dirs) {
9a53c3a7 3461 drop_nlink(old_dir);
d8c76e6f 3462 inc_nlink(new_dir);
1da177e4
LT
3463 }
3464
3465 old_dir->i_size -= BOGO_DIRENT_SIZE;
3466 new_dir->i_size += BOGO_DIRENT_SIZE;
944d0d9d 3467 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
36f05cab
JL
3468 inode_inc_iversion(old_dir);
3469 inode_inc_iversion(new_dir);
1da177e4
LT
3470 return 0;
3471}
3472
7a77db95 3473static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
549c7297 3474 struct dentry *dentry, const char *symname)
1da177e4
LT
3475{
3476 int error;
3477 int len;
3478 struct inode *inode;
7ad0414b 3479 struct folio *folio;
1da177e4
LT
3480
3481 len = strlen(symname) + 1;
09cbfeaf 3482 if (len > PAGE_SIZE)
1da177e4
LT
3483 return -ENAMETOOLONG;
3484
7a80e5b8 3485 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
0825a6f9 3486 VM_NORESERVE);
71480663
CM
3487 if (IS_ERR(inode))
3488 return PTR_ERR(inode);
1da177e4 3489
9d8f13ba 3490 error = security_inode_init_security(inode, dir, &dentry->d_name,
6d9d88d0 3491 shmem_initxattrs, NULL);
23a31d87
CL
3492 if (error && error != -EOPNOTSUPP)
3493 goto out_iput;
570bc1c2 3494
a2e45955
CL
3495 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3496 if (error)
3497 goto out_iput;
570bc1c2 3498
1da177e4 3499 inode->i_size = len-1;
69f07ec9 3500 if (len <= SHORT_SYMLINK_LEN) {
3ed47db3
AV
3501 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3502 if (!inode->i_link) {
23a31d87 3503 error = -ENOMEM;
a2e45955 3504 goto out_remove_offset;
69f07ec9
HD
3505 }
3506 inode->i_op = &shmem_short_symlink_operations;
1da177e4 3507 } else {
e8ecde25 3508 inode_nohighmem(inode);
7ad0414b 3509 error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
23a31d87 3510 if (error)
a2e45955 3511 goto out_remove_offset;
14fcc23f 3512 inode->i_mapping->a_ops = &shmem_aops;
1da177e4 3513 inode->i_op = &shmem_symlink_inode_operations;
7ad0414b
MWO
3514 memcpy(folio_address(folio), symname, len);
3515 folio_mark_uptodate(folio);
3516 folio_mark_dirty(folio);
3517 folio_unlock(folio);
3518 folio_put(folio);
1da177e4 3519 }
1da177e4 3520 dir->i_size += BOGO_DIRENT_SIZE;
cf2766bb 3521 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
36f05cab 3522 inode_inc_iversion(dir);
1da177e4
LT
3523 d_instantiate(dentry, inode);
3524 dget(dentry);
3525 return 0;
a2e45955
CL
3526
3527out_remove_offset:
3528 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
23a31d87
CL
3529out_iput:
3530 iput(inode);
3531 return error;
1da177e4
LT
3532}
3533
fceef393 3534static void shmem_put_link(void *arg)
1da177e4 3535{
e4b57722
MWO
3536 folio_mark_accessed(arg);
3537 folio_put(arg);
1da177e4
LT
3538}
3539
9be7d5b0 3540static const char *shmem_get_link(struct dentry *dentry, struct inode *inode,
fceef393 3541 struct delayed_call *done)
1da177e4 3542{
e4b57722 3543 struct folio *folio = NULL;
6b255391 3544 int error;
e4b57722 3545
6a6c9904 3546 if (!dentry) {
e4b57722 3547 folio = filemap_get_folio(inode->i_mapping, 0);
66dabbb6 3548 if (IS_ERR(folio))
6a6c9904 3549 return ERR_PTR(-ECHILD);
7459c149 3550 if (PageHWPoison(folio_page(folio, 0)) ||
e4b57722
MWO
3551 !folio_test_uptodate(folio)) {
3552 folio_put(folio);
6a6c9904
AV
3553 return ERR_PTR(-ECHILD);
3554 }
3555 } else {
e4b57722 3556 error = shmem_get_folio(inode, 0, &folio, SGP_READ);
6a6c9904
AV
3557 if (error)
3558 return ERR_PTR(error);
e4b57722 3559 if (!folio)
a7605426 3560 return ERR_PTR(-ECHILD);
7459c149 3561 if (PageHWPoison(folio_page(folio, 0))) {
e4b57722
MWO
3562 folio_unlock(folio);
3563 folio_put(folio);
a7605426
YS
3564 return ERR_PTR(-ECHILD);
3565 }
e4b57722 3566 folio_unlock(folio);
6a6c9904 3567 }
e4b57722
MWO
3568 set_delayed_call(done, shmem_put_link, folio);
3569 return folio_address(folio);
1da177e4
LT
3570}
3571
b09e0fa4 3572#ifdef CONFIG_TMPFS_XATTR
e408e695
TT
3573
3574static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3575{
3576 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3577
3578 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3579
3580 return 0;
3581}
3582
8782a9ae 3583static int shmem_fileattr_set(struct mnt_idmap *idmap,
e408e695
TT
3584 struct dentry *dentry, struct fileattr *fa)
3585{
3586 struct inode *inode = d_inode(dentry);
3587 struct shmem_inode_info *info = SHMEM_I(inode);
3588
3589 if (fileattr_has_fsx(fa))
3590 return -EOPNOTSUPP;
cb241339
HD
3591 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3592 return -EOPNOTSUPP;
e408e695
TT
3593
3594 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3595 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3596
cb241339 3597 shmem_set_inode_flags(inode, info->fsflags);
65287334 3598 inode_set_ctime_current(inode);
36f05cab 3599 inode_inc_iversion(inode);
e408e695
TT
3600 return 0;
3601}
3602
46711810 3603/*
b09e0fa4
EP
3604 * Superblocks without xattr inode operations may get some security.* xattr
3605 * support from the LSM "for free". As soon as we have any other xattrs
39f0247d
AG
3606 * like ACLs, we also need to implement the security.* handlers at
3607 * filesystem level, though.
3608 */
3609
6d9d88d0
JS
3610/*
3611 * Callback for security_inode_init_security() for acquiring xattrs.
3612 */
3613static int shmem_initxattrs(struct inode *inode,
9be7d5b0 3614 const struct xattr *xattr_array, void *fs_info)
6d9d88d0
JS
3615{
3616 struct shmem_inode_info *info = SHMEM_I(inode);
2daf18a7 3617 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
6d9d88d0 3618 const struct xattr *xattr;
38f38657 3619 struct simple_xattr *new_xattr;
2daf18a7 3620 size_t ispace = 0;
6d9d88d0
JS
3621 size_t len;
3622
2daf18a7
HD
3623 if (sbinfo->max_inodes) {
3624 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3625 ispace += simple_xattr_space(xattr->name,
3626 xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
3627 }
3628 if (ispace) {
3629 raw_spin_lock(&sbinfo->stat_lock);
3630 if (sbinfo->free_ispace < ispace)
3631 ispace = 0;
3632 else
3633 sbinfo->free_ispace -= ispace;
3634 raw_spin_unlock(&sbinfo->stat_lock);
3635 if (!ispace)
3636 return -ENOSPC;
3637 }
3638 }
3639
6d9d88d0 3640 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
38f38657 3641 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
6d9d88d0 3642 if (!new_xattr)
2daf18a7 3643 break;
6d9d88d0
JS
3644
3645 len = strlen(xattr->name) + 1;
3646 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
572a3d1e 3647 GFP_KERNEL_ACCOUNT);
6d9d88d0 3648 if (!new_xattr->name) {
3bef735a 3649 kvfree(new_xattr);
2daf18a7 3650 break;
6d9d88d0
JS
3651 }
3652
3653 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3654 XATTR_SECURITY_PREFIX_LEN);
3655 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3656 xattr->name, len);
3657
3b4c7bc0 3658 simple_xattr_add(&info->xattrs, new_xattr);
6d9d88d0
JS
3659 }
3660
2daf18a7
HD
3661 if (xattr->name != NULL) {
3662 if (ispace) {
3663 raw_spin_lock(&sbinfo->stat_lock);
3664 sbinfo->free_ispace += ispace;
3665 raw_spin_unlock(&sbinfo->stat_lock);
3666 }
3667 simple_xattrs_free(&info->xattrs, NULL);
3668 return -ENOMEM;
3669 }
3670
6d9d88d0
JS
3671 return 0;
3672}
3673
aa7c5241 3674static int shmem_xattr_handler_get(const struct xattr_handler *handler,
b296821a
AV
3675 struct dentry *unused, struct inode *inode,
3676 const char *name, void *buffer, size_t size)
b09e0fa4 3677{
b296821a 3678 struct shmem_inode_info *info = SHMEM_I(inode);
b09e0fa4 3679
aa7c5241 3680 name = xattr_full_name(handler, name);
38f38657 3681 return simple_xattr_get(&info->xattrs, name, buffer, size);
b09e0fa4
EP
3682}
3683
aa7c5241 3684static int shmem_xattr_handler_set(const struct xattr_handler *handler,
39f60c1c 3685 struct mnt_idmap *idmap,
59301226
AV
3686 struct dentry *unused, struct inode *inode,
3687 const char *name, const void *value,
3688 size_t size, int flags)
b09e0fa4 3689{
59301226 3690 struct shmem_inode_info *info = SHMEM_I(inode);
2daf18a7 3691 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
5de75970 3692 struct simple_xattr *old_xattr;
2daf18a7 3693 size_t ispace = 0;
b09e0fa4 3694
aa7c5241 3695 name = xattr_full_name(handler, name);
2daf18a7
HD
3696 if (value && sbinfo->max_inodes) {
3697 ispace = simple_xattr_space(name, size);
3698 raw_spin_lock(&sbinfo->stat_lock);
3699 if (sbinfo->free_ispace < ispace)
3700 ispace = 0;
3701 else
3702 sbinfo->free_ispace -= ispace;
3703 raw_spin_unlock(&sbinfo->stat_lock);
3704 if (!ispace)
3705 return -ENOSPC;
3706 }
3707
5de75970
HD
3708 old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
3709 if (!IS_ERR(old_xattr)) {
2daf18a7
HD
3710 ispace = 0;
3711 if (old_xattr && sbinfo->max_inodes)
3712 ispace = simple_xattr_space(old_xattr->name,
3713 old_xattr->size);
5de75970
HD
3714 simple_xattr_free(old_xattr);
3715 old_xattr = NULL;
65287334 3716 inode_set_ctime_current(inode);
36f05cab
JL
3717 inode_inc_iversion(inode);
3718 }
2daf18a7
HD
3719 if (ispace) {
3720 raw_spin_lock(&sbinfo->stat_lock);
3721 sbinfo->free_ispace += ispace;
3722 raw_spin_unlock(&sbinfo->stat_lock);
3723 }
5de75970 3724 return PTR_ERR(old_xattr);
b09e0fa4
EP
3725}
3726
aa7c5241
AG
3727static const struct xattr_handler shmem_security_xattr_handler = {
3728 .prefix = XATTR_SECURITY_PREFIX,
3729 .get = shmem_xattr_handler_get,
3730 .set = shmem_xattr_handler_set,
3731};
b09e0fa4 3732
aa7c5241
AG
3733static const struct xattr_handler shmem_trusted_xattr_handler = {
3734 .prefix = XATTR_TRUSTED_PREFIX,
3735 .get = shmem_xattr_handler_get,
3736 .set = shmem_xattr_handler_set,
3737};
b09e0fa4 3738
2daf18a7
HD
3739static const struct xattr_handler shmem_user_xattr_handler = {
3740 .prefix = XATTR_USER_PREFIX,
3741 .get = shmem_xattr_handler_get,
3742 .set = shmem_xattr_handler_set,
3743};
3744
2f502860 3745static const struct xattr_handler * const shmem_xattr_handlers[] = {
aa7c5241
AG
3746 &shmem_security_xattr_handler,
3747 &shmem_trusted_xattr_handler,
2daf18a7 3748 &shmem_user_xattr_handler,
aa7c5241
AG
3749 NULL
3750};
b09e0fa4
EP
3751
3752static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3753{
75c3cfa8 3754 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
786534b9 3755 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
b09e0fa4
EP
3756}
3757#endif /* CONFIG_TMPFS_XATTR */
3758
69f07ec9 3759static const struct inode_operations shmem_short_symlink_operations = {
f7cd16a5 3760 .getattr = shmem_getattr,
e09764cf 3761 .setattr = shmem_setattr,
6b255391 3762 .get_link = simple_get_link,
b09e0fa4 3763#ifdef CONFIG_TMPFS_XATTR
b09e0fa4 3764 .listxattr = shmem_listxattr,
b09e0fa4
EP
3765#endif
3766};
3767
3768static const struct inode_operations shmem_symlink_inode_operations = {
f7cd16a5 3769 .getattr = shmem_getattr,
e09764cf 3770 .setattr = shmem_setattr,
6b255391 3771 .get_link = shmem_get_link,
b09e0fa4 3772#ifdef CONFIG_TMPFS_XATTR
b09e0fa4 3773 .listxattr = shmem_listxattr,
39f0247d 3774#endif
b09e0fa4 3775};
39f0247d 3776
91828a40
DG
3777static struct dentry *shmem_get_parent(struct dentry *child)
3778{
3779 return ERR_PTR(-ESTALE);
3780}
3781
3782static int shmem_match(struct inode *ino, void *vfh)
3783{
3784 __u32 *fh = vfh;
3785 __u64 inum = fh[2];
3786 inum = (inum << 32) | fh[1];
3787 return ino->i_ino == inum && fh[0] == ino->i_generation;
3788}
3789
12ba780d
AG
3790/* Find any alias of inode, but prefer a hashed alias */
3791static struct dentry *shmem_find_alias(struct inode *inode)
3792{
3793 struct dentry *alias = d_find_alias(inode);
3794
3795 return alias ?: d_find_any_alias(inode);
3796}
3797
480b116c
CH
3798static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3799 struct fid *fid, int fh_len, int fh_type)
91828a40 3800{
91828a40 3801 struct inode *inode;
480b116c 3802 struct dentry *dentry = NULL;
35c2a7f4 3803 u64 inum;
480b116c
CH
3804
3805 if (fh_len < 3)
3806 return NULL;
91828a40 3807
35c2a7f4
HD
3808 inum = fid->raw[2];
3809 inum = (inum << 32) | fid->raw[1];
3810
480b116c
CH
3811 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3812 shmem_match, fid->raw);
91828a40 3813 if (inode) {
12ba780d 3814 dentry = shmem_find_alias(inode);
91828a40
DG
3815 iput(inode);
3816 }
3817
480b116c 3818 return dentry;
91828a40
DG
3819}
3820
b0b0382b
AV
3821static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3822 struct inode *parent)
91828a40 3823{
5fe0c237
AK
3824 if (*len < 3) {
3825 *len = 3;
94e07a75 3826 return FILEID_INVALID;
5fe0c237 3827 }
91828a40 3828
1d3382cb 3829 if (inode_unhashed(inode)) {
91828a40
DG
3830 /* Unfortunately insert_inode_hash is not idempotent,
3831 * so as we hash inodes here rather than at creation
3832 * time, we need a lock to ensure we only try
3833 * to do it once
3834 */
3835 static DEFINE_SPINLOCK(lock);
3836 spin_lock(&lock);
1d3382cb 3837 if (inode_unhashed(inode))
91828a40
DG
3838 __insert_inode_hash(inode,
3839 inode->i_ino + inode->i_generation);
3840 spin_unlock(&lock);
3841 }
3842
3843 fh[0] = inode->i_generation;
3844 fh[1] = inode->i_ino;
3845 fh[2] = ((__u64)inode->i_ino) >> 32;
3846
3847 *len = 3;
3848 return 1;
3849}
3850
39655164 3851static const struct export_operations shmem_export_ops = {
91828a40 3852 .get_parent = shmem_get_parent,
91828a40 3853 .encode_fh = shmem_encode_fh,
480b116c 3854 .fh_to_dentry = shmem_fh_to_dentry,
91828a40
DG
3855};
3856
626c3920
AV
3857enum shmem_param {
3858 Opt_gid,
3859 Opt_huge,
3860 Opt_mode,
3861 Opt_mpol,
3862 Opt_nr_blocks,
3863 Opt_nr_inodes,
3864 Opt_size,
3865 Opt_uid,
ea3271f7
CD
3866 Opt_inode32,
3867 Opt_inode64,
2c6efe9c 3868 Opt_noswap,
e09764cf
CM
3869 Opt_quota,
3870 Opt_usrquota,
3871 Opt_grpquota,
de4c0e7c
LC
3872 Opt_usrquota_block_hardlimit,
3873 Opt_usrquota_inode_hardlimit,
3874 Opt_grpquota_block_hardlimit,
3875 Opt_grpquota_inode_hardlimit,
626c3920
AV
3876};
3877
5eede625 3878static const struct constant_table shmem_param_enums_huge[] = {
2710c957
AV
3879 {"never", SHMEM_HUGE_NEVER },
3880 {"always", SHMEM_HUGE_ALWAYS },
3881 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3882 {"advise", SHMEM_HUGE_ADVISE },
2710c957
AV
3883 {}
3884};
3885
d7167b14 3886const struct fs_parameter_spec shmem_fs_parameters[] = {
626c3920 3887 fsparam_u32 ("gid", Opt_gid),
2710c957 3888 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
626c3920
AV
3889 fsparam_u32oct("mode", Opt_mode),
3890 fsparam_string("mpol", Opt_mpol),
3891 fsparam_string("nr_blocks", Opt_nr_blocks),
3892 fsparam_string("nr_inodes", Opt_nr_inodes),
3893 fsparam_string("size", Opt_size),
3894 fsparam_u32 ("uid", Opt_uid),
ea3271f7
CD
3895 fsparam_flag ("inode32", Opt_inode32),
3896 fsparam_flag ("inode64", Opt_inode64),
2c6efe9c 3897 fsparam_flag ("noswap", Opt_noswap),
e09764cf
CM
3898#ifdef CONFIG_TMPFS_QUOTA
3899 fsparam_flag ("quota", Opt_quota),
3900 fsparam_flag ("usrquota", Opt_usrquota),
3901 fsparam_flag ("grpquota", Opt_grpquota),
de4c0e7c
LC
3902 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
3903 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
3904 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
3905 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
e09764cf 3906#endif
626c3920
AV
3907 {}
3908};
3909
f3235626 3910static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
1da177e4 3911{
f3235626 3912 struct shmem_options *ctx = fc->fs_private;
626c3920
AV
3913 struct fs_parse_result result;
3914 unsigned long long size;
e04dc423 3915 char *rest;
626c3920 3916 int opt;
0200679f
CB
3917 kuid_t kuid;
3918 kgid_t kgid;
626c3920 3919
d7167b14 3920 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
f3235626 3921 if (opt < 0)
626c3920 3922 return opt;
1da177e4 3923
626c3920
AV
3924 switch (opt) {
3925 case Opt_size:
3926 size = memparse(param->string, &rest);
e04dc423
AV
3927 if (*rest == '%') {
3928 size <<= PAGE_SHIFT;
3929 size *= totalram_pages();
3930 do_div(size, 100);
3931 rest++;
3932 }
3933 if (*rest)
626c3920 3934 goto bad_value;
e04dc423
AV
3935 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3936 ctx->seen |= SHMEM_SEEN_BLOCKS;
626c3920
AV
3937 break;
3938 case Opt_nr_blocks:
3939 ctx->blocks = memparse(param->string, &rest);
e07c469e 3940 if (*rest || ctx->blocks > LONG_MAX)
626c3920 3941 goto bad_value;
e04dc423 3942 ctx->seen |= SHMEM_SEEN_BLOCKS;
626c3920
AV
3943 break;
3944 case Opt_nr_inodes:
3945 ctx->inodes = memparse(param->string, &rest);
e07c469e 3946 if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
626c3920 3947 goto bad_value;
e04dc423 3948 ctx->seen |= SHMEM_SEEN_INODES;
626c3920
AV
3949 break;
3950 case Opt_mode:
3951 ctx->mode = result.uint_32 & 07777;
3952 break;
3953 case Opt_uid:
0200679f
CB
3954 kuid = make_kuid(current_user_ns(), result.uint_32);
3955 if (!uid_valid(kuid))
626c3920 3956 goto bad_value;
0200679f
CB
3957
3958 /*
3959 * The requested uid must be representable in the
3960 * filesystem's idmapping.
3961 */
3962 if (!kuid_has_mapping(fc->user_ns, kuid))
626c3920 3963 goto bad_value;
0200679f
CB
3964
3965 ctx->uid = kuid;
626c3920
AV
3966 break;
3967 case Opt_gid:
0200679f
CB
3968 kgid = make_kgid(current_user_ns(), result.uint_32);
3969 if (!gid_valid(kgid))
626c3920 3970 goto bad_value;
0200679f
CB
3971
3972 /*
3973 * The requested gid must be representable in the
3974 * filesystem's idmapping.
3975 */
3976 if (!kgid_has_mapping(fc->user_ns, kgid))
626c3920 3977 goto bad_value;
0200679f
CB
3978
3979 ctx->gid = kgid;
626c3920
AV
3980 break;
3981 case Opt_huge:
3982 ctx->huge = result.uint_32;
3983 if (ctx->huge != SHMEM_HUGE_NEVER &&
396bcc52 3984 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
626c3920
AV
3985 has_transparent_hugepage()))
3986 goto unsupported_parameter;
e04dc423 3987 ctx->seen |= SHMEM_SEEN_HUGE;
626c3920
AV
3988 break;
3989 case Opt_mpol:
3990 if (IS_ENABLED(CONFIG_NUMA)) {
3991 mpol_put(ctx->mpol);
3992 ctx->mpol = NULL;
3993 if (mpol_parse_str(param->string, &ctx->mpol))
3994 goto bad_value;
3995 break;
3996 }
3997 goto unsupported_parameter;
ea3271f7
CD
3998 case Opt_inode32:
3999 ctx->full_inums = false;
4000 ctx->seen |= SHMEM_SEEN_INUMS;
4001 break;
4002 case Opt_inode64:
4003 if (sizeof(ino_t) < 8) {
4004 return invalfc(fc,
4005 "Cannot use inode64 with <64bit inums in kernel\n");
4006 }
4007 ctx->full_inums = true;
4008 ctx->seen |= SHMEM_SEEN_INUMS;
4009 break;
2c6efe9c 4010 case Opt_noswap:
01106e14
CB
4011 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
4012 return invalfc(fc,
4013 "Turning off swap in unprivileged tmpfs mounts unsupported");
4014 }
2c6efe9c
LC
4015 ctx->noswap = true;
4016 ctx->seen |= SHMEM_SEEN_NOSWAP;
4017 break;
e09764cf
CM
4018 case Opt_quota:
4019 if (fc->user_ns != &init_user_ns)
4020 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4021 ctx->seen |= SHMEM_SEEN_QUOTA;
4022 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
4023 break;
4024 case Opt_usrquota:
4025 if (fc->user_ns != &init_user_ns)
4026 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4027 ctx->seen |= SHMEM_SEEN_QUOTA;
4028 ctx->quota_types |= QTYPE_MASK_USR;
4029 break;
4030 case Opt_grpquota:
4031 if (fc->user_ns != &init_user_ns)
4032 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4033 ctx->seen |= SHMEM_SEEN_QUOTA;
4034 ctx->quota_types |= QTYPE_MASK_GRP;
4035 break;
de4c0e7c
LC
4036 case Opt_usrquota_block_hardlimit:
4037 size = memparse(param->string, &rest);
4038 if (*rest || !size)
4039 goto bad_value;
4040 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4041 return invalfc(fc,
4042 "User quota block hardlimit too large.");
4043 ctx->qlimits.usrquota_bhardlimit = size;
4044 break;
4045 case Opt_grpquota_block_hardlimit:
4046 size = memparse(param->string, &rest);
4047 if (*rest || !size)
4048 goto bad_value;
4049 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4050 return invalfc(fc,
4051 "Group quota block hardlimit too large.");
4052 ctx->qlimits.grpquota_bhardlimit = size;
4053 break;
4054 case Opt_usrquota_inode_hardlimit:
4055 size = memparse(param->string, &rest);
4056 if (*rest || !size)
4057 goto bad_value;
4058 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4059 return invalfc(fc,
4060 "User quota inode hardlimit too large.");
4061 ctx->qlimits.usrquota_ihardlimit = size;
4062 break;
4063 case Opt_grpquota_inode_hardlimit:
4064 size = memparse(param->string, &rest);
4065 if (*rest || !size)
4066 goto bad_value;
4067 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4068 return invalfc(fc,
4069 "Group quota inode hardlimit too large.");
4070 ctx->qlimits.grpquota_ihardlimit = size;
4071 break;
e04dc423
AV
4072 }
4073 return 0;
4074
626c3920 4075unsupported_parameter:
f35aa2bc 4076 return invalfc(fc, "Unsupported parameter '%s'", param->key);
626c3920 4077bad_value:
f35aa2bc 4078 return invalfc(fc, "Bad value for '%s'", param->key);
e04dc423
AV
4079}
4080
f3235626 4081static int shmem_parse_options(struct fs_context *fc, void *data)
e04dc423 4082{
f3235626
DH
4083 char *options = data;
4084
33f37c64
AV
4085 if (options) {
4086 int err = security_sb_eat_lsm_opts(options, &fc->security);
4087 if (err)
4088 return err;
4089 }
4090
b00dc3ad 4091 while (options != NULL) {
626c3920 4092 char *this_char = options;
b00dc3ad
HD
4093 for (;;) {
4094 /*
4095 * NUL-terminate this option: unfortunately,
4096 * mount options form a comma-separated list,
4097 * but mpol's nodelist may also contain commas.
4098 */
4099 options = strchr(options, ',');
4100 if (options == NULL)
4101 break;
4102 options++;
4103 if (!isdigit(*options)) {
4104 options[-1] = '\0';
4105 break;
4106 }
4107 }
626c3920 4108 if (*this_char) {
68d68ff6 4109 char *value = strchr(this_char, '=');
f3235626 4110 size_t len = 0;
626c3920
AV
4111 int err;
4112
4113 if (value) {
4114 *value++ = '\0';
f3235626 4115 len = strlen(value);
626c3920 4116 }
f3235626
DH
4117 err = vfs_parse_fs_string(fc, this_char, value, len);
4118 if (err < 0)
4119 return err;
1da177e4 4120 }
1da177e4
LT
4121 }
4122 return 0;
1da177e4
LT
4123}
4124
f3235626
DH
4125/*
4126 * Reconfigure a shmem filesystem.
f3235626
DH
4127 */
4128static int shmem_reconfigure(struct fs_context *fc)
1da177e4 4129{
f3235626
DH
4130 struct shmem_options *ctx = fc->fs_private;
4131 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
e07c469e 4132 unsigned long used_isp;
bf11b9a8 4133 struct mempolicy *mpol = NULL;
f3235626 4134 const char *err;
1da177e4 4135
bf11b9a8 4136 raw_spin_lock(&sbinfo->stat_lock);
e07c469e 4137 used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
0c98c8e1 4138
f3235626
DH
4139 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4140 if (!sbinfo->max_blocks) {
4141 err = "Cannot retroactively limit size";
0b5071dd 4142 goto out;
f3235626 4143 }
0b5071dd 4144 if (percpu_counter_compare(&sbinfo->used_blocks,
f3235626
DH
4145 ctx->blocks) > 0) {
4146 err = "Too small a size for current use";
0b5071dd 4147 goto out;
f3235626 4148 }
0b5071dd 4149 }
f3235626
DH
4150 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4151 if (!sbinfo->max_inodes) {
4152 err = "Cannot retroactively limit inodes";
0b5071dd 4153 goto out;
f3235626 4154 }
e07c469e 4155 if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
f3235626 4156 err = "Too few inodes for current use";
0b5071dd 4157 goto out;
f3235626 4158 }
0b5071dd 4159 }
0edd73b3 4160
ea3271f7
CD
4161 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4162 sbinfo->next_ino > UINT_MAX) {
4163 err = "Current inum too high to switch to 32-bit inums";
4164 goto out;
4165 }
2c6efe9c
LC
4166 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4167 err = "Cannot disable swap on remount";
4168 goto out;
4169 }
4170 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4171 err = "Cannot enable swap on remount if it was disabled on first mount";
4172 goto out;
4173 }
ea3271f7 4174
e09764cf
CM
4175 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4176 !sb_any_quota_loaded(fc->root->d_sb)) {
4177 err = "Cannot enable quota on remount";
4178 goto out;
4179 }
4180
de4c0e7c
LC
4181#ifdef CONFIG_TMPFS_QUOTA
4182#define CHANGED_LIMIT(name) \
4183 (ctx->qlimits.name## hardlimit && \
4184 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4185
4186 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4187 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4188 err = "Cannot change global quota limit on remount";
4189 goto out;
4190 }
4191#endif /* CONFIG_TMPFS_QUOTA */
4192
f3235626
DH
4193 if (ctx->seen & SHMEM_SEEN_HUGE)
4194 sbinfo->huge = ctx->huge;
ea3271f7
CD
4195 if (ctx->seen & SHMEM_SEEN_INUMS)
4196 sbinfo->full_inums = ctx->full_inums;
f3235626
DH
4197 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4198 sbinfo->max_blocks = ctx->blocks;
4199 if (ctx->seen & SHMEM_SEEN_INODES) {
4200 sbinfo->max_inodes = ctx->inodes;
e07c469e 4201 sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
0b5071dd 4202 }
71fe804b 4203
5f00110f
GT
4204 /*
4205 * Preserve previous mempolicy unless mpol remount option was specified.
4206 */
f3235626 4207 if (ctx->mpol) {
bf11b9a8 4208 mpol = sbinfo->mpol;
f3235626
DH
4209 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4210 ctx->mpol = NULL;
5f00110f 4211 }
2c6efe9c
LC
4212
4213 if (ctx->noswap)
4214 sbinfo->noswap = true;
4215
bf11b9a8
SAS
4216 raw_spin_unlock(&sbinfo->stat_lock);
4217 mpol_put(mpol);
f3235626 4218 return 0;
0edd73b3 4219out:
bf11b9a8 4220 raw_spin_unlock(&sbinfo->stat_lock);
f35aa2bc 4221 return invalfc(fc, "%s", err);
1da177e4 4222}
680d794b 4223
34c80b1d 4224static int shmem_show_options(struct seq_file *seq, struct dentry *root)
680d794b 4225{
34c80b1d 4226 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
283ebdee 4227 struct mempolicy *mpol;
680d794b 4228
4229 if (sbinfo->max_blocks != shmem_default_max_blocks())
b91742d8 4230 seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks));
680d794b 4231 if (sbinfo->max_inodes != shmem_default_max_inodes())
4232 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
0825a6f9 4233 if (sbinfo->mode != (0777 | S_ISVTX))
09208d15 4234 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
8751e039
EB
4235 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4236 seq_printf(seq, ",uid=%u",
4237 from_kuid_munged(&init_user_ns, sbinfo->uid));
4238 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4239 seq_printf(seq, ",gid=%u",
4240 from_kgid_munged(&init_user_ns, sbinfo->gid));
ea3271f7
CD
4241
4242 /*
4243 * Showing inode{64,32} might be useful even if it's the system default,
4244 * since then people don't have to resort to checking both here and
4245 * /proc/config.gz to confirm 64-bit inums were successfully applied
4246 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4247 *
4248 * We hide it when inode64 isn't the default and we are using 32-bit
4249 * inodes, since that probably just means the feature isn't even under
4250 * consideration.
4251 *
4252 * As such:
4253 *
4254 * +-----------------+-----------------+
4255 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4256 * +------------------+-----------------+-----------------+
4257 * | full_inums=true | show | show |
4258 * | full_inums=false | show | hide |
4259 * +------------------+-----------------+-----------------+
4260 *
4261 */
4262 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4263 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
396bcc52 4264#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5a6e75f8
KS
4265 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4266 if (sbinfo->huge)
4267 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4268#endif
283ebdee
TJ
4269 mpol = shmem_get_sbmpol(sbinfo);
4270 shmem_show_mpol(seq, mpol);
4271 mpol_put(mpol);
2c6efe9c
LC
4272 if (sbinfo->noswap)
4273 seq_printf(seq, ",noswap");
680d794b 4274 return 0;
4275}
9183df25 4276
680d794b 4277#endif /* CONFIG_TMPFS */
1da177e4
LT
4278
4279static void shmem_put_super(struct super_block *sb)
4280{
602586a8
HD
4281 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4282
e09764cf
CM
4283#ifdef CONFIG_TMPFS_QUOTA
4284 shmem_disable_quotas(sb);
4285#endif
e809d5f0 4286 free_percpu(sbinfo->ino_batch);
602586a8 4287 percpu_counter_destroy(&sbinfo->used_blocks);
49cd0a5c 4288 mpol_put(sbinfo->mpol);
602586a8 4289 kfree(sbinfo);
1da177e4
LT
4290 sb->s_fs_info = NULL;
4291}
4292
f3235626 4293static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
1da177e4 4294{
f3235626 4295 struct shmem_options *ctx = fc->fs_private;
1da177e4 4296 struct inode *inode;
0edd73b3 4297 struct shmem_sb_info *sbinfo;
71480663 4298 int error = -ENOMEM;
680d794b 4299
4300 /* Round up to L1_CACHE_BYTES to resist false sharing */
425fbf04 4301 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
680d794b 4302 L1_CACHE_BYTES), GFP_KERNEL);
4303 if (!sbinfo)
71480663 4304 return error;
680d794b 4305
680d794b 4306 sb->s_fs_info = sbinfo;
1da177e4 4307
0edd73b3 4308#ifdef CONFIG_TMPFS
1da177e4
LT
4309 /*
4310 * Per default we only allow half of the physical ram per
4311 * tmpfs instance, limiting inodes to one per page of lowmem;
4312 * but the internal instance is left unlimited.
4313 */
1751e8a6 4314 if (!(sb->s_flags & SB_KERNMOUNT)) {
f3235626
DH
4315 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4316 ctx->blocks = shmem_default_max_blocks();
4317 if (!(ctx->seen & SHMEM_SEEN_INODES))
4318 ctx->inodes = shmem_default_max_inodes();
ea3271f7
CD
4319 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4320 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
2c6efe9c 4321 sbinfo->noswap = ctx->noswap;
ca4e0519 4322 } else {
1751e8a6 4323 sb->s_flags |= SB_NOUSER;
1da177e4 4324 }
91828a40 4325 sb->s_export_op = &shmem_export_ops;
36f05cab 4326 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
1da177e4 4327#else
1751e8a6 4328 sb->s_flags |= SB_NOUSER;
1da177e4 4329#endif
f3235626 4330 sbinfo->max_blocks = ctx->blocks;
e07c469e
HD
4331 sbinfo->max_inodes = ctx->inodes;
4332 sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
e809d5f0
CD
4333 if (sb->s_flags & SB_KERNMOUNT) {
4334 sbinfo->ino_batch = alloc_percpu(ino_t);
4335 if (!sbinfo->ino_batch)
4336 goto failed;
4337 }
f3235626
DH
4338 sbinfo->uid = ctx->uid;
4339 sbinfo->gid = ctx->gid;
ea3271f7 4340 sbinfo->full_inums = ctx->full_inums;
f3235626
DH
4341 sbinfo->mode = ctx->mode;
4342 sbinfo->huge = ctx->huge;
4343 sbinfo->mpol = ctx->mpol;
4344 ctx->mpol = NULL;
1da177e4 4345
bf11b9a8 4346 raw_spin_lock_init(&sbinfo->stat_lock);
908c7f19 4347 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
602586a8 4348 goto failed;
779750d2
KS
4349 spin_lock_init(&sbinfo->shrinklist_lock);
4350 INIT_LIST_HEAD(&sbinfo->shrinklist);
0edd73b3 4351
285b2c4f 4352 sb->s_maxbytes = MAX_LFS_FILESIZE;
09cbfeaf
KS
4353 sb->s_blocksize = PAGE_SIZE;
4354 sb->s_blocksize_bits = PAGE_SHIFT;
1da177e4
LT
4355 sb->s_magic = TMPFS_MAGIC;
4356 sb->s_op = &shmem_ops;
cfd95a9c 4357 sb->s_time_gran = 1;
b09e0fa4 4358#ifdef CONFIG_TMPFS_XATTR
39f0247d 4359 sb->s_xattr = shmem_xattr_handlers;
b09e0fa4
EP
4360#endif
4361#ifdef CONFIG_TMPFS_POSIX_ACL
1751e8a6 4362 sb->s_flags |= SB_POSIXACL;
39f0247d 4363#endif
2b4db796 4364 uuid_gen(&sb->s_uuid);
0edd73b3 4365
e09764cf
CM
4366#ifdef CONFIG_TMPFS_QUOTA
4367 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4368 sb->dq_op = &shmem_quota_operations;
4369 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4370 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4371
de4c0e7c
LC
4372 /* Copy the default limits from ctx into sbinfo */
4373 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4374 sizeof(struct shmem_quota_limits));
4375
e09764cf
CM
4376 if (shmem_enable_quotas(sb, ctx->quota_types))
4377 goto failed;
4378 }
4379#endif /* CONFIG_TMPFS_QUOTA */
4380
9be7d5b0
HD
4381 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL,
4382 S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
71480663
CM
4383 if (IS_ERR(inode)) {
4384 error = PTR_ERR(inode);
1da177e4 4385 goto failed;
71480663 4386 }
680d794b 4387 inode->i_uid = sbinfo->uid;
4388 inode->i_gid = sbinfo->gid;
318ceed0
AV
4389 sb->s_root = d_make_root(inode);
4390 if (!sb->s_root)
48fde701 4391 goto failed;
1da177e4
LT
4392 return 0;
4393
1da177e4
LT
4394failed:
4395 shmem_put_super(sb);
71480663 4396 return error;
1da177e4
LT
4397}
4398
f3235626
DH
4399static int shmem_get_tree(struct fs_context *fc)
4400{
4401 return get_tree_nodev(fc, shmem_fill_super);
4402}
4403
4404static void shmem_free_fc(struct fs_context *fc)
4405{
4406 struct shmem_options *ctx = fc->fs_private;
4407
4408 if (ctx) {
4409 mpol_put(ctx->mpol);
4410 kfree(ctx);
4411 }
4412}
4413
4414static const struct fs_context_operations shmem_fs_context_ops = {
4415 .free = shmem_free_fc,
4416 .get_tree = shmem_get_tree,
4417#ifdef CONFIG_TMPFS
4418 .parse_monolithic = shmem_parse_options,
4419 .parse_param = shmem_parse_one,
4420 .reconfigure = shmem_reconfigure,
4421#endif
4422};
4423
68279f9c 4424static struct kmem_cache *shmem_inode_cachep __ro_after_init;
1da177e4
LT
4425
4426static struct inode *shmem_alloc_inode(struct super_block *sb)
4427{
41ffe5d5 4428 struct shmem_inode_info *info;
fd60b288 4429 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
41ffe5d5 4430 if (!info)
1da177e4 4431 return NULL;
41ffe5d5 4432 return &info->vfs_inode;
1da177e4
LT
4433}
4434
74b1da56 4435static void shmem_free_in_core_inode(struct inode *inode)
fa0d7e3d 4436{
84e710da
AV
4437 if (S_ISLNK(inode->i_mode))
4438 kfree(inode->i_link);
fa0d7e3d
NP
4439 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4440}
4441
1da177e4
LT
4442static void shmem_destroy_inode(struct inode *inode)
4443{
09208d15 4444 if (S_ISREG(inode->i_mode))
1da177e4 4445 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
a2e45955
CL
4446 if (S_ISDIR(inode->i_mode))
4447 simple_offset_destroy(shmem_get_offset_ctx(inode));
1da177e4
LT
4448}
4449
41ffe5d5 4450static void shmem_init_inode(void *foo)
1da177e4 4451{
41ffe5d5
HD
4452 struct shmem_inode_info *info = foo;
4453 inode_init_once(&info->vfs_inode);
1da177e4
LT
4454}
4455
68279f9c 4456static void __init shmem_init_inodecache(void)
1da177e4
LT
4457{
4458 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4459 sizeof(struct shmem_inode_info),
5d097056 4460 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
1da177e4
LT
4461}
4462
68279f9c 4463static void __init shmem_destroy_inodecache(void)
1da177e4 4464{
1a1d92c1 4465 kmem_cache_destroy(shmem_inode_cachep);
1da177e4
LT
4466}
4467
a7605426 4468/* Keep the page in page cache instead of truncating it */
af7628d6
MWO
4469static int shmem_error_remove_folio(struct address_space *mapping,
4470 struct folio *folio)
a7605426
YS
4471{
4472 return 0;
4473}
4474
aefacb20 4475static const struct address_space_operations shmem_aops = {
1da177e4 4476 .writepage = shmem_writepage,
46de8b97 4477 .dirty_folio = noop_dirty_folio,
1da177e4 4478#ifdef CONFIG_TMPFS
800d15a5
NP
4479 .write_begin = shmem_write_begin,
4480 .write_end = shmem_write_end,
1da177e4 4481#endif
1c93923c 4482#ifdef CONFIG_MIGRATION
54184650 4483 .migrate_folio = migrate_folio,
1c93923c 4484#endif
af7628d6 4485 .error_remove_folio = shmem_error_remove_folio,
1da177e4
LT
4486};
4487
15ad7cdc 4488static const struct file_operations shmem_file_operations = {
1da177e4 4489 .mmap = shmem_mmap,
e88e0d36 4490 .open = shmem_file_open,
c01d5b30 4491 .get_unmapped_area = shmem_get_unmapped_area,
1da177e4 4492#ifdef CONFIG_TMPFS
220f2ac9 4493 .llseek = shmem_file_llseek,
2ba5bbed 4494 .read_iter = shmem_file_read_iter,
e88e0d36 4495 .write_iter = shmem_file_write_iter,
1b061d92 4496 .fsync = noop_fsync,
bd194b18 4497 .splice_read = shmem_file_splice_read,
f6cb85d0 4498 .splice_write = iter_file_splice_write,
83e4fa9c 4499 .fallocate = shmem_fallocate,
1da177e4
LT
4500#endif
4501};
4502
92e1d5be 4503static const struct inode_operations shmem_inode_operations = {
44a30220 4504 .getattr = shmem_getattr,
94c1e62d 4505 .setattr = shmem_setattr,
b09e0fa4 4506#ifdef CONFIG_TMPFS_XATTR
b09e0fa4 4507 .listxattr = shmem_listxattr,
feda821e 4508 .set_acl = simple_set_acl,
e408e695
TT
4509 .fileattr_get = shmem_fileattr_get,
4510 .fileattr_set = shmem_fileattr_set,
b09e0fa4 4511#endif
1da177e4
LT
4512};
4513
92e1d5be 4514static const struct inode_operations shmem_dir_inode_operations = {
1da177e4 4515#ifdef CONFIG_TMPFS
f7cd16a5 4516 .getattr = shmem_getattr,
1da177e4
LT
4517 .create = shmem_create,
4518 .lookup = simple_lookup,
4519 .link = shmem_link,
4520 .unlink = shmem_unlink,
4521 .symlink = shmem_symlink,
4522 .mkdir = shmem_mkdir,
4523 .rmdir = shmem_rmdir,
4524 .mknod = shmem_mknod,
2773bf00 4525 .rename = shmem_rename2,
60545d0d 4526 .tmpfile = shmem_tmpfile,
a2e45955 4527 .get_offset_ctx = shmem_get_offset_ctx,
1da177e4 4528#endif
b09e0fa4 4529#ifdef CONFIG_TMPFS_XATTR
b09e0fa4 4530 .listxattr = shmem_listxattr,
e408e695
TT
4531 .fileattr_get = shmem_fileattr_get,
4532 .fileattr_set = shmem_fileattr_set,
b09e0fa4 4533#endif
39f0247d 4534#ifdef CONFIG_TMPFS_POSIX_ACL
94c1e62d 4535 .setattr = shmem_setattr,
feda821e 4536 .set_acl = simple_set_acl,
39f0247d
AG
4537#endif
4538};
4539
92e1d5be 4540static const struct inode_operations shmem_special_inode_operations = {
f7cd16a5 4541 .getattr = shmem_getattr,
b09e0fa4 4542#ifdef CONFIG_TMPFS_XATTR
b09e0fa4 4543 .listxattr = shmem_listxattr,
b09e0fa4 4544#endif
39f0247d 4545#ifdef CONFIG_TMPFS_POSIX_ACL
94c1e62d 4546 .setattr = shmem_setattr,
feda821e 4547 .set_acl = simple_set_acl,
39f0247d 4548#endif
1da177e4
LT
4549};
4550
759b9775 4551static const struct super_operations shmem_ops = {
1da177e4 4552 .alloc_inode = shmem_alloc_inode,
74b1da56 4553 .free_inode = shmem_free_in_core_inode,
1da177e4
LT
4554 .destroy_inode = shmem_destroy_inode,
4555#ifdef CONFIG_TMPFS
4556 .statfs = shmem_statfs,
680d794b 4557 .show_options = shmem_show_options,
e09764cf
CM
4558#endif
4559#ifdef CONFIG_TMPFS_QUOTA
4560 .get_dquots = shmem_get_dquots,
1da177e4 4561#endif
1f895f75 4562 .evict_inode = shmem_evict_inode,
1da177e4
LT
4563 .drop_inode = generic_delete_inode,
4564 .put_super = shmem_put_super,
396bcc52 4565#ifdef CONFIG_TRANSPARENT_HUGEPAGE
779750d2
KS
4566 .nr_cached_objects = shmem_unused_huge_count,
4567 .free_cached_objects = shmem_unused_huge_scan,
4568#endif
1da177e4
LT
4569};
4570
f0f37e2f 4571static const struct vm_operations_struct shmem_vm_ops = {
54cb8821 4572 .fault = shmem_fault,
d7c17551 4573 .map_pages = filemap_map_pages,
1da177e4
LT
4574#ifdef CONFIG_NUMA
4575 .set_policy = shmem_set_policy,
4576 .get_policy = shmem_get_policy,
4577#endif
4578};
4579
d09e8ca6
PT
4580static const struct vm_operations_struct shmem_anon_vm_ops = {
4581 .fault = shmem_fault,
4582 .map_pages = filemap_map_pages,
4583#ifdef CONFIG_NUMA
4584 .set_policy = shmem_set_policy,
4585 .get_policy = shmem_get_policy,
4586#endif
4587};
4588
f3235626 4589int shmem_init_fs_context(struct fs_context *fc)
1da177e4 4590{
f3235626
DH
4591 struct shmem_options *ctx;
4592
4593 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4594 if (!ctx)
4595 return -ENOMEM;
4596
4597 ctx->mode = 0777 | S_ISVTX;
4598 ctx->uid = current_fsuid();
4599 ctx->gid = current_fsgid();
4600
4601 fc->fs_private = ctx;
4602 fc->ops = &shmem_fs_context_ops;
4603 return 0;
1da177e4
LT
4604}
4605
41ffe5d5 4606static struct file_system_type shmem_fs_type = {
1da177e4
LT
4607 .owner = THIS_MODULE,
4608 .name = "tmpfs",
f3235626
DH
4609 .init_fs_context = shmem_init_fs_context,
4610#ifdef CONFIG_TMPFS
d7167b14 4611 .parameters = shmem_fs_parameters,
f3235626 4612#endif
1da177e4 4613 .kill_sb = kill_litter_super,
db58b5ee 4614 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
1da177e4 4615};
1da177e4 4616
9096bbe9 4617void __init shmem_init(void)
1da177e4
LT
4618{
4619 int error;
4620
9a8ec03e 4621 shmem_init_inodecache();
1da177e4 4622
e09764cf
CM
4623#ifdef CONFIG_TMPFS_QUOTA
4624 error = register_quota_format(&shmem_quota_format);
4625 if (error < 0) {
4626 pr_err("Could not register quota format\n");
4627 goto out3;
4628 }
4629#endif
4630
41ffe5d5 4631 error = register_filesystem(&shmem_fs_type);
1da177e4 4632 if (error) {
1170532b 4633 pr_err("Could not register tmpfs\n");
1da177e4
LT
4634 goto out2;
4635 }
95dc112a 4636
ca4e0519 4637 shm_mnt = kern_mount(&shmem_fs_type);
1da177e4
LT
4638 if (IS_ERR(shm_mnt)) {
4639 error = PTR_ERR(shm_mnt);
1170532b 4640 pr_err("Could not kern_mount tmpfs\n");
1da177e4
LT
4641 goto out1;
4642 }
5a6e75f8 4643
396bcc52 4644#ifdef CONFIG_TRANSPARENT_HUGEPAGE
435c0b87 4645 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
5a6e75f8
KS
4646 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4647 else
5e6e5a12 4648 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
5a6e75f8 4649#endif
9096bbe9 4650 return;
1da177e4
LT
4651
4652out1:
41ffe5d5 4653 unregister_filesystem(&shmem_fs_type);
1da177e4 4654out2:
e09764cf
CM
4655#ifdef CONFIG_TMPFS_QUOTA
4656 unregister_quota_format(&shmem_quota_format);
4657out3:
4658#endif
41ffe5d5 4659 shmem_destroy_inodecache();
1da177e4 4660 shm_mnt = ERR_PTR(error);
1da177e4 4661}
853ac43a 4662
396bcc52 4663#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
5a6e75f8 4664static ssize_t shmem_enabled_show(struct kobject *kobj,
79d4d38a 4665 struct kobj_attribute *attr, char *buf)
5a6e75f8 4666{
26083eb6 4667 static const int values[] = {
5a6e75f8
KS
4668 SHMEM_HUGE_ALWAYS,
4669 SHMEM_HUGE_WITHIN_SIZE,
4670 SHMEM_HUGE_ADVISE,
4671 SHMEM_HUGE_NEVER,
4672 SHMEM_HUGE_DENY,
4673 SHMEM_HUGE_FORCE,
4674 };
79d4d38a
JP
4675 int len = 0;
4676 int i;
5a6e75f8 4677
79d4d38a
JP
4678 for (i = 0; i < ARRAY_SIZE(values); i++) {
4679 len += sysfs_emit_at(buf, len,
9be7d5b0
HD
4680 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4681 i ? " " : "", shmem_format_huge(values[i]));
5a6e75f8 4682 }
79d4d38a
JP
4683 len += sysfs_emit_at(buf, len, "\n");
4684
4685 return len;
5a6e75f8
KS
4686}
4687
4688static ssize_t shmem_enabled_store(struct kobject *kobj,
4689 struct kobj_attribute *attr, const char *buf, size_t count)
4690{
4691 char tmp[16];
4692 int huge;
4693
4694 if (count + 1 > sizeof(tmp))
4695 return -EINVAL;
4696 memcpy(tmp, buf, count);
4697 tmp[count] = '\0';
4698 if (count && tmp[count - 1] == '\n')
4699 tmp[count - 1] = '\0';
4700
4701 huge = shmem_parse_huge(tmp);
4702 if (huge == -EINVAL)
4703 return -EINVAL;
4704 if (!has_transparent_hugepage() &&
4705 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4706 return -EINVAL;
4707
4708 shmem_huge = huge;
435c0b87 4709 if (shmem_huge > SHMEM_HUGE_DENY)
5a6e75f8
KS
4710 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4711 return count;
4712}
4713
4bfa8ada 4714struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
396bcc52 4715#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
f3f0e1d2 4716
853ac43a
MM
4717#else /* !CONFIG_SHMEM */
4718
4719/*
4720 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4721 *
4722 * This is intended for small system where the benefits of the full
4723 * shmem code (swap-backed and resource-limited) are outweighed by
4724 * their complexity. On systems without swap this code should be
4725 * effectively equivalent, but much lighter weight.
4726 */
4727
41ffe5d5 4728static struct file_system_type shmem_fs_type = {
853ac43a 4729 .name = "tmpfs",
f3235626 4730 .init_fs_context = ramfs_init_fs_context,
d7167b14 4731 .parameters = ramfs_fs_parameters,
36ce9d76 4732 .kill_sb = ramfs_kill_sb,
2b8576cb 4733 .fs_flags = FS_USERNS_MOUNT,
853ac43a
MM
4734};
4735
9096bbe9 4736void __init shmem_init(void)
853ac43a 4737{
41ffe5d5 4738 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
853ac43a 4739
41ffe5d5 4740 shm_mnt = kern_mount(&shmem_fs_type);
853ac43a 4741 BUG_ON(IS_ERR(shm_mnt));
853ac43a
MM
4742}
4743
10a9c496 4744int shmem_unuse(unsigned int type)
853ac43a
MM
4745{
4746 return 0;
4747}
4748
d7c9e99a 4749int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
3f96b79a
HD
4750{
4751 return 0;
4752}
4753
24513264
HD
4754void shmem_unlock_mapping(struct address_space *mapping)
4755{
4756}
4757
c01d5b30
HD
4758#ifdef CONFIG_MMU
4759unsigned long shmem_get_unmapped_area(struct file *file,
4760 unsigned long addr, unsigned long len,
4761 unsigned long pgoff, unsigned long flags)
4762{
4763 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4764}
4765#endif
4766
41ffe5d5 4767void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
94c1e62d 4768{
41ffe5d5 4769 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
94c1e62d
HD
4770}
4771EXPORT_SYMBOL_GPL(shmem_truncate_range);
4772
0b0a0806 4773#define shmem_vm_ops generic_file_vm_ops
d09e8ca6 4774#define shmem_anon_vm_ops generic_file_vm_ops
0b0a0806 4775#define shmem_file_operations ramfs_file_operations
0b0a0806
HD
4776#define shmem_acct_size(flags, size) 0
4777#define shmem_unacct_size(flags, size) do {} while (0)
853ac43a 4778
9be7d5b0
HD
4779static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
4780 struct super_block *sb, struct inode *dir,
4781 umode_t mode, dev_t dev, unsigned long flags)
71480663
CM
4782{
4783 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
4784 return inode ? inode : ERR_PTR(-ENOSPC);
4785}
4786
853ac43a
MM
4787#endif /* CONFIG_SHMEM */
4788
4789/* common code */
1da177e4 4790
9be7d5b0
HD
4791static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name,
4792 loff_t size, unsigned long flags, unsigned int i_flags)
1da177e4 4793{
1da177e4 4794 struct inode *inode;
93dec2da 4795 struct file *res;
1da177e4 4796
703321b6
MA
4797 if (IS_ERR(mnt))
4798 return ERR_CAST(mnt);
1da177e4 4799
285b2c4f 4800 if (size < 0 || size > MAX_LFS_FILESIZE)
1da177e4
LT
4801 return ERR_PTR(-EINVAL);
4802
4803 if (shmem_acct_size(flags, size))
4804 return ERR_PTR(-ENOMEM);
4805
7a80e5b8
GS
4806 if (is_idmapped_mnt(mnt))
4807 return ERR_PTR(-EINVAL);
4808
4809 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4810 S_IFREG | S_IRWXUGO, 0, flags);
71480663 4811 if (IS_ERR(inode)) {
dac2d1f6 4812 shmem_unacct_size(flags, size);
71480663 4813 return ERR_CAST(inode);
dac2d1f6 4814 }
c7277090 4815 inode->i_flags |= i_flags;
1da177e4 4816 inode->i_size = size;
6d6b77f1 4817 clear_nlink(inode); /* It is unlinked */
26567cdb 4818 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
93dec2da
AV
4819 if (!IS_ERR(res))
4820 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4821 &shmem_file_operations);
26567cdb 4822 if (IS_ERR(res))
93dec2da 4823 iput(inode);
6b4d0b27 4824 return res;
1da177e4 4825}
c7277090
EP
4826
4827/**
4828 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4829 * kernel internal. There will be NO LSM permission checks against the
4830 * underlying inode. So users of this interface must do LSM checks at a
e1832f29
SS
4831 * higher layer. The users are the big_key and shm implementations. LSM
4832 * checks are provided at the key or shm level rather than the inode.
c7277090
EP
4833 * @name: name for dentry (to be seen in /proc/<pid>/maps
4834 * @size: size to be set for the file
4835 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4836 */
4837struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4838{
703321b6 4839 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
c7277090
EP
4840}
4841
4842/**
4843 * shmem_file_setup - get an unlinked file living in tmpfs
4844 * @name: name for dentry (to be seen in /proc/<pid>/maps
4845 * @size: size to be set for the file
4846 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4847 */
4848struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4849{
703321b6 4850 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
c7277090 4851}
395e0ddc 4852EXPORT_SYMBOL_GPL(shmem_file_setup);
1da177e4 4853
703321b6
MA
4854/**
4855 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4856 * @mnt: the tmpfs mount where the file will be created
4857 * @name: name for dentry (to be seen in /proc/<pid>/maps
4858 * @size: size to be set for the file
4859 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4860 */
4861struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4862 loff_t size, unsigned long flags)
4863{
4864 return __shmem_file_setup(mnt, name, size, flags, 0);
4865}
4866EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4867
46711810 4868/**
1da177e4 4869 * shmem_zero_setup - setup a shared anonymous mapping
45e55300 4870 * @vma: the vma to be mmapped is prepared by do_mmap
1da177e4
LT
4871 */
4872int shmem_zero_setup(struct vm_area_struct *vma)
4873{
4874 struct file *file;
4875 loff_t size = vma->vm_end - vma->vm_start;
4876
66fc1303 4877 /*
c1e8d7c6 4878 * Cloning a new file under mmap_lock leads to a lock ordering conflict
66fc1303
HD
4879 * between XFS directory reading and selinux: since this file is only
4880 * accessible to the user through its mapping, use S_PRIVATE flag to
4881 * bypass file security, in the same way as shmem_kernel_file_setup().
4882 */
703321b6 4883 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
1da177e4
LT
4884 if (IS_ERR(file))
4885 return PTR_ERR(file);
4886
4887 if (vma->vm_file)
4888 fput(vma->vm_file);
4889 vma->vm_file = file;
d09e8ca6 4890 vma->vm_ops = &shmem_anon_vm_ops;
f3f0e1d2 4891
1da177e4
LT
4892 return 0;
4893}
d9d90e5e
HD
4894
4895/**
f01b2b3e
MWO
4896 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4897 * @mapping: the folio's address_space
4898 * @index: the folio index
d9d90e5e
HD
4899 * @gfp: the page allocator flags to use if allocating
4900 *
4901 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4902 * with any new page allocations done using the specified allocation flags.
7e0a1265 4903 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
d9d90e5e
HD
4904 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4905 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4906 *
68da9f05
HD
4907 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4908 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
d9d90e5e 4909 */
f01b2b3e
MWO
4910struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4911 pgoff_t index, gfp_t gfp)
d9d90e5e 4912{
68da9f05
HD
4913#ifdef CONFIG_SHMEM
4914 struct inode *inode = mapping->host;
a3a9c397 4915 struct folio *folio;
68da9f05
HD
4916 int error;
4917
30e6a51d 4918 BUG_ON(!shmem_mapping(mapping));
a3a9c397 4919 error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
e3e1a506 4920 gfp, NULL, NULL);
68da9f05 4921 if (error)
a7605426
YS
4922 return ERR_PTR(error);
4923
a3a9c397 4924 folio_unlock(folio);
f01b2b3e
MWO
4925 return folio;
4926#else
4927 /*
4928 * The tiny !SHMEM case uses ramfs without swap
4929 */
4930 return mapping_read_folio_gfp(mapping, index, gfp);
4931#endif
4932}
4933EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4934
4935struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4936 pgoff_t index, gfp_t gfp)
4937{
4938 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4939 struct page *page;
4940
4941 if (IS_ERR(folio))
4942 return &folio->page;
4943
a3a9c397 4944 page = folio_file_page(folio, index);
a7605426 4945 if (PageHWPoison(page)) {
a3a9c397 4946 folio_put(folio);
a7605426
YS
4947 return ERR_PTR(-EIO);
4948 }
4949
68da9f05 4950 return page;
d9d90e5e
HD
4951}
4952EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);