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.
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
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 *
18 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
21 * This file is released under the GPL.
22 */
23
24 #include <linux/fs.h>
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
31 #include <linux/fileattr.h>
32 #include <linux/mm.h>
33 #include <linux/random.h>
34 #include <linux/sched/signal.h>
35 #include <linux/export.h>
36 #include <linux/shmem_fs.h>
37 #include <linux/swap.h>
38 #include <linux/uio.h>
39 #include <linux/hugetlb.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
42 #include <linux/iversion.h>
43 #include "swap.h"
44
45 static struct vfsmount *shm_mnt;
46
47 #ifdef CONFIG_SHMEM
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
54 #include <linux/xattr.h>
55 #include <linux/exportfs.h>
56 #include <linux/posix_acl.h>
57 #include <linux/posix_acl_xattr.h>
58 #include <linux/mman.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/backing-dev.h>
62 #include <linux/writeback.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
81 #include <linux/quotaops.h>
82
83 #include <linux/uaccess.h>
84
85 #include "internal.h"
86
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
92
93 /* Pretend that one inode + its dentry occupy this much memory */
94 #define BOGO_INODE_SIZE 1024
95
96 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
97 #define SHORT_SYMLINK_LEN 128
98
99 /*
100 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
101 * inode->i_private (with i_rwsem making sure that it has only one user at
102 * a time): we would prefer not to enlarge the shmem inode just for that.
103 */
104 struct shmem_falloc {
105 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
106 pgoff_t start; /* start of range currently being fallocated */
107 pgoff_t next; /* the next page offset to be fallocated */
108 pgoff_t nr_falloced; /* how many new pages have been fallocated */
109 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
110 };
111
112 struct shmem_options {
113 unsigned long long blocks;
114 unsigned long long inodes;
115 struct mempolicy *mpol;
116 kuid_t uid;
117 kgid_t gid;
118 umode_t mode;
119 bool full_inums;
120 int huge;
121 int seen;
122 bool noswap;
123 unsigned short quota_types;
124 struct shmem_quota_limits qlimits;
125 #define SHMEM_SEEN_BLOCKS 1
126 #define SHMEM_SEEN_INODES 2
127 #define SHMEM_SEEN_HUGE 4
128 #define SHMEM_SEEN_INUMS 8
129 #define SHMEM_SEEN_NOSWAP 16
130 #define SHMEM_SEEN_QUOTA 32
131 };
132
133 #ifdef CONFIG_TMPFS
shmem_default_max_blocks(void)134 static unsigned long shmem_default_max_blocks(void)
135 {
136 return totalram_pages() / 2;
137 }
138
shmem_default_max_inodes(void)139 static unsigned long shmem_default_max_inodes(void)
140 {
141 unsigned long nr_pages = totalram_pages();
142
143 return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
144 ULONG_MAX / BOGO_INODE_SIZE);
145 }
146 #endif
147
148 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
149 struct folio **foliop, enum sgp_type sgp,
150 gfp_t gfp, struct vm_area_struct *vma,
151 vm_fault_t *fault_type);
152
SHMEM_SB(struct super_block * sb)153 static 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 */
shmem_acct_size(unsigned long flags,loff_t size)164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
165 {
166 return (flags & VM_NORESERVE) ?
167 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
168 }
169
shmem_unacct_size(unsigned long flags,loff_t size)170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
171 {
172 if (!(flags & VM_NORESERVE))
173 vm_unacct_memory(VM_ACCT(size));
174 }
175
shmem_reacct_size(unsigned long flags,loff_t oldsize,loff_t newsize)176 static 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
189 /*
190 * ... whereas tmpfs objects are accounted incrementally as
191 * pages are allocated, in order to allow large sparse files.
192 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
194 */
shmem_acct_block(unsigned long flags,long pages)195 static inline int shmem_acct_block(unsigned long flags, long pages)
196 {
197 if (!(flags & VM_NORESERVE))
198 return 0;
199
200 return security_vm_enough_memory_mm(current->mm,
201 pages * VM_ACCT(PAGE_SIZE));
202 }
203
shmem_unacct_blocks(unsigned long flags,long pages)204 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
205 {
206 if (flags & VM_NORESERVE)
207 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
208 }
209
shmem_inode_acct_block(struct inode * inode,long pages)210 static int shmem_inode_acct_block(struct inode *inode, long pages)
211 {
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
214 int err = -ENOSPC;
215
216 if (shmem_acct_block(info->flags, pages))
217 return err;
218
219 might_sleep(); /* when quotas */
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
223 goto unacct;
224
225 err = dquot_alloc_block_nodirty(inode, pages);
226 if (err)
227 goto unacct;
228
229 percpu_counter_add(&sbinfo->used_blocks, pages);
230 } else {
231 err = dquot_alloc_block_nodirty(inode, pages);
232 if (err)
233 goto unacct;
234 }
235
236 return 0;
237
238 unacct:
239 shmem_unacct_blocks(info->flags, pages);
240 return err;
241 }
242
shmem_inode_unacct_blocks(struct inode * inode,long pages)243 static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
244 {
245 struct shmem_inode_info *info = SHMEM_I(inode);
246 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
247
248 might_sleep(); /* when quotas */
249 dquot_free_block_nodirty(inode, pages);
250
251 if (sbinfo->max_blocks)
252 percpu_counter_sub(&sbinfo->used_blocks, pages);
253 shmem_unacct_blocks(info->flags, pages);
254 }
255
256 static const struct super_operations shmem_ops;
257 const struct address_space_operations shmem_aops;
258 static const struct file_operations shmem_file_operations;
259 static const struct inode_operations shmem_inode_operations;
260 static const struct inode_operations shmem_dir_inode_operations;
261 static const struct inode_operations shmem_special_inode_operations;
262 static const struct vm_operations_struct shmem_vm_ops;
263 static const struct vm_operations_struct shmem_anon_vm_ops;
264 static struct file_system_type shmem_fs_type;
265
vma_is_anon_shmem(struct vm_area_struct * vma)266 bool vma_is_anon_shmem(struct vm_area_struct *vma)
267 {
268 return vma->vm_ops == &shmem_anon_vm_ops;
269 }
270
vma_is_shmem(struct vm_area_struct * vma)271 bool vma_is_shmem(struct vm_area_struct *vma)
272 {
273 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
274 }
275
276 static LIST_HEAD(shmem_swaplist);
277 static DEFINE_MUTEX(shmem_swaplist_mutex);
278
279 #ifdef CONFIG_TMPFS_QUOTA
280
shmem_enable_quotas(struct super_block * sb,unsigned short quota_types)281 static int shmem_enable_quotas(struct super_block *sb,
282 unsigned short quota_types)
283 {
284 int type, err = 0;
285
286 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
287 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
288 if (!(quota_types & (1 << type)))
289 continue;
290 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
291 DQUOT_USAGE_ENABLED |
292 DQUOT_LIMITS_ENABLED);
293 if (err)
294 goto out_err;
295 }
296 return 0;
297
298 out_err:
299 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
300 type, err);
301 for (type--; type >= 0; type--)
302 dquot_quota_off(sb, type);
303 return err;
304 }
305
shmem_disable_quotas(struct super_block * sb)306 static void shmem_disable_quotas(struct super_block *sb)
307 {
308 int type;
309
310 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
311 dquot_quota_off(sb, type);
312 }
313
shmem_get_dquots(struct inode * inode)314 static struct dquot __rcu **shmem_get_dquots(struct inode *inode)
315 {
316 return SHMEM_I(inode)->i_dquot;
317 }
318 #endif /* CONFIG_TMPFS_QUOTA */
319
320 /*
321 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
322 * produces a novel ino for the newly allocated inode.
323 *
324 * It may also be called when making a hard link to permit the space needed by
325 * each dentry. However, in that case, no new inode number is needed since that
326 * internally draws from another pool of inode numbers (currently global
327 * get_next_ino()). This case is indicated by passing NULL as inop.
328 */
329 #define SHMEM_INO_BATCH 1024
shmem_reserve_inode(struct super_block * sb,ino_t * inop)330 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
331 {
332 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
333 ino_t ino;
334
335 if (!(sb->s_flags & SB_KERNMOUNT)) {
336 raw_spin_lock(&sbinfo->stat_lock);
337 if (sbinfo->max_inodes) {
338 if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
339 raw_spin_unlock(&sbinfo->stat_lock);
340 return -ENOSPC;
341 }
342 sbinfo->free_ispace -= BOGO_INODE_SIZE;
343 }
344 if (inop) {
345 ino = sbinfo->next_ino++;
346 if (unlikely(is_zero_ino(ino)))
347 ino = sbinfo->next_ino++;
348 if (unlikely(!sbinfo->full_inums &&
349 ino > UINT_MAX)) {
350 /*
351 * Emulate get_next_ino uint wraparound for
352 * compatibility
353 */
354 if (IS_ENABLED(CONFIG_64BIT))
355 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
356 __func__, MINOR(sb->s_dev));
357 sbinfo->next_ino = 1;
358 ino = sbinfo->next_ino++;
359 }
360 *inop = ino;
361 }
362 raw_spin_unlock(&sbinfo->stat_lock);
363 } else if (inop) {
364 /*
365 * __shmem_file_setup, one of our callers, is lock-free: it
366 * doesn't hold stat_lock in shmem_reserve_inode since
367 * max_inodes is always 0, and is called from potentially
368 * unknown contexts. As such, use a per-cpu batched allocator
369 * which doesn't require the per-sb stat_lock unless we are at
370 * the batch boundary.
371 *
372 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
373 * shmem mounts are not exposed to userspace, so we don't need
374 * to worry about things like glibc compatibility.
375 */
376 ino_t *next_ino;
377
378 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
379 ino = *next_ino;
380 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
381 raw_spin_lock(&sbinfo->stat_lock);
382 ino = sbinfo->next_ino;
383 sbinfo->next_ino += SHMEM_INO_BATCH;
384 raw_spin_unlock(&sbinfo->stat_lock);
385 if (unlikely(is_zero_ino(ino)))
386 ino++;
387 }
388 *inop = ino;
389 *next_ino = ++ino;
390 put_cpu();
391 }
392
393 return 0;
394 }
395
shmem_free_inode(struct super_block * sb,size_t freed_ispace)396 static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
397 {
398 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
399 if (sbinfo->max_inodes) {
400 raw_spin_lock(&sbinfo->stat_lock);
401 sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
402 raw_spin_unlock(&sbinfo->stat_lock);
403 }
404 }
405
406 /**
407 * shmem_recalc_inode - recalculate the block usage of an inode
408 * @inode: inode to recalc
409 * @alloced: the change in number of pages allocated to inode
410 * @swapped: the change in number of pages swapped from inode
411 *
412 * We have to calculate the free blocks since the mm can drop
413 * undirtied hole pages behind our back.
414 *
415 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
416 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
417 */
shmem_recalc_inode(struct inode * inode,long alloced,long swapped)418 static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
419 {
420 struct shmem_inode_info *info = SHMEM_I(inode);
421 long freed;
422
423 spin_lock(&info->lock);
424 info->alloced += alloced;
425 info->swapped += swapped;
426 freed = info->alloced - info->swapped -
427 READ_ONCE(inode->i_mapping->nrpages);
428 /*
429 * Special case: whereas normally shmem_recalc_inode() is called
430 * after i_mapping->nrpages has already been adjusted (up or down),
431 * shmem_writepage() has to raise swapped before nrpages is lowered -
432 * to stop a racing shmem_recalc_inode() from thinking that a page has
433 * been freed. Compensate here, to avoid the need for a followup call.
434 */
435 if (swapped > 0)
436 freed += swapped;
437 if (freed > 0)
438 info->alloced -= freed;
439 spin_unlock(&info->lock);
440
441 /* The quota case may block */
442 if (freed > 0)
443 shmem_inode_unacct_blocks(inode, freed);
444 }
445
shmem_charge(struct inode * inode,long pages)446 bool shmem_charge(struct inode *inode, long pages)
447 {
448 struct address_space *mapping = inode->i_mapping;
449
450 if (shmem_inode_acct_block(inode, pages))
451 return false;
452
453 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
454 xa_lock_irq(&mapping->i_pages);
455 mapping->nrpages += pages;
456 xa_unlock_irq(&mapping->i_pages);
457
458 shmem_recalc_inode(inode, pages, 0);
459 return true;
460 }
461
shmem_uncharge(struct inode * inode,long pages)462 void shmem_uncharge(struct inode *inode, long pages)
463 {
464 /* pages argument is currently unused: keep it to help debugging */
465 /* nrpages adjustment done by __filemap_remove_folio() or caller */
466
467 shmem_recalc_inode(inode, 0, 0);
468 }
469
470 /*
471 * Replace item expected in xarray by a new item, while holding xa_lock.
472 */
shmem_replace_entry(struct address_space * mapping,pgoff_t index,void * expected,void * replacement)473 static int shmem_replace_entry(struct address_space *mapping,
474 pgoff_t index, void *expected, void *replacement)
475 {
476 XA_STATE(xas, &mapping->i_pages, index);
477 void *item;
478
479 VM_BUG_ON(!expected);
480 VM_BUG_ON(!replacement);
481 item = xas_load(&xas);
482 if (item != expected)
483 return -ENOENT;
484 xas_store(&xas, replacement);
485 return 0;
486 }
487
488 /*
489 * Sometimes, before we decide whether to proceed or to fail, we must check
490 * that an entry was not already brought back from swap by a racing thread.
491 *
492 * Checking page is not enough: by the time a SwapCache page is locked, it
493 * might be reused, and again be SwapCache, using the same swap as before.
494 */
shmem_confirm_swap(struct address_space * mapping,pgoff_t index,swp_entry_t swap)495 static bool shmem_confirm_swap(struct address_space *mapping,
496 pgoff_t index, swp_entry_t swap)
497 {
498 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
499 }
500
501 /*
502 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
503 *
504 * SHMEM_HUGE_NEVER:
505 * disables huge pages for the mount;
506 * SHMEM_HUGE_ALWAYS:
507 * enables huge pages for the mount;
508 * SHMEM_HUGE_WITHIN_SIZE:
509 * only allocate huge pages if the page will be fully within i_size,
510 * also respect fadvise()/madvise() hints;
511 * SHMEM_HUGE_ADVISE:
512 * only allocate huge pages if requested with fadvise()/madvise();
513 */
514
515 #define SHMEM_HUGE_NEVER 0
516 #define SHMEM_HUGE_ALWAYS 1
517 #define SHMEM_HUGE_WITHIN_SIZE 2
518 #define SHMEM_HUGE_ADVISE 3
519
520 /*
521 * Special values.
522 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
523 *
524 * SHMEM_HUGE_DENY:
525 * disables huge on shm_mnt and all mounts, for emergency use;
526 * SHMEM_HUGE_FORCE:
527 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
528 *
529 */
530 #define SHMEM_HUGE_DENY (-1)
531 #define SHMEM_HUGE_FORCE (-2)
532
533 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
534 /* ifdef here to avoid bloating shmem.o when not necessary */
535
536 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
537
shmem_is_huge(struct inode * inode,pgoff_t index,bool shmem_huge_force,struct mm_struct * mm,unsigned long vm_flags)538 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
539 struct mm_struct *mm, unsigned long vm_flags)
540 {
541 loff_t i_size;
542
543 if (!S_ISREG(inode->i_mode))
544 return false;
545 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
546 return false;
547 if (shmem_huge == SHMEM_HUGE_DENY)
548 return false;
549 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
550 return true;
551
552 switch (SHMEM_SB(inode->i_sb)->huge) {
553 case SHMEM_HUGE_ALWAYS:
554 return true;
555 case SHMEM_HUGE_WITHIN_SIZE:
556 index = round_up(index + 1, HPAGE_PMD_NR);
557 i_size = round_up(i_size_read(inode), PAGE_SIZE);
558 if (i_size >> PAGE_SHIFT >= index)
559 return true;
560 fallthrough;
561 case SHMEM_HUGE_ADVISE:
562 if (mm && (vm_flags & VM_HUGEPAGE))
563 return true;
564 fallthrough;
565 default:
566 return false;
567 }
568 }
569
570 #if defined(CONFIG_SYSFS)
shmem_parse_huge(const char * str)571 static int shmem_parse_huge(const char *str)
572 {
573 if (!strcmp(str, "never"))
574 return SHMEM_HUGE_NEVER;
575 if (!strcmp(str, "always"))
576 return SHMEM_HUGE_ALWAYS;
577 if (!strcmp(str, "within_size"))
578 return SHMEM_HUGE_WITHIN_SIZE;
579 if (!strcmp(str, "advise"))
580 return SHMEM_HUGE_ADVISE;
581 if (!strcmp(str, "deny"))
582 return SHMEM_HUGE_DENY;
583 if (!strcmp(str, "force"))
584 return SHMEM_HUGE_FORCE;
585 return -EINVAL;
586 }
587 #endif
588
589 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
shmem_format_huge(int huge)590 static const char *shmem_format_huge(int huge)
591 {
592 switch (huge) {
593 case SHMEM_HUGE_NEVER:
594 return "never";
595 case SHMEM_HUGE_ALWAYS:
596 return "always";
597 case SHMEM_HUGE_WITHIN_SIZE:
598 return "within_size";
599 case SHMEM_HUGE_ADVISE:
600 return "advise";
601 case SHMEM_HUGE_DENY:
602 return "deny";
603 case SHMEM_HUGE_FORCE:
604 return "force";
605 default:
606 VM_BUG_ON(1);
607 return "bad_val";
608 }
609 }
610 #endif
611
shmem_unused_huge_shrink(struct shmem_sb_info * sbinfo,struct shrink_control * sc,unsigned long nr_to_split)612 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
613 struct shrink_control *sc, unsigned long nr_to_split)
614 {
615 LIST_HEAD(list), *pos, *next;
616 LIST_HEAD(to_remove);
617 struct inode *inode;
618 struct shmem_inode_info *info;
619 struct folio *folio;
620 unsigned long batch = sc ? sc->nr_to_scan : 128;
621 int split = 0;
622
623 if (list_empty(&sbinfo->shrinklist))
624 return SHRINK_STOP;
625
626 spin_lock(&sbinfo->shrinklist_lock);
627 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
628 info = list_entry(pos, struct shmem_inode_info, shrinklist);
629
630 /* pin the inode */
631 inode = igrab(&info->vfs_inode);
632
633 /* inode is about to be evicted */
634 if (!inode) {
635 list_del_init(&info->shrinklist);
636 goto next;
637 }
638
639 /* Check if there's anything to gain */
640 if (round_up(inode->i_size, PAGE_SIZE) ==
641 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
642 list_move(&info->shrinklist, &to_remove);
643 goto next;
644 }
645
646 list_move(&info->shrinklist, &list);
647 next:
648 sbinfo->shrinklist_len--;
649 if (!--batch)
650 break;
651 }
652 spin_unlock(&sbinfo->shrinklist_lock);
653
654 list_for_each_safe(pos, next, &to_remove) {
655 info = list_entry(pos, struct shmem_inode_info, shrinklist);
656 inode = &info->vfs_inode;
657 list_del_init(&info->shrinklist);
658 iput(inode);
659 }
660
661 list_for_each_safe(pos, next, &list) {
662 int ret;
663 pgoff_t index;
664
665 info = list_entry(pos, struct shmem_inode_info, shrinklist);
666 inode = &info->vfs_inode;
667
668 if (nr_to_split && split >= nr_to_split)
669 goto move_back;
670
671 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
672 folio = filemap_get_folio(inode->i_mapping, index);
673 if (IS_ERR(folio))
674 goto drop;
675
676 /* No huge page at the end of the file: nothing to split */
677 if (!folio_test_large(folio)) {
678 folio_put(folio);
679 goto drop;
680 }
681
682 /*
683 * Move the inode on the list back to shrinklist if we failed
684 * to lock the page at this time.
685 *
686 * Waiting for the lock may lead to deadlock in the
687 * reclaim path.
688 */
689 if (!folio_trylock(folio)) {
690 folio_put(folio);
691 goto move_back;
692 }
693
694 ret = split_folio(folio);
695 folio_unlock(folio);
696 folio_put(folio);
697
698 /* If split failed move the inode on the list back to shrinklist */
699 if (ret)
700 goto move_back;
701
702 split++;
703 drop:
704 list_del_init(&info->shrinklist);
705 goto put;
706 move_back:
707 /*
708 * Make sure the inode is either on the global list or deleted
709 * from any local list before iput() since it could be deleted
710 * in another thread once we put the inode (then the local list
711 * is corrupted).
712 */
713 spin_lock(&sbinfo->shrinklist_lock);
714 list_move(&info->shrinklist, &sbinfo->shrinklist);
715 sbinfo->shrinklist_len++;
716 spin_unlock(&sbinfo->shrinklist_lock);
717 put:
718 iput(inode);
719 }
720
721 return split;
722 }
723
shmem_unused_huge_scan(struct super_block * sb,struct shrink_control * sc)724 static long shmem_unused_huge_scan(struct super_block *sb,
725 struct shrink_control *sc)
726 {
727 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
728
729 if (!READ_ONCE(sbinfo->shrinklist_len))
730 return SHRINK_STOP;
731
732 return shmem_unused_huge_shrink(sbinfo, sc, 0);
733 }
734
shmem_unused_huge_count(struct super_block * sb,struct shrink_control * sc)735 static long shmem_unused_huge_count(struct super_block *sb,
736 struct shrink_control *sc)
737 {
738 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
739 return READ_ONCE(sbinfo->shrinklist_len);
740 }
741 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
742
743 #define shmem_huge SHMEM_HUGE_DENY
744
shmem_unused_huge_shrink(struct shmem_sb_info * sbinfo,struct shrink_control * sc,unsigned long nr_to_split)745 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
746 struct shrink_control *sc, unsigned long nr_to_split)
747 {
748 return 0;
749 }
750 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
751
752 /*
753 * Like filemap_add_folio, but error if expected item has gone.
754 */
shmem_add_to_page_cache(struct folio * folio,struct address_space * mapping,pgoff_t index,void * expected,gfp_t gfp,struct mm_struct * charge_mm)755 static int shmem_add_to_page_cache(struct folio *folio,
756 struct address_space *mapping,
757 pgoff_t index, void *expected, gfp_t gfp,
758 struct mm_struct *charge_mm)
759 {
760 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
761 long nr = folio_nr_pages(folio);
762 int error;
763
764 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
765 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
766 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
767 VM_BUG_ON(expected && folio_test_large(folio));
768
769 folio_ref_add(folio, nr);
770 folio->mapping = mapping;
771 folio->index = index;
772
773 if (!folio_test_swapcache(folio)) {
774 error = mem_cgroup_charge(folio, charge_mm, gfp);
775 if (error) {
776 if (folio_test_pmd_mappable(folio)) {
777 count_vm_event(THP_FILE_FALLBACK);
778 count_vm_event(THP_FILE_FALLBACK_CHARGE);
779 }
780 goto error;
781 }
782 }
783 folio_throttle_swaprate(folio, gfp);
784
785 do {
786 xas_lock_irq(&xas);
787 if (expected != xas_find_conflict(&xas)) {
788 xas_set_err(&xas, -EEXIST);
789 goto unlock;
790 }
791 if (expected && xas_find_conflict(&xas)) {
792 xas_set_err(&xas, -EEXIST);
793 goto unlock;
794 }
795 xas_store(&xas, folio);
796 if (xas_error(&xas))
797 goto unlock;
798 if (folio_test_pmd_mappable(folio)) {
799 count_vm_event(THP_FILE_ALLOC);
800 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
801 }
802 mapping->nrpages += nr;
803 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
804 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
805 unlock:
806 xas_unlock_irq(&xas);
807 } while (xas_nomem(&xas, gfp));
808
809 if (xas_error(&xas)) {
810 error = xas_error(&xas);
811 goto error;
812 }
813
814 return 0;
815 error:
816 folio->mapping = NULL;
817 folio_ref_sub(folio, nr);
818 return error;
819 }
820
821 /*
822 * Like delete_from_page_cache, but substitutes swap for @folio.
823 */
shmem_delete_from_page_cache(struct folio * folio,void * radswap)824 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
825 {
826 struct address_space *mapping = folio->mapping;
827 long nr = folio_nr_pages(folio);
828 int error;
829
830 xa_lock_irq(&mapping->i_pages);
831 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
832 folio->mapping = NULL;
833 mapping->nrpages -= nr;
834 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
835 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
836 xa_unlock_irq(&mapping->i_pages);
837 folio_put(folio);
838 BUG_ON(error);
839 }
840
841 /*
842 * Remove swap entry from page cache, free the swap and its page cache.
843 */
shmem_free_swap(struct address_space * mapping,pgoff_t index,void * radswap)844 static int shmem_free_swap(struct address_space *mapping,
845 pgoff_t index, void *radswap)
846 {
847 void *old;
848
849 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
850 if (old != radswap)
851 return -ENOENT;
852 free_swap_and_cache(radix_to_swp_entry(radswap));
853 return 0;
854 }
855
856 /*
857 * Determine (in bytes) how many of the shmem object's pages mapped by the
858 * given offsets are swapped out.
859 *
860 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
861 * as long as the inode doesn't go away and racy results are not a problem.
862 */
shmem_partial_swap_usage(struct address_space * mapping,pgoff_t start,pgoff_t end)863 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
864 pgoff_t start, pgoff_t end)
865 {
866 XA_STATE(xas, &mapping->i_pages, start);
867 struct page *page;
868 unsigned long swapped = 0;
869 unsigned long max = end - 1;
870
871 rcu_read_lock();
872 xas_for_each(&xas, page, max) {
873 if (xas_retry(&xas, page))
874 continue;
875 if (xa_is_value(page))
876 swapped++;
877 if (xas.xa_index == max)
878 break;
879 if (need_resched()) {
880 xas_pause(&xas);
881 cond_resched_rcu();
882 }
883 }
884
885 rcu_read_unlock();
886
887 return swapped << PAGE_SHIFT;
888 }
889
890 /*
891 * Determine (in bytes) how many of the shmem object's pages mapped by the
892 * given vma is swapped out.
893 *
894 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
895 * as long as the inode doesn't go away and racy results are not a problem.
896 */
shmem_swap_usage(struct vm_area_struct * vma)897 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
898 {
899 struct inode *inode = file_inode(vma->vm_file);
900 struct shmem_inode_info *info = SHMEM_I(inode);
901 struct address_space *mapping = inode->i_mapping;
902 unsigned long swapped;
903
904 /* Be careful as we don't hold info->lock */
905 swapped = READ_ONCE(info->swapped);
906
907 /*
908 * The easier cases are when the shmem object has nothing in swap, or
909 * the vma maps it whole. Then we can simply use the stats that we
910 * already track.
911 */
912 if (!swapped)
913 return 0;
914
915 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
916 return swapped << PAGE_SHIFT;
917
918 /* Here comes the more involved part */
919 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
920 vma->vm_pgoff + vma_pages(vma));
921 }
922
923 /*
924 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
925 */
shmem_unlock_mapping(struct address_space * mapping)926 void shmem_unlock_mapping(struct address_space *mapping)
927 {
928 struct folio_batch fbatch;
929 pgoff_t index = 0;
930
931 folio_batch_init(&fbatch);
932 /*
933 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
934 */
935 while (!mapping_unevictable(mapping) &&
936 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
937 check_move_unevictable_folios(&fbatch);
938 folio_batch_release(&fbatch);
939 cond_resched();
940 }
941 }
942
shmem_get_partial_folio(struct inode * inode,pgoff_t index)943 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
944 {
945 struct folio *folio;
946
947 /*
948 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
949 * beyond i_size, and reports fallocated folios as holes.
950 */
951 folio = filemap_get_entry(inode->i_mapping, index);
952 if (!folio)
953 return folio;
954 if (!xa_is_value(folio)) {
955 folio_lock(folio);
956 if (folio->mapping == inode->i_mapping)
957 return folio;
958 /* The folio has been swapped out */
959 folio_unlock(folio);
960 folio_put(folio);
961 }
962 /*
963 * But read a folio back from swap if any of it is within i_size
964 * (although in some cases this is just a waste of time).
965 */
966 folio = NULL;
967 shmem_get_folio(inode, index, &folio, SGP_READ);
968 return folio;
969 }
970
971 /*
972 * Remove range of pages and swap entries from page cache, and free them.
973 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
974 */
shmem_undo_range(struct inode * inode,loff_t lstart,loff_t lend,bool unfalloc)975 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
976 bool unfalloc)
977 {
978 struct address_space *mapping = inode->i_mapping;
979 struct shmem_inode_info *info = SHMEM_I(inode);
980 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
981 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
982 struct folio_batch fbatch;
983 pgoff_t indices[PAGEVEC_SIZE];
984 struct folio *folio;
985 bool same_folio;
986 long nr_swaps_freed = 0;
987 pgoff_t index;
988 int i;
989
990 if (lend == -1)
991 end = -1; /* unsigned, so actually very big */
992
993 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
994 info->fallocend = start;
995
996 folio_batch_init(&fbatch);
997 index = start;
998 while (index < end && find_lock_entries(mapping, &index, end - 1,
999 &fbatch, indices)) {
1000 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1001 folio = fbatch.folios[i];
1002
1003 if (xa_is_value(folio)) {
1004 if (unfalloc)
1005 continue;
1006 nr_swaps_freed += !shmem_free_swap(mapping,
1007 indices[i], folio);
1008 continue;
1009 }
1010
1011 if (!unfalloc || !folio_test_uptodate(folio))
1012 truncate_inode_folio(mapping, folio);
1013 folio_unlock(folio);
1014 }
1015 folio_batch_remove_exceptionals(&fbatch);
1016 folio_batch_release(&fbatch);
1017 cond_resched();
1018 }
1019
1020 /*
1021 * When undoing a failed fallocate, we want none of the partial folio
1022 * zeroing and splitting below, but shall want to truncate the whole
1023 * folio when !uptodate indicates that it was added by this fallocate,
1024 * even when [lstart, lend] covers only a part of the folio.
1025 */
1026 if (unfalloc)
1027 goto whole_folios;
1028
1029 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1030 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1031 if (folio) {
1032 same_folio = lend < folio_pos(folio) + folio_size(folio);
1033 folio_mark_dirty(folio);
1034 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
1035 start = folio_next_index(folio);
1036 if (same_folio)
1037 end = folio->index;
1038 }
1039 folio_unlock(folio);
1040 folio_put(folio);
1041 folio = NULL;
1042 }
1043
1044 if (!same_folio)
1045 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1046 if (folio) {
1047 folio_mark_dirty(folio);
1048 if (!truncate_inode_partial_folio(folio, lstart, lend))
1049 end = folio->index;
1050 folio_unlock(folio);
1051 folio_put(folio);
1052 }
1053
1054 whole_folios:
1055
1056 index = start;
1057 while (index < end) {
1058 cond_resched();
1059
1060 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1061 indices)) {
1062 /* If all gone or hole-punch or unfalloc, we're done */
1063 if (index == start || end != -1)
1064 break;
1065 /* But if truncating, restart to make sure all gone */
1066 index = start;
1067 continue;
1068 }
1069 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1070 folio = fbatch.folios[i];
1071
1072 if (xa_is_value(folio)) {
1073 if (unfalloc)
1074 continue;
1075 if (shmem_free_swap(mapping, indices[i], folio)) {
1076 /* Swap was replaced by page: retry */
1077 index = indices[i];
1078 break;
1079 }
1080 nr_swaps_freed++;
1081 continue;
1082 }
1083
1084 folio_lock(folio);
1085
1086 if (!unfalloc || !folio_test_uptodate(folio)) {
1087 if (folio_mapping(folio) != mapping) {
1088 /* Page was replaced by swap: retry */
1089 folio_unlock(folio);
1090 index = indices[i];
1091 break;
1092 }
1093 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1094 folio);
1095
1096 if (!folio_test_large(folio)) {
1097 truncate_inode_folio(mapping, folio);
1098 } else if (truncate_inode_partial_folio(folio, lstart, lend)) {
1099 /*
1100 * If we split a page, reset the loop so
1101 * that we pick up the new sub pages.
1102 * Otherwise the THP was entirely
1103 * dropped or the target range was
1104 * zeroed, so just continue the loop as
1105 * is.
1106 */
1107 if (!folio_test_large(folio)) {
1108 folio_unlock(folio);
1109 index = start;
1110 break;
1111 }
1112 }
1113 }
1114 folio_unlock(folio);
1115 }
1116 folio_batch_remove_exceptionals(&fbatch);
1117 folio_batch_release(&fbatch);
1118 }
1119
1120 shmem_recalc_inode(inode, 0, -nr_swaps_freed);
1121 }
1122
shmem_truncate_range(struct inode * inode,loff_t lstart,loff_t lend)1123 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1124 {
1125 shmem_undo_range(inode, lstart, lend, false);
1126 inode->i_mtime = inode_set_ctime_current(inode);
1127 inode_inc_iversion(inode);
1128 }
1129 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1130
shmem_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)1131 static int shmem_getattr(struct mnt_idmap *idmap,
1132 const struct path *path, struct kstat *stat,
1133 u32 request_mask, unsigned int query_flags)
1134 {
1135 struct inode *inode = path->dentry->d_inode;
1136 struct shmem_inode_info *info = SHMEM_I(inode);
1137
1138 if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1139 shmem_recalc_inode(inode, 0, 0);
1140
1141 if (info->fsflags & FS_APPEND_FL)
1142 stat->attributes |= STATX_ATTR_APPEND;
1143 if (info->fsflags & FS_IMMUTABLE_FL)
1144 stat->attributes |= STATX_ATTR_IMMUTABLE;
1145 if (info->fsflags & FS_NODUMP_FL)
1146 stat->attributes |= STATX_ATTR_NODUMP;
1147 stat->attributes_mask |= (STATX_ATTR_APPEND |
1148 STATX_ATTR_IMMUTABLE |
1149 STATX_ATTR_NODUMP);
1150 generic_fillattr(idmap, request_mask, inode, stat);
1151
1152 if (shmem_is_huge(inode, 0, false, NULL, 0))
1153 stat->blksize = HPAGE_PMD_SIZE;
1154
1155 if (request_mask & STATX_BTIME) {
1156 stat->result_mask |= STATX_BTIME;
1157 stat->btime.tv_sec = info->i_crtime.tv_sec;
1158 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1159 }
1160
1161 return 0;
1162 }
1163
shmem_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)1164 static int shmem_setattr(struct mnt_idmap *idmap,
1165 struct dentry *dentry, struct iattr *attr)
1166 {
1167 struct inode *inode = d_inode(dentry);
1168 struct shmem_inode_info *info = SHMEM_I(inode);
1169 int error;
1170 bool update_mtime = false;
1171 bool update_ctime = true;
1172
1173 error = setattr_prepare(idmap, dentry, attr);
1174 if (error)
1175 return error;
1176
1177 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1178 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1179 return -EPERM;
1180 }
1181 }
1182
1183 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1184 loff_t oldsize = inode->i_size;
1185 loff_t newsize = attr->ia_size;
1186
1187 /* protected by i_rwsem */
1188 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1189 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1190 return -EPERM;
1191
1192 if (newsize != oldsize) {
1193 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1194 oldsize, newsize);
1195 if (error)
1196 return error;
1197 i_size_write(inode, newsize);
1198 update_mtime = true;
1199 } else {
1200 update_ctime = false;
1201 }
1202 if (newsize <= oldsize) {
1203 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1204 if (oldsize > holebegin)
1205 unmap_mapping_range(inode->i_mapping,
1206 holebegin, 0, 1);
1207 if (info->alloced)
1208 shmem_truncate_range(inode,
1209 newsize, (loff_t)-1);
1210 /* unmap again to remove racily COWed private pages */
1211 if (oldsize > holebegin)
1212 unmap_mapping_range(inode->i_mapping,
1213 holebegin, 0, 1);
1214 }
1215 }
1216
1217 if (is_quota_modification(idmap, inode, attr)) {
1218 error = dquot_initialize(inode);
1219 if (error)
1220 return error;
1221 }
1222
1223 /* Transfer quota accounting */
1224 if (i_uid_needs_update(idmap, attr, inode) ||
1225 i_gid_needs_update(idmap, attr, inode)) {
1226 error = dquot_transfer(idmap, inode, attr);
1227
1228 if (error)
1229 return error;
1230 }
1231
1232 setattr_copy(idmap, inode, attr);
1233 if (attr->ia_valid & ATTR_MODE)
1234 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1235 if (!error && update_ctime) {
1236 inode_set_ctime_current(inode);
1237 if (update_mtime)
1238 inode->i_mtime = inode_get_ctime(inode);
1239 inode_inc_iversion(inode);
1240 }
1241 return error;
1242 }
1243
shmem_evict_inode(struct inode * inode)1244 static void shmem_evict_inode(struct inode *inode)
1245 {
1246 struct shmem_inode_info *info = SHMEM_I(inode);
1247 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1248 size_t freed = 0;
1249
1250 if (shmem_mapping(inode->i_mapping)) {
1251 shmem_unacct_size(info->flags, inode->i_size);
1252 inode->i_size = 0;
1253 mapping_set_exiting(inode->i_mapping);
1254 shmem_truncate_range(inode, 0, (loff_t)-1);
1255 if (!list_empty(&info->shrinklist)) {
1256 spin_lock(&sbinfo->shrinklist_lock);
1257 if (!list_empty(&info->shrinklist)) {
1258 list_del_init(&info->shrinklist);
1259 sbinfo->shrinklist_len--;
1260 }
1261 spin_unlock(&sbinfo->shrinklist_lock);
1262 }
1263 while (!list_empty(&info->swaplist)) {
1264 /* Wait while shmem_unuse() is scanning this inode... */
1265 wait_var_event(&info->stop_eviction,
1266 !atomic_read(&info->stop_eviction));
1267 mutex_lock(&shmem_swaplist_mutex);
1268 /* ...but beware of the race if we peeked too early */
1269 if (!atomic_read(&info->stop_eviction))
1270 list_del_init(&info->swaplist);
1271 mutex_unlock(&shmem_swaplist_mutex);
1272 }
1273 }
1274
1275 simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL);
1276 shmem_free_inode(inode->i_sb, freed);
1277 WARN_ON(inode->i_blocks);
1278 clear_inode(inode);
1279 #ifdef CONFIG_TMPFS_QUOTA
1280 dquot_free_inode(inode);
1281 dquot_drop(inode);
1282 #endif
1283 }
1284
shmem_find_swap_entries(struct address_space * mapping,pgoff_t start,struct folio_batch * fbatch,pgoff_t * indices,unsigned int type)1285 static int shmem_find_swap_entries(struct address_space *mapping,
1286 pgoff_t start, struct folio_batch *fbatch,
1287 pgoff_t *indices, unsigned int type)
1288 {
1289 XA_STATE(xas, &mapping->i_pages, start);
1290 struct folio *folio;
1291 swp_entry_t entry;
1292
1293 rcu_read_lock();
1294 xas_for_each(&xas, folio, ULONG_MAX) {
1295 if (xas_retry(&xas, folio))
1296 continue;
1297
1298 if (!xa_is_value(folio))
1299 continue;
1300
1301 entry = radix_to_swp_entry(folio);
1302 /*
1303 * swapin error entries can be found in the mapping. But they're
1304 * deliberately ignored here as we've done everything we can do.
1305 */
1306 if (swp_type(entry) != type)
1307 continue;
1308
1309 indices[folio_batch_count(fbatch)] = xas.xa_index;
1310 if (!folio_batch_add(fbatch, folio))
1311 break;
1312
1313 if (need_resched()) {
1314 xas_pause(&xas);
1315 cond_resched_rcu();
1316 }
1317 }
1318 rcu_read_unlock();
1319
1320 return xas.xa_index;
1321 }
1322
1323 /*
1324 * Move the swapped pages for an inode to page cache. Returns the count
1325 * of pages swapped in, or the error in case of failure.
1326 */
shmem_unuse_swap_entries(struct inode * inode,struct folio_batch * fbatch,pgoff_t * indices)1327 static int shmem_unuse_swap_entries(struct inode *inode,
1328 struct folio_batch *fbatch, pgoff_t *indices)
1329 {
1330 int i = 0;
1331 int ret = 0;
1332 int error = 0;
1333 struct address_space *mapping = inode->i_mapping;
1334
1335 for (i = 0; i < folio_batch_count(fbatch); i++) {
1336 struct folio *folio = fbatch->folios[i];
1337
1338 if (!xa_is_value(folio))
1339 continue;
1340 error = shmem_swapin_folio(inode, indices[i],
1341 &folio, SGP_CACHE,
1342 mapping_gfp_mask(mapping),
1343 NULL, NULL);
1344 if (error == 0) {
1345 folio_unlock(folio);
1346 folio_put(folio);
1347 ret++;
1348 }
1349 if (error == -ENOMEM)
1350 break;
1351 error = 0;
1352 }
1353 return error ? error : ret;
1354 }
1355
1356 /*
1357 * If swap found in inode, free it and move page from swapcache to filecache.
1358 */
shmem_unuse_inode(struct inode * inode,unsigned int type)1359 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1360 {
1361 struct address_space *mapping = inode->i_mapping;
1362 pgoff_t start = 0;
1363 struct folio_batch fbatch;
1364 pgoff_t indices[PAGEVEC_SIZE];
1365 int ret = 0;
1366
1367 do {
1368 folio_batch_init(&fbatch);
1369 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1370 if (folio_batch_count(&fbatch) == 0) {
1371 ret = 0;
1372 break;
1373 }
1374
1375 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1376 if (ret < 0)
1377 break;
1378
1379 start = indices[folio_batch_count(&fbatch) - 1];
1380 } while (true);
1381
1382 return ret;
1383 }
1384
1385 /*
1386 * Read all the shared memory data that resides in the swap
1387 * device 'type' back into memory, so the swap device can be
1388 * unused.
1389 */
shmem_unuse(unsigned int type)1390 int shmem_unuse(unsigned int type)
1391 {
1392 struct shmem_inode_info *info, *next;
1393 int error = 0;
1394
1395 if (list_empty(&shmem_swaplist))
1396 return 0;
1397
1398 mutex_lock(&shmem_swaplist_mutex);
1399 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1400 if (!info->swapped) {
1401 list_del_init(&info->swaplist);
1402 continue;
1403 }
1404 /*
1405 * Drop the swaplist mutex while searching the inode for swap;
1406 * but before doing so, make sure shmem_evict_inode() will not
1407 * remove placeholder inode from swaplist, nor let it be freed
1408 * (igrab() would protect from unlink, but not from unmount).
1409 */
1410 atomic_inc(&info->stop_eviction);
1411 mutex_unlock(&shmem_swaplist_mutex);
1412
1413 error = shmem_unuse_inode(&info->vfs_inode, type);
1414 cond_resched();
1415
1416 mutex_lock(&shmem_swaplist_mutex);
1417 next = list_next_entry(info, swaplist);
1418 if (!info->swapped)
1419 list_del_init(&info->swaplist);
1420 if (atomic_dec_and_test(&info->stop_eviction))
1421 wake_up_var(&info->stop_eviction);
1422 if (error)
1423 break;
1424 }
1425 mutex_unlock(&shmem_swaplist_mutex);
1426
1427 return error;
1428 }
1429
1430 /*
1431 * Move the page from the page cache to the swap cache.
1432 */
shmem_writepage(struct page * page,struct writeback_control * wbc)1433 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1434 {
1435 struct folio *folio = page_folio(page);
1436 struct address_space *mapping = folio->mapping;
1437 struct inode *inode = mapping->host;
1438 struct shmem_inode_info *info = SHMEM_I(inode);
1439 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1440 swp_entry_t swap;
1441 pgoff_t index;
1442
1443 /*
1444 * Our capabilities prevent regular writeback or sync from ever calling
1445 * shmem_writepage; but a stacking filesystem might use ->writepage of
1446 * its underlying filesystem, in which case tmpfs should write out to
1447 * swap only in response to memory pressure, and not for the writeback
1448 * threads or sync.
1449 */
1450 if (WARN_ON_ONCE(!wbc->for_reclaim))
1451 goto redirty;
1452
1453 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1454 goto redirty;
1455
1456 if (!total_swap_pages)
1457 goto redirty;
1458
1459 /*
1460 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1461 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1462 * and its shmem_writeback() needs them to be split when swapping.
1463 */
1464 if (folio_test_large(folio)) {
1465 /* Ensure the subpages are still dirty */
1466 folio_test_set_dirty(folio);
1467 if (split_huge_page(page) < 0)
1468 goto redirty;
1469 folio = page_folio(page);
1470 folio_clear_dirty(folio);
1471 }
1472
1473 index = folio->index;
1474
1475 /*
1476 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1477 * value into swapfile.c, the only way we can correctly account for a
1478 * fallocated folio arriving here is now to initialize it and write it.
1479 *
1480 * That's okay for a folio already fallocated earlier, but if we have
1481 * not yet completed the fallocation, then (a) we want to keep track
1482 * of this folio in case we have to undo it, and (b) it may not be a
1483 * good idea to continue anyway, once we're pushing into swap. So
1484 * reactivate the folio, and let shmem_fallocate() quit when too many.
1485 */
1486 if (!folio_test_uptodate(folio)) {
1487 if (inode->i_private) {
1488 struct shmem_falloc *shmem_falloc;
1489 spin_lock(&inode->i_lock);
1490 shmem_falloc = inode->i_private;
1491 if (shmem_falloc &&
1492 !shmem_falloc->waitq &&
1493 index >= shmem_falloc->start &&
1494 index < shmem_falloc->next)
1495 shmem_falloc->nr_unswapped++;
1496 else
1497 shmem_falloc = NULL;
1498 spin_unlock(&inode->i_lock);
1499 if (shmem_falloc)
1500 goto redirty;
1501 }
1502 folio_zero_range(folio, 0, folio_size(folio));
1503 flush_dcache_folio(folio);
1504 folio_mark_uptodate(folio);
1505 }
1506
1507 swap = folio_alloc_swap(folio);
1508 if (!swap.val)
1509 goto redirty;
1510
1511 /*
1512 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1513 * if it's not already there. Do it now before the folio is
1514 * moved to swap cache, when its pagelock no longer protects
1515 * the inode from eviction. But don't unlock the mutex until
1516 * we've incremented swapped, because shmem_unuse_inode() will
1517 * prune a !swapped inode from the swaplist under this mutex.
1518 */
1519 mutex_lock(&shmem_swaplist_mutex);
1520 if (list_empty(&info->swaplist))
1521 list_add(&info->swaplist, &shmem_swaplist);
1522
1523 if (add_to_swap_cache(folio, swap,
1524 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1525 NULL) == 0) {
1526 shmem_recalc_inode(inode, 0, 1);
1527 swap_shmem_alloc(swap);
1528 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1529
1530 mutex_unlock(&shmem_swaplist_mutex);
1531 BUG_ON(folio_mapped(folio));
1532 swap_writepage(&folio->page, wbc);
1533 return 0;
1534 }
1535
1536 mutex_unlock(&shmem_swaplist_mutex);
1537 put_swap_folio(folio, swap);
1538 redirty:
1539 folio_mark_dirty(folio);
1540 if (wbc->for_reclaim)
1541 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1542 folio_unlock(folio);
1543 return 0;
1544 }
1545
1546 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
shmem_show_mpol(struct seq_file * seq,struct mempolicy * mpol)1547 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1548 {
1549 char buffer[64];
1550
1551 if (!mpol || mpol->mode == MPOL_DEFAULT)
1552 return; /* show nothing */
1553
1554 mpol_to_str(buffer, sizeof(buffer), mpol);
1555
1556 seq_printf(seq, ",mpol=%s", buffer);
1557 }
1558
shmem_get_sbmpol(struct shmem_sb_info * sbinfo)1559 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1560 {
1561 struct mempolicy *mpol = NULL;
1562 if (sbinfo->mpol) {
1563 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1564 mpol = sbinfo->mpol;
1565 mpol_get(mpol);
1566 raw_spin_unlock(&sbinfo->stat_lock);
1567 }
1568 return mpol;
1569 }
1570 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
shmem_show_mpol(struct seq_file * seq,struct mempolicy * mpol)1571 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1572 {
1573 }
shmem_get_sbmpol(struct shmem_sb_info * sbinfo)1574 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1575 {
1576 return NULL;
1577 }
1578 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1579 #ifndef CONFIG_NUMA
1580 #define vm_policy vm_private_data
1581 #endif
1582
shmem_pseudo_vma_init(struct vm_area_struct * vma,struct shmem_inode_info * info,pgoff_t index)1583 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1584 struct shmem_inode_info *info, pgoff_t index)
1585 {
1586 /* Create a pseudo vma that just contains the policy */
1587 vma_init(vma, NULL);
1588 /* Bias interleave by inode number to distribute better across nodes */
1589 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1590 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1591 }
1592
shmem_pseudo_vma_destroy(struct vm_area_struct * vma)1593 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1594 {
1595 /* Drop reference taken by mpol_shared_policy_lookup() */
1596 mpol_cond_put(vma->vm_policy);
1597 }
1598
shmem_swapin(swp_entry_t swap,gfp_t gfp,struct shmem_inode_info * info,pgoff_t index)1599 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1600 struct shmem_inode_info *info, pgoff_t index)
1601 {
1602 struct vm_area_struct pvma;
1603 struct page *page;
1604 struct vm_fault vmf = {
1605 .vma = &pvma,
1606 };
1607
1608 shmem_pseudo_vma_init(&pvma, info, index);
1609 page = swap_cluster_readahead(swap, gfp, &vmf);
1610 shmem_pseudo_vma_destroy(&pvma);
1611
1612 if (!page)
1613 return NULL;
1614 return page_folio(page);
1615 }
1616
1617 /*
1618 * Make sure huge_gfp is always more limited than limit_gfp.
1619 * Some of the flags set permissions, while others set limitations.
1620 */
limit_gfp_mask(gfp_t huge_gfp,gfp_t limit_gfp)1621 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1622 {
1623 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1624 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1625 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1626 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1627
1628 /* Allow allocations only from the originally specified zones. */
1629 result |= zoneflags;
1630
1631 /*
1632 * Minimize the result gfp by taking the union with the deny flags,
1633 * and the intersection of the allow flags.
1634 */
1635 result |= (limit_gfp & denyflags);
1636 result |= (huge_gfp & limit_gfp) & allowflags;
1637
1638 return result;
1639 }
1640
shmem_alloc_hugefolio(gfp_t gfp,struct shmem_inode_info * info,pgoff_t index)1641 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1642 struct shmem_inode_info *info, pgoff_t index)
1643 {
1644 struct vm_area_struct pvma;
1645 struct address_space *mapping = info->vfs_inode.i_mapping;
1646 pgoff_t hindex;
1647 struct folio *folio;
1648
1649 hindex = round_down(index, HPAGE_PMD_NR);
1650 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1651 XA_PRESENT))
1652 return NULL;
1653
1654 shmem_pseudo_vma_init(&pvma, info, hindex);
1655 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1656 shmem_pseudo_vma_destroy(&pvma);
1657 if (!folio)
1658 count_vm_event(THP_FILE_FALLBACK);
1659 return folio;
1660 }
1661
shmem_alloc_folio(gfp_t gfp,struct shmem_inode_info * info,pgoff_t index)1662 static struct folio *shmem_alloc_folio(gfp_t gfp,
1663 struct shmem_inode_info *info, pgoff_t index)
1664 {
1665 struct vm_area_struct pvma;
1666 struct folio *folio;
1667
1668 shmem_pseudo_vma_init(&pvma, info, index);
1669 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1670 shmem_pseudo_vma_destroy(&pvma);
1671
1672 return folio;
1673 }
1674
shmem_alloc_and_acct_folio(gfp_t gfp,struct inode * inode,pgoff_t index,bool huge)1675 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1676 pgoff_t index, bool huge)
1677 {
1678 struct shmem_inode_info *info = SHMEM_I(inode);
1679 struct folio *folio;
1680 int nr;
1681 int err;
1682
1683 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1684 huge = false;
1685 nr = huge ? HPAGE_PMD_NR : 1;
1686
1687 err = shmem_inode_acct_block(inode, nr);
1688 if (err)
1689 goto failed;
1690
1691 if (huge)
1692 folio = shmem_alloc_hugefolio(gfp, info, index);
1693 else
1694 folio = shmem_alloc_folio(gfp, info, index);
1695 if (folio) {
1696 __folio_set_locked(folio);
1697 __folio_set_swapbacked(folio);
1698 return folio;
1699 }
1700
1701 err = -ENOMEM;
1702 shmem_inode_unacct_blocks(inode, nr);
1703 failed:
1704 return ERR_PTR(err);
1705 }
1706
1707 /*
1708 * When a page is moved from swapcache to shmem filecache (either by the
1709 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1710 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1711 * ignorance of the mapping it belongs to. If that mapping has special
1712 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1713 * we may need to copy to a suitable page before moving to filecache.
1714 *
1715 * In a future release, this may well be extended to respect cpuset and
1716 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1717 * but for now it is a simple matter of zone.
1718 */
shmem_should_replace_folio(struct folio * folio,gfp_t gfp)1719 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1720 {
1721 return folio_zonenum(folio) > gfp_zone(gfp);
1722 }
1723
shmem_replace_folio(struct folio ** foliop,gfp_t gfp,struct shmem_inode_info * info,pgoff_t index)1724 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1725 struct shmem_inode_info *info, pgoff_t index)
1726 {
1727 struct folio *old, *new;
1728 struct address_space *swap_mapping;
1729 swp_entry_t entry;
1730 pgoff_t swap_index;
1731 int error;
1732
1733 old = *foliop;
1734 entry = old->swap;
1735 swap_index = swp_offset(entry);
1736 swap_mapping = swap_address_space(entry);
1737
1738 /*
1739 * We have arrived here because our zones are constrained, so don't
1740 * limit chance of success by further cpuset and node constraints.
1741 */
1742 gfp &= ~GFP_CONSTRAINT_MASK;
1743 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1744 new = shmem_alloc_folio(gfp, info, index);
1745 if (!new)
1746 return -ENOMEM;
1747
1748 folio_get(new);
1749 folio_copy(new, old);
1750 flush_dcache_folio(new);
1751
1752 __folio_set_locked(new);
1753 __folio_set_swapbacked(new);
1754 folio_mark_uptodate(new);
1755 new->swap = entry;
1756 folio_set_swapcache(new);
1757
1758 /*
1759 * Our caller will very soon move newpage out of swapcache, but it's
1760 * a nice clean interface for us to replace oldpage by newpage there.
1761 */
1762 xa_lock_irq(&swap_mapping->i_pages);
1763 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1764 if (!error) {
1765 mem_cgroup_migrate(old, new);
1766 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1767 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1768 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1769 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1770 }
1771 xa_unlock_irq(&swap_mapping->i_pages);
1772
1773 if (unlikely(error)) {
1774 /*
1775 * Is this possible? I think not, now that our callers check
1776 * both PageSwapCache and page_private after getting page lock;
1777 * but be defensive. Reverse old to newpage for clear and free.
1778 */
1779 old = new;
1780 } else {
1781 folio_add_lru(new);
1782 *foliop = new;
1783 }
1784
1785 folio_clear_swapcache(old);
1786 old->private = NULL;
1787
1788 folio_unlock(old);
1789 folio_put_refs(old, 2);
1790 return error;
1791 }
1792
shmem_set_folio_swapin_error(struct inode * inode,pgoff_t index,struct folio * folio,swp_entry_t swap)1793 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1794 struct folio *folio, swp_entry_t swap)
1795 {
1796 struct address_space *mapping = inode->i_mapping;
1797 swp_entry_t swapin_error;
1798 void *old;
1799
1800 swapin_error = make_poisoned_swp_entry();
1801 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1802 swp_to_radix_entry(swap),
1803 swp_to_radix_entry(swapin_error), 0);
1804 if (old != swp_to_radix_entry(swap))
1805 return;
1806
1807 folio_wait_writeback(folio);
1808 delete_from_swap_cache(folio);
1809 /*
1810 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
1811 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
1812 * in shmem_evict_inode().
1813 */
1814 shmem_recalc_inode(inode, -1, -1);
1815 swap_free(swap);
1816 }
1817
1818 /*
1819 * Swap in the folio pointed to by *foliop.
1820 * Caller has to make sure that *foliop contains a valid swapped folio.
1821 * Returns 0 and the folio in foliop if success. On failure, returns the
1822 * error code and NULL in *foliop.
1823 */
shmem_swapin_folio(struct inode * inode,pgoff_t index,struct folio ** foliop,enum sgp_type sgp,gfp_t gfp,struct vm_area_struct * vma,vm_fault_t * fault_type)1824 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1825 struct folio **foliop, enum sgp_type sgp,
1826 gfp_t gfp, struct vm_area_struct *vma,
1827 vm_fault_t *fault_type)
1828 {
1829 struct address_space *mapping = inode->i_mapping;
1830 struct shmem_inode_info *info = SHMEM_I(inode);
1831 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1832 struct swap_info_struct *si;
1833 struct folio *folio = NULL;
1834 swp_entry_t swap;
1835 int error;
1836
1837 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1838 swap = radix_to_swp_entry(*foliop);
1839 *foliop = NULL;
1840
1841 if (is_poisoned_swp_entry(swap))
1842 return -EIO;
1843
1844 si = get_swap_device(swap);
1845 if (!si) {
1846 if (!shmem_confirm_swap(mapping, index, swap))
1847 return -EEXIST;
1848 else
1849 return -EINVAL;
1850 }
1851
1852 /* Look it up and read it in.. */
1853 folio = swap_cache_get_folio(swap, NULL, 0);
1854 if (!folio) {
1855 /* Or update major stats only when swapin succeeds?? */
1856 if (fault_type) {
1857 *fault_type |= VM_FAULT_MAJOR;
1858 count_vm_event(PGMAJFAULT);
1859 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1860 }
1861 /* Here we actually start the io */
1862 folio = shmem_swapin(swap, gfp, info, index);
1863 if (!folio) {
1864 error = -ENOMEM;
1865 goto failed;
1866 }
1867 }
1868
1869 /* We have to do this with folio locked to prevent races */
1870 folio_lock(folio);
1871 if (!folio_test_swapcache(folio) ||
1872 folio->swap.val != swap.val ||
1873 !shmem_confirm_swap(mapping, index, swap)) {
1874 error = -EEXIST;
1875 goto unlock;
1876 }
1877 if (!folio_test_uptodate(folio)) {
1878 error = -EIO;
1879 goto failed;
1880 }
1881 folio_wait_writeback(folio);
1882
1883 /*
1884 * Some architectures may have to restore extra metadata to the
1885 * folio after reading from swap.
1886 */
1887 arch_swap_restore(swap, folio);
1888
1889 if (shmem_should_replace_folio(folio, gfp)) {
1890 error = shmem_replace_folio(&folio, gfp, info, index);
1891 if (error)
1892 goto failed;
1893 }
1894
1895 error = shmem_add_to_page_cache(folio, mapping, index,
1896 swp_to_radix_entry(swap), gfp,
1897 charge_mm);
1898 if (error)
1899 goto failed;
1900
1901 shmem_recalc_inode(inode, 0, -1);
1902
1903 if (sgp == SGP_WRITE)
1904 folio_mark_accessed(folio);
1905
1906 delete_from_swap_cache(folio);
1907 folio_mark_dirty(folio);
1908 swap_free(swap);
1909 put_swap_device(si);
1910
1911 *foliop = folio;
1912 return 0;
1913 failed:
1914 if (!shmem_confirm_swap(mapping, index, swap))
1915 error = -EEXIST;
1916 if (error == -EIO)
1917 shmem_set_folio_swapin_error(inode, index, folio, swap);
1918 unlock:
1919 if (folio) {
1920 folio_unlock(folio);
1921 folio_put(folio);
1922 }
1923 put_swap_device(si);
1924
1925 return error;
1926 }
1927
1928 /*
1929 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1930 *
1931 * If we allocate a new one we do not mark it dirty. That's up to the
1932 * vm. If we swap it in we mark it dirty since we also free the swap
1933 * entry since a page cannot live in both the swap and page cache.
1934 *
1935 * vma, vmf, and fault_type are only supplied by shmem_fault:
1936 * otherwise they are NULL.
1937 */
shmem_get_folio_gfp(struct inode * inode,pgoff_t index,struct folio ** foliop,enum sgp_type sgp,gfp_t gfp,struct vm_area_struct * vma,struct vm_fault * vmf,vm_fault_t * fault_type)1938 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1939 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1940 struct vm_area_struct *vma, struct vm_fault *vmf,
1941 vm_fault_t *fault_type)
1942 {
1943 struct address_space *mapping = inode->i_mapping;
1944 struct shmem_inode_info *info = SHMEM_I(inode);
1945 struct shmem_sb_info *sbinfo;
1946 struct mm_struct *charge_mm;
1947 struct folio *folio;
1948 pgoff_t hindex;
1949 gfp_t huge_gfp;
1950 int error;
1951 int once = 0;
1952 int alloced = 0;
1953
1954 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1955 return -EFBIG;
1956 repeat:
1957 if (sgp <= SGP_CACHE &&
1958 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1959 return -EINVAL;
1960 }
1961
1962 sbinfo = SHMEM_SB(inode->i_sb);
1963 charge_mm = vma ? vma->vm_mm : NULL;
1964
1965 folio = filemap_get_entry(mapping, index);
1966 if (folio && vma && userfaultfd_minor(vma)) {
1967 if (!xa_is_value(folio))
1968 folio_put(folio);
1969 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1970 return 0;
1971 }
1972
1973 if (xa_is_value(folio)) {
1974 error = shmem_swapin_folio(inode, index, &folio,
1975 sgp, gfp, vma, fault_type);
1976 if (error == -EEXIST)
1977 goto repeat;
1978
1979 *foliop = folio;
1980 return error;
1981 }
1982
1983 if (folio) {
1984 folio_lock(folio);
1985
1986 /* Has the folio been truncated or swapped out? */
1987 if (unlikely(folio->mapping != mapping)) {
1988 folio_unlock(folio);
1989 folio_put(folio);
1990 goto repeat;
1991 }
1992 if (sgp == SGP_WRITE)
1993 folio_mark_accessed(folio);
1994 if (folio_test_uptodate(folio))
1995 goto out;
1996 /* fallocated folio */
1997 if (sgp != SGP_READ)
1998 goto clear;
1999 folio_unlock(folio);
2000 folio_put(folio);
2001 }
2002
2003 /*
2004 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
2005 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
2006 */
2007 *foliop = NULL;
2008 if (sgp == SGP_READ)
2009 return 0;
2010 if (sgp == SGP_NOALLOC)
2011 return -ENOENT;
2012
2013 /*
2014 * Fast cache lookup and swap lookup did not find it: allocate.
2015 */
2016
2017 if (vma && userfaultfd_missing(vma)) {
2018 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2019 return 0;
2020 }
2021
2022 if (!shmem_is_huge(inode, index, false,
2023 vma ? vma->vm_mm : NULL, vma ? vma->vm_flags : 0))
2024 goto alloc_nohuge;
2025
2026 huge_gfp = vma_thp_gfp_mask(vma);
2027 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2028 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
2029 if (IS_ERR(folio)) {
2030 alloc_nohuge:
2031 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
2032 }
2033 if (IS_ERR(folio)) {
2034 int retry = 5;
2035
2036 error = PTR_ERR(folio);
2037 folio = NULL;
2038 if (error != -ENOSPC)
2039 goto unlock;
2040 /*
2041 * Try to reclaim some space by splitting a large folio
2042 * beyond i_size on the filesystem.
2043 */
2044 while (retry--) {
2045 int ret;
2046
2047 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
2048 if (ret == SHRINK_STOP)
2049 break;
2050 if (ret)
2051 goto alloc_nohuge;
2052 }
2053 goto unlock;
2054 }
2055
2056 hindex = round_down(index, folio_nr_pages(folio));
2057
2058 if (sgp == SGP_WRITE)
2059 __folio_set_referenced(folio);
2060
2061 error = shmem_add_to_page_cache(folio, mapping, hindex,
2062 NULL, gfp & GFP_RECLAIM_MASK,
2063 charge_mm);
2064 if (error)
2065 goto unacct;
2066
2067 folio_add_lru(folio);
2068 shmem_recalc_inode(inode, folio_nr_pages(folio), 0);
2069 alloced = true;
2070
2071 if (folio_test_pmd_mappable(folio) &&
2072 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2073 folio_next_index(folio) - 1) {
2074 /*
2075 * Part of the large folio is beyond i_size: subject
2076 * to shrink under memory pressure.
2077 */
2078 spin_lock(&sbinfo->shrinklist_lock);
2079 /*
2080 * _careful to defend against unlocked access to
2081 * ->shrink_list in shmem_unused_huge_shrink()
2082 */
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 }
2090
2091 /*
2092 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2093 */
2094 if (sgp == SGP_FALLOC)
2095 sgp = SGP_WRITE;
2096 clear:
2097 /*
2098 * Let SGP_WRITE caller clear ends if write does not fill folio;
2099 * but SGP_FALLOC on a folio fallocated earlier must initialize
2100 * it now, lest undo on failure cancel our earlier guarantee.
2101 */
2102 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2103 long i, n = folio_nr_pages(folio);
2104
2105 for (i = 0; i < n; i++)
2106 clear_highpage(folio_page(folio, i));
2107 flush_dcache_folio(folio);
2108 folio_mark_uptodate(folio);
2109 }
2110
2111 /* Perhaps the file has been truncated since we checked */
2112 if (sgp <= SGP_CACHE &&
2113 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2114 if (alloced) {
2115 folio_clear_dirty(folio);
2116 filemap_remove_folio(folio);
2117 shmem_recalc_inode(inode, 0, 0);
2118 }
2119 error = -EINVAL;
2120 goto unlock;
2121 }
2122 out:
2123 *foliop = folio;
2124 return 0;
2125
2126 /*
2127 * Error recovery.
2128 */
2129 unacct:
2130 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2131
2132 if (folio_test_large(folio)) {
2133 folio_unlock(folio);
2134 folio_put(folio);
2135 goto alloc_nohuge;
2136 }
2137 unlock:
2138 if (folio) {
2139 folio_unlock(folio);
2140 folio_put(folio);
2141 }
2142 if (error == -ENOSPC && !once++) {
2143 shmem_recalc_inode(inode, 0, 0);
2144 goto repeat;
2145 }
2146 if (error == -EEXIST)
2147 goto repeat;
2148 return error;
2149 }
2150
shmem_get_folio(struct inode * inode,pgoff_t index,struct folio ** foliop,enum sgp_type sgp)2151 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2152 enum sgp_type sgp)
2153 {
2154 return shmem_get_folio_gfp(inode, index, foliop, sgp,
2155 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2156 }
2157
2158 /*
2159 * This is like autoremove_wake_function, but it removes the wait queue
2160 * entry unconditionally - even if something else had already woken the
2161 * target.
2162 */
synchronous_wake_function(wait_queue_entry_t * wait,unsigned mode,int sync,void * key)2163 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2164 {
2165 int ret = default_wake_function(wait, mode, sync, key);
2166 list_del_init(&wait->entry);
2167 return ret;
2168 }
2169
shmem_fault(struct vm_fault * vmf)2170 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2171 {
2172 struct vm_area_struct *vma = vmf->vma;
2173 struct inode *inode = file_inode(vma->vm_file);
2174 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2175 struct folio *folio = NULL;
2176 int err;
2177 vm_fault_t ret = VM_FAULT_LOCKED;
2178
2179 /*
2180 * Trinity finds that probing a hole which tmpfs is punching can
2181 * prevent the hole-punch from ever completing: which in turn
2182 * locks writers out with its hold on i_rwsem. So refrain from
2183 * faulting pages into the hole while it's being punched. Although
2184 * shmem_undo_range() does remove the additions, it may be unable to
2185 * keep up, as each new page needs its own unmap_mapping_range() call,
2186 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2187 *
2188 * It does not matter if we sometimes reach this check just before the
2189 * hole-punch begins, so that one fault then races with the punch:
2190 * we just need to make racing faults a rare case.
2191 *
2192 * The implementation below would be much simpler if we just used a
2193 * standard mutex or completion: but we cannot take i_rwsem in fault,
2194 * and bloating every shmem inode for this unlikely case would be sad.
2195 */
2196 if (unlikely(inode->i_private)) {
2197 struct shmem_falloc *shmem_falloc;
2198
2199 spin_lock(&inode->i_lock);
2200 shmem_falloc = inode->i_private;
2201 if (shmem_falloc &&
2202 shmem_falloc->waitq &&
2203 vmf->pgoff >= shmem_falloc->start &&
2204 vmf->pgoff < shmem_falloc->next) {
2205 struct file *fpin;
2206 wait_queue_head_t *shmem_falloc_waitq;
2207 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2208
2209 ret = VM_FAULT_NOPAGE;
2210 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2211 if (fpin)
2212 ret = VM_FAULT_RETRY;
2213
2214 shmem_falloc_waitq = shmem_falloc->waitq;
2215 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2216 TASK_UNINTERRUPTIBLE);
2217 spin_unlock(&inode->i_lock);
2218 schedule();
2219
2220 /*
2221 * shmem_falloc_waitq points into the shmem_fallocate()
2222 * stack of the hole-punching task: shmem_falloc_waitq
2223 * is usually invalid by the time we reach here, but
2224 * finish_wait() does not dereference it in that case;
2225 * though i_lock needed lest racing with wake_up_all().
2226 */
2227 spin_lock(&inode->i_lock);
2228 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2229 spin_unlock(&inode->i_lock);
2230
2231 if (fpin)
2232 fput(fpin);
2233 return ret;
2234 }
2235 spin_unlock(&inode->i_lock);
2236 }
2237
2238 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2239 gfp, vma, vmf, &ret);
2240 if (err)
2241 return vmf_error(err);
2242 if (folio)
2243 vmf->page = folio_file_page(folio, vmf->pgoff);
2244 return ret;
2245 }
2246
shmem_get_unmapped_area(struct file * file,unsigned long uaddr,unsigned long len,unsigned long pgoff,unsigned long flags)2247 unsigned long shmem_get_unmapped_area(struct file *file,
2248 unsigned long uaddr, unsigned long len,
2249 unsigned long pgoff, unsigned long flags)
2250 {
2251 unsigned long (*get_area)(struct file *,
2252 unsigned long, unsigned long, unsigned long, unsigned long);
2253 unsigned long addr;
2254 unsigned long offset;
2255 unsigned long inflated_len;
2256 unsigned long inflated_addr;
2257 unsigned long inflated_offset;
2258
2259 if (len > TASK_SIZE)
2260 return -ENOMEM;
2261
2262 get_area = current->mm->get_unmapped_area;
2263 addr = get_area(file, uaddr, len, pgoff, flags);
2264
2265 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2266 return addr;
2267 if (IS_ERR_VALUE(addr))
2268 return addr;
2269 if (addr & ~PAGE_MASK)
2270 return addr;
2271 if (addr > TASK_SIZE - len)
2272 return addr;
2273
2274 if (shmem_huge == SHMEM_HUGE_DENY)
2275 return addr;
2276 if (len < HPAGE_PMD_SIZE)
2277 return addr;
2278 if (flags & MAP_FIXED)
2279 return addr;
2280 /*
2281 * Our priority is to support MAP_SHARED mapped hugely;
2282 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2283 * But if caller specified an address hint and we allocated area there
2284 * successfully, respect that as before.
2285 */
2286 if (uaddr == addr)
2287 return addr;
2288
2289 if (shmem_huge != SHMEM_HUGE_FORCE) {
2290 struct super_block *sb;
2291
2292 if (file) {
2293 VM_BUG_ON(file->f_op != &shmem_file_operations);
2294 sb = file_inode(file)->i_sb;
2295 } else {
2296 /*
2297 * Called directly from mm/mmap.c, or drivers/char/mem.c
2298 * for "/dev/zero", to create a shared anonymous object.
2299 */
2300 if (IS_ERR(shm_mnt))
2301 return addr;
2302 sb = shm_mnt->mnt_sb;
2303 }
2304 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2305 return addr;
2306 }
2307
2308 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2309 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2310 return addr;
2311 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2312 return addr;
2313
2314 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2315 if (inflated_len > TASK_SIZE)
2316 return addr;
2317 if (inflated_len < len)
2318 return addr;
2319
2320 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2321 if (IS_ERR_VALUE(inflated_addr))
2322 return addr;
2323 if (inflated_addr & ~PAGE_MASK)
2324 return addr;
2325
2326 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2327 inflated_addr += offset - inflated_offset;
2328 if (inflated_offset > offset)
2329 inflated_addr += HPAGE_PMD_SIZE;
2330
2331 if (inflated_addr > TASK_SIZE - len)
2332 return addr;
2333 return inflated_addr;
2334 }
2335
2336 #ifdef CONFIG_NUMA
shmem_set_policy(struct vm_area_struct * vma,struct mempolicy * mpol)2337 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2338 {
2339 struct inode *inode = file_inode(vma->vm_file);
2340 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2341 }
2342
shmem_get_policy(struct vm_area_struct * vma,unsigned long addr)2343 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2344 unsigned long addr)
2345 {
2346 struct inode *inode = file_inode(vma->vm_file);
2347 pgoff_t index;
2348
2349 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2350 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2351 }
2352 #endif
2353
shmem_lock(struct file * file,int lock,struct ucounts * ucounts)2354 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2355 {
2356 struct inode *inode = file_inode(file);
2357 struct shmem_inode_info *info = SHMEM_I(inode);
2358 int retval = -ENOMEM;
2359
2360 /*
2361 * What serializes the accesses to info->flags?
2362 * ipc_lock_object() when called from shmctl_do_lock(),
2363 * no serialization needed when called from shm_destroy().
2364 */
2365 if (lock && !(info->flags & VM_LOCKED)) {
2366 if (!user_shm_lock(inode->i_size, ucounts))
2367 goto out_nomem;
2368 info->flags |= VM_LOCKED;
2369 mapping_set_unevictable(file->f_mapping);
2370 }
2371 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2372 user_shm_unlock(inode->i_size, ucounts);
2373 info->flags &= ~VM_LOCKED;
2374 mapping_clear_unevictable(file->f_mapping);
2375 }
2376 retval = 0;
2377
2378 out_nomem:
2379 return retval;
2380 }
2381
shmem_mmap(struct file * file,struct vm_area_struct * vma)2382 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2383 {
2384 struct inode *inode = file_inode(file);
2385 struct shmem_inode_info *info = SHMEM_I(inode);
2386 int ret;
2387
2388 ret = seal_check_future_write(info->seals, vma);
2389 if (ret)
2390 return ret;
2391
2392 /* arm64 - allow memory tagging on RAM-based files */
2393 vm_flags_set(vma, VM_MTE_ALLOWED);
2394
2395 file_accessed(file);
2396 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2397 if (inode->i_nlink)
2398 vma->vm_ops = &shmem_vm_ops;
2399 else
2400 vma->vm_ops = &shmem_anon_vm_ops;
2401 return 0;
2402 }
2403
shmem_file_open(struct inode * inode,struct file * file)2404 static int shmem_file_open(struct inode *inode, struct file *file)
2405 {
2406 file->f_mode |= FMODE_CAN_ODIRECT;
2407 return generic_file_open(inode, file);
2408 }
2409
2410 #ifdef CONFIG_TMPFS_XATTR
2411 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2412
2413 /*
2414 * chattr's fsflags are unrelated to extended attributes,
2415 * but tmpfs has chosen to enable them under the same config option.
2416 */
shmem_set_inode_flags(struct inode * inode,unsigned int fsflags)2417 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2418 {
2419 unsigned int i_flags = 0;
2420
2421 if (fsflags & FS_NOATIME_FL)
2422 i_flags |= S_NOATIME;
2423 if (fsflags & FS_APPEND_FL)
2424 i_flags |= S_APPEND;
2425 if (fsflags & FS_IMMUTABLE_FL)
2426 i_flags |= S_IMMUTABLE;
2427 /*
2428 * But FS_NODUMP_FL does not require any action in i_flags.
2429 */
2430 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2431 }
2432 #else
shmem_set_inode_flags(struct inode * inode,unsigned int fsflags)2433 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2434 {
2435 }
2436 #define shmem_initxattrs NULL
2437 #endif
2438
shmem_get_offset_ctx(struct inode * inode)2439 static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2440 {
2441 return &SHMEM_I(inode)->dir_offsets;
2442 }
2443
__shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)2444 static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2445 struct super_block *sb,
2446 struct inode *dir, umode_t mode,
2447 dev_t dev, unsigned long flags)
2448 {
2449 struct inode *inode;
2450 struct shmem_inode_info *info;
2451 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2452 ino_t ino;
2453 int err;
2454
2455 err = shmem_reserve_inode(sb, &ino);
2456 if (err)
2457 return ERR_PTR(err);
2458
2459
2460 inode = new_inode(sb);
2461 if (!inode) {
2462 shmem_free_inode(sb, 0);
2463 return ERR_PTR(-ENOSPC);
2464 }
2465
2466 inode->i_ino = ino;
2467 inode_init_owner(idmap, inode, dir, mode);
2468 inode->i_blocks = 0;
2469 inode->i_atime = inode->i_mtime = inode_set_ctime_current(inode);
2470 inode->i_generation = get_random_u32();
2471 info = SHMEM_I(inode);
2472 memset(info, 0, (char *)inode - (char *)info);
2473 spin_lock_init(&info->lock);
2474 atomic_set(&info->stop_eviction, 0);
2475 info->seals = F_SEAL_SEAL;
2476 info->flags = flags & VM_NORESERVE;
2477 info->i_crtime = inode->i_mtime;
2478 info->fsflags = (dir == NULL) ? 0 :
2479 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2480 if (info->fsflags)
2481 shmem_set_inode_flags(inode, info->fsflags);
2482 INIT_LIST_HEAD(&info->shrinklist);
2483 INIT_LIST_HEAD(&info->swaplist);
2484 INIT_LIST_HEAD(&info->swaplist);
2485 if (sbinfo->noswap)
2486 mapping_set_unevictable(inode->i_mapping);
2487 simple_xattrs_init(&info->xattrs);
2488 cache_no_acl(inode);
2489 mapping_set_large_folios(inode->i_mapping);
2490
2491 switch (mode & S_IFMT) {
2492 default:
2493 inode->i_op = &shmem_special_inode_operations;
2494 init_special_inode(inode, mode, dev);
2495 break;
2496 case S_IFREG:
2497 inode->i_mapping->a_ops = &shmem_aops;
2498 inode->i_op = &shmem_inode_operations;
2499 inode->i_fop = &shmem_file_operations;
2500 mpol_shared_policy_init(&info->policy,
2501 shmem_get_sbmpol(sbinfo));
2502 break;
2503 case S_IFDIR:
2504 inc_nlink(inode);
2505 /* Some things misbehave if size == 0 on a directory */
2506 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2507 inode->i_op = &shmem_dir_inode_operations;
2508 inode->i_fop = &simple_offset_dir_operations;
2509 simple_offset_init(shmem_get_offset_ctx(inode));
2510 break;
2511 case S_IFLNK:
2512 /*
2513 * Must not load anything in the rbtree,
2514 * mpol_free_shared_policy will not be called.
2515 */
2516 mpol_shared_policy_init(&info->policy, NULL);
2517 break;
2518 }
2519
2520 lockdep_annotate_inode_mutex_key(inode);
2521 return inode;
2522 }
2523
2524 #ifdef CONFIG_TMPFS_QUOTA
shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)2525 static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2526 struct super_block *sb, struct inode *dir,
2527 umode_t mode, dev_t dev, unsigned long flags)
2528 {
2529 int err;
2530 struct inode *inode;
2531
2532 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2533 if (IS_ERR(inode))
2534 return inode;
2535
2536 err = dquot_initialize(inode);
2537 if (err)
2538 goto errout;
2539
2540 err = dquot_alloc_inode(inode);
2541 if (err) {
2542 dquot_drop(inode);
2543 goto errout;
2544 }
2545 return inode;
2546
2547 errout:
2548 inode->i_flags |= S_NOQUOTA;
2549 iput(inode);
2550 return ERR_PTR(err);
2551 }
2552 #else
shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)2553 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2554 struct super_block *sb, struct inode *dir,
2555 umode_t mode, dev_t dev, unsigned long flags)
2556 {
2557 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2558 }
2559 #endif /* CONFIG_TMPFS_QUOTA */
2560
2561 #ifdef CONFIG_USERFAULTFD
shmem_mfill_atomic_pte(pmd_t * dst_pmd,struct vm_area_struct * dst_vma,unsigned long dst_addr,unsigned long src_addr,uffd_flags_t flags,struct folio ** foliop)2562 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2563 struct vm_area_struct *dst_vma,
2564 unsigned long dst_addr,
2565 unsigned long src_addr,
2566 uffd_flags_t flags,
2567 struct folio **foliop)
2568 {
2569 struct inode *inode = file_inode(dst_vma->vm_file);
2570 struct shmem_inode_info *info = SHMEM_I(inode);
2571 struct address_space *mapping = inode->i_mapping;
2572 gfp_t gfp = mapping_gfp_mask(mapping);
2573 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2574 void *page_kaddr;
2575 struct folio *folio;
2576 int ret;
2577 pgoff_t max_off;
2578
2579 if (shmem_inode_acct_block(inode, 1)) {
2580 /*
2581 * We may have got a page, returned -ENOENT triggering a retry,
2582 * and now we find ourselves with -ENOMEM. Release the page, to
2583 * avoid a BUG_ON in our caller.
2584 */
2585 if (unlikely(*foliop)) {
2586 folio_put(*foliop);
2587 *foliop = NULL;
2588 }
2589 return -ENOMEM;
2590 }
2591
2592 if (!*foliop) {
2593 ret = -ENOMEM;
2594 folio = shmem_alloc_folio(gfp, info, pgoff);
2595 if (!folio)
2596 goto out_unacct_blocks;
2597
2598 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2599 page_kaddr = kmap_local_folio(folio, 0);
2600 /*
2601 * The read mmap_lock is held here. Despite the
2602 * mmap_lock being read recursive a deadlock is still
2603 * possible if a writer has taken a lock. For example:
2604 *
2605 * process A thread 1 takes read lock on own mmap_lock
2606 * process A thread 2 calls mmap, blocks taking write lock
2607 * process B thread 1 takes page fault, read lock on own mmap lock
2608 * process B thread 2 calls mmap, blocks taking write lock
2609 * process A thread 1 blocks taking read lock on process B
2610 * process B thread 1 blocks taking read lock on process A
2611 *
2612 * Disable page faults to prevent potential deadlock
2613 * and retry the copy outside the mmap_lock.
2614 */
2615 pagefault_disable();
2616 ret = copy_from_user(page_kaddr,
2617 (const void __user *)src_addr,
2618 PAGE_SIZE);
2619 pagefault_enable();
2620 kunmap_local(page_kaddr);
2621
2622 /* fallback to copy_from_user outside mmap_lock */
2623 if (unlikely(ret)) {
2624 *foliop = folio;
2625 ret = -ENOENT;
2626 /* don't free the page */
2627 goto out_unacct_blocks;
2628 }
2629
2630 flush_dcache_folio(folio);
2631 } else { /* ZEROPAGE */
2632 clear_user_highpage(&folio->page, dst_addr);
2633 }
2634 } else {
2635 folio = *foliop;
2636 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2637 *foliop = NULL;
2638 }
2639
2640 VM_BUG_ON(folio_test_locked(folio));
2641 VM_BUG_ON(folio_test_swapbacked(folio));
2642 __folio_set_locked(folio);
2643 __folio_set_swapbacked(folio);
2644 __folio_mark_uptodate(folio);
2645
2646 ret = -EFAULT;
2647 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2648 if (unlikely(pgoff >= max_off))
2649 goto out_release;
2650
2651 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2652 gfp & GFP_RECLAIM_MASK, dst_vma->vm_mm);
2653 if (ret)
2654 goto out_release;
2655
2656 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
2657 &folio->page, true, flags);
2658 if (ret)
2659 goto out_delete_from_cache;
2660
2661 shmem_recalc_inode(inode, 1, 0);
2662 folio_unlock(folio);
2663 return 0;
2664 out_delete_from_cache:
2665 filemap_remove_folio(folio);
2666 out_release:
2667 folio_unlock(folio);
2668 folio_put(folio);
2669 out_unacct_blocks:
2670 shmem_inode_unacct_blocks(inode, 1);
2671 return ret;
2672 }
2673 #endif /* CONFIG_USERFAULTFD */
2674
2675 #ifdef CONFIG_TMPFS
2676 static const struct inode_operations shmem_symlink_inode_operations;
2677 static const struct inode_operations shmem_short_symlink_operations;
2678
2679 static int
shmem_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct page ** pagep,void ** fsdata)2680 shmem_write_begin(struct file *file, struct address_space *mapping,
2681 loff_t pos, unsigned len,
2682 struct page **pagep, void **fsdata)
2683 {
2684 struct inode *inode = mapping->host;
2685 struct shmem_inode_info *info = SHMEM_I(inode);
2686 pgoff_t index = pos >> PAGE_SHIFT;
2687 struct folio *folio;
2688 int ret = 0;
2689
2690 /* i_rwsem is held by caller */
2691 if (unlikely(info->seals & (F_SEAL_GROW |
2692 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2693 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2694 return -EPERM;
2695 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2696 return -EPERM;
2697 }
2698
2699 ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2700
2701 if (ret)
2702 return ret;
2703
2704 *pagep = folio_file_page(folio, index);
2705 if (PageHWPoison(*pagep)) {
2706 folio_unlock(folio);
2707 folio_put(folio);
2708 *pagep = NULL;
2709 return -EIO;
2710 }
2711
2712 return 0;
2713 }
2714
2715 static int
shmem_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)2716 shmem_write_end(struct file *file, struct address_space *mapping,
2717 loff_t pos, unsigned len, unsigned copied,
2718 struct page *page, void *fsdata)
2719 {
2720 struct folio *folio = page_folio(page);
2721 struct inode *inode = mapping->host;
2722
2723 if (pos + copied > inode->i_size)
2724 i_size_write(inode, pos + copied);
2725
2726 if (!folio_test_uptodate(folio)) {
2727 if (copied < folio_size(folio)) {
2728 size_t from = offset_in_folio(folio, pos);
2729 folio_zero_segments(folio, 0, from,
2730 from + copied, folio_size(folio));
2731 }
2732 folio_mark_uptodate(folio);
2733 }
2734 folio_mark_dirty(folio);
2735 folio_unlock(folio);
2736 folio_put(folio);
2737
2738 return copied;
2739 }
2740
shmem_file_read_iter(struct kiocb * iocb,struct iov_iter * to)2741 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2742 {
2743 struct file *file = iocb->ki_filp;
2744 struct inode *inode = file_inode(file);
2745 struct address_space *mapping = inode->i_mapping;
2746 pgoff_t index;
2747 unsigned long offset;
2748 int error = 0;
2749 ssize_t retval = 0;
2750 loff_t *ppos = &iocb->ki_pos;
2751
2752 index = *ppos >> PAGE_SHIFT;
2753 offset = *ppos & ~PAGE_MASK;
2754
2755 for (;;) {
2756 struct folio *folio = NULL;
2757 struct page *page = NULL;
2758 pgoff_t end_index;
2759 unsigned long nr, ret;
2760 loff_t i_size = i_size_read(inode);
2761
2762 end_index = i_size >> PAGE_SHIFT;
2763 if (index > end_index)
2764 break;
2765 if (index == end_index) {
2766 nr = i_size & ~PAGE_MASK;
2767 if (nr <= offset)
2768 break;
2769 }
2770
2771 error = shmem_get_folio(inode, index, &folio, SGP_READ);
2772 if (error) {
2773 if (error == -EINVAL)
2774 error = 0;
2775 break;
2776 }
2777 if (folio) {
2778 folio_unlock(folio);
2779
2780 page = folio_file_page(folio, index);
2781 if (PageHWPoison(page)) {
2782 folio_put(folio);
2783 error = -EIO;
2784 break;
2785 }
2786 }
2787
2788 /*
2789 * We must evaluate after, since reads (unlike writes)
2790 * are called without i_rwsem protection against truncate
2791 */
2792 nr = PAGE_SIZE;
2793 i_size = i_size_read(inode);
2794 end_index = i_size >> PAGE_SHIFT;
2795 if (index == end_index) {
2796 nr = i_size & ~PAGE_MASK;
2797 if (nr <= offset) {
2798 if (folio)
2799 folio_put(folio);
2800 break;
2801 }
2802 }
2803 nr -= offset;
2804
2805 if (folio) {
2806 /*
2807 * If users can be writing to this page using arbitrary
2808 * virtual addresses, take care about potential aliasing
2809 * before reading the page on the kernel side.
2810 */
2811 if (mapping_writably_mapped(mapping))
2812 flush_dcache_page(page);
2813 /*
2814 * Mark the page accessed if we read the beginning.
2815 */
2816 if (!offset)
2817 folio_mark_accessed(folio);
2818 /*
2819 * Ok, we have the page, and it's up-to-date, so
2820 * now we can copy it to user space...
2821 */
2822 ret = copy_page_to_iter(page, offset, nr, to);
2823 folio_put(folio);
2824
2825 } else if (user_backed_iter(to)) {
2826 /*
2827 * Copy to user tends to be so well optimized, but
2828 * clear_user() not so much, that it is noticeably
2829 * faster to copy the zero page instead of clearing.
2830 */
2831 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2832 } else {
2833 /*
2834 * But submitting the same page twice in a row to
2835 * splice() - or others? - can result in confusion:
2836 * so don't attempt that optimization on pipes etc.
2837 */
2838 ret = iov_iter_zero(nr, to);
2839 }
2840
2841 retval += ret;
2842 offset += ret;
2843 index += offset >> PAGE_SHIFT;
2844 offset &= ~PAGE_MASK;
2845
2846 if (!iov_iter_count(to))
2847 break;
2848 if (ret < nr) {
2849 error = -EFAULT;
2850 break;
2851 }
2852 cond_resched();
2853 }
2854
2855 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2856 file_accessed(file);
2857 return retval ? retval : error;
2858 }
2859
shmem_file_write_iter(struct kiocb * iocb,struct iov_iter * from)2860 static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2861 {
2862 struct file *file = iocb->ki_filp;
2863 struct inode *inode = file->f_mapping->host;
2864 ssize_t ret;
2865
2866 inode_lock(inode);
2867 ret = generic_write_checks(iocb, from);
2868 if (ret <= 0)
2869 goto unlock;
2870 ret = file_remove_privs(file);
2871 if (ret)
2872 goto unlock;
2873 ret = file_update_time(file);
2874 if (ret)
2875 goto unlock;
2876 ret = generic_perform_write(iocb, from);
2877 unlock:
2878 inode_unlock(inode);
2879 return ret;
2880 }
2881
zero_pipe_buf_get(struct pipe_inode_info * pipe,struct pipe_buffer * buf)2882 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
2883 struct pipe_buffer *buf)
2884 {
2885 return true;
2886 }
2887
zero_pipe_buf_release(struct pipe_inode_info * pipe,struct pipe_buffer * buf)2888 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
2889 struct pipe_buffer *buf)
2890 {
2891 }
2892
zero_pipe_buf_try_steal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)2893 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
2894 struct pipe_buffer *buf)
2895 {
2896 return false;
2897 }
2898
2899 static const struct pipe_buf_operations zero_pipe_buf_ops = {
2900 .release = zero_pipe_buf_release,
2901 .try_steal = zero_pipe_buf_try_steal,
2902 .get = zero_pipe_buf_get,
2903 };
2904
splice_zeropage_into_pipe(struct pipe_inode_info * pipe,loff_t fpos,size_t size)2905 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
2906 loff_t fpos, size_t size)
2907 {
2908 size_t offset = fpos & ~PAGE_MASK;
2909
2910 size = min_t(size_t, size, PAGE_SIZE - offset);
2911
2912 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2913 struct pipe_buffer *buf = pipe_head_buf(pipe);
2914
2915 *buf = (struct pipe_buffer) {
2916 .ops = &zero_pipe_buf_ops,
2917 .page = ZERO_PAGE(0),
2918 .offset = offset,
2919 .len = size,
2920 };
2921 pipe->head++;
2922 }
2923
2924 return size;
2925 }
2926
shmem_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)2927 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2928 struct pipe_inode_info *pipe,
2929 size_t len, unsigned int flags)
2930 {
2931 struct inode *inode = file_inode(in);
2932 struct address_space *mapping = inode->i_mapping;
2933 struct folio *folio = NULL;
2934 size_t total_spliced = 0, used, npages, n, part;
2935 loff_t isize;
2936 int error = 0;
2937
2938 /* Work out how much data we can actually add into the pipe */
2939 used = pipe_occupancy(pipe->head, pipe->tail);
2940 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2941 len = min_t(size_t, len, npages * PAGE_SIZE);
2942
2943 do {
2944 if (*ppos >= i_size_read(inode))
2945 break;
2946
2947 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
2948 SGP_READ);
2949 if (error) {
2950 if (error == -EINVAL)
2951 error = 0;
2952 break;
2953 }
2954 if (folio) {
2955 folio_unlock(folio);
2956
2957 if (folio_test_hwpoison(folio) ||
2958 (folio_test_large(folio) &&
2959 folio_test_has_hwpoisoned(folio))) {
2960 error = -EIO;
2961 break;
2962 }
2963 }
2964
2965 /*
2966 * i_size must be checked after we know the pages are Uptodate.
2967 *
2968 * Checking i_size after the check allows us to calculate
2969 * the correct value for "nr", which means the zero-filled
2970 * part of the page is not copied back to userspace (unless
2971 * another truncate extends the file - this is desired though).
2972 */
2973 isize = i_size_read(inode);
2974 if (unlikely(*ppos >= isize))
2975 break;
2976 part = min_t(loff_t, isize - *ppos, len);
2977
2978 if (folio) {
2979 /*
2980 * If users can be writing to this page using arbitrary
2981 * virtual addresses, take care about potential aliasing
2982 * before reading the page on the kernel side.
2983 */
2984 if (mapping_writably_mapped(mapping))
2985 flush_dcache_folio(folio);
2986 folio_mark_accessed(folio);
2987 /*
2988 * Ok, we have the page, and it's up-to-date, so we can
2989 * now splice it into the pipe.
2990 */
2991 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
2992 folio_put(folio);
2993 folio = NULL;
2994 } else {
2995 n = splice_zeropage_into_pipe(pipe, *ppos, part);
2996 }
2997
2998 if (!n)
2999 break;
3000 len -= n;
3001 total_spliced += n;
3002 *ppos += n;
3003 in->f_ra.prev_pos = *ppos;
3004 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
3005 break;
3006
3007 cond_resched();
3008 } while (len);
3009
3010 if (folio)
3011 folio_put(folio);
3012
3013 file_accessed(in);
3014 return total_spliced ? total_spliced : error;
3015 }
3016
shmem_file_llseek(struct file * file,loff_t offset,int whence)3017 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
3018 {
3019 struct address_space *mapping = file->f_mapping;
3020 struct inode *inode = mapping->host;
3021
3022 if (whence != SEEK_DATA && whence != SEEK_HOLE)
3023 return generic_file_llseek_size(file, offset, whence,
3024 MAX_LFS_FILESIZE, i_size_read(inode));
3025 if (offset < 0)
3026 return -ENXIO;
3027
3028 inode_lock(inode);
3029 /* We're holding i_rwsem so we can access i_size directly */
3030 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
3031 if (offset >= 0)
3032 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
3033 inode_unlock(inode);
3034 return offset;
3035 }
3036
shmem_fallocate(struct file * file,int mode,loff_t offset,loff_t len)3037 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3038 loff_t len)
3039 {
3040 struct inode *inode = file_inode(file);
3041 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3042 struct shmem_inode_info *info = SHMEM_I(inode);
3043 struct shmem_falloc shmem_falloc;
3044 pgoff_t start, index, end, undo_fallocend;
3045 int error;
3046
3047 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3048 return -EOPNOTSUPP;
3049
3050 inode_lock(inode);
3051
3052 if (mode & FALLOC_FL_PUNCH_HOLE) {
3053 struct address_space *mapping = file->f_mapping;
3054 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3055 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
3056 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
3057
3058 /* protected by i_rwsem */
3059 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
3060 error = -EPERM;
3061 goto out;
3062 }
3063
3064 shmem_falloc.waitq = &shmem_falloc_waitq;
3065 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
3066 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3067 spin_lock(&inode->i_lock);
3068 inode->i_private = &shmem_falloc;
3069 spin_unlock(&inode->i_lock);
3070
3071 if ((u64)unmap_end > (u64)unmap_start)
3072 unmap_mapping_range(mapping, unmap_start,
3073 1 + unmap_end - unmap_start, 0);
3074 shmem_truncate_range(inode, offset, offset + len - 1);
3075 /* No need to unmap again: hole-punching leaves COWed pages */
3076
3077 spin_lock(&inode->i_lock);
3078 inode->i_private = NULL;
3079 wake_up_all(&shmem_falloc_waitq);
3080 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
3081 spin_unlock(&inode->i_lock);
3082 error = 0;
3083 goto out;
3084 }
3085
3086 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3087 error = inode_newsize_ok(inode, offset + len);
3088 if (error)
3089 goto out;
3090
3091 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3092 error = -EPERM;
3093 goto out;
3094 }
3095
3096 start = offset >> PAGE_SHIFT;
3097 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
3098 /* Try to avoid a swapstorm if len is impossible to satisfy */
3099 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3100 error = -ENOSPC;
3101 goto out;
3102 }
3103
3104 shmem_falloc.waitq = NULL;
3105 shmem_falloc.start = start;
3106 shmem_falloc.next = start;
3107 shmem_falloc.nr_falloced = 0;
3108 shmem_falloc.nr_unswapped = 0;
3109 spin_lock(&inode->i_lock);
3110 inode->i_private = &shmem_falloc;
3111 spin_unlock(&inode->i_lock);
3112
3113 /*
3114 * info->fallocend is only relevant when huge pages might be
3115 * involved: to prevent split_huge_page() freeing fallocated
3116 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3117 */
3118 undo_fallocend = info->fallocend;
3119 if (info->fallocend < end)
3120 info->fallocend = end;
3121
3122 for (index = start; index < end; ) {
3123 struct folio *folio;
3124
3125 /*
3126 * Good, the fallocate(2) manpage permits EINTR: we may have
3127 * been interrupted because we are using up too much memory.
3128 */
3129 if (signal_pending(current))
3130 error = -EINTR;
3131 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3132 error = -ENOMEM;
3133 else
3134 error = shmem_get_folio(inode, index, &folio,
3135 SGP_FALLOC);
3136 if (error) {
3137 info->fallocend = undo_fallocend;
3138 /* Remove the !uptodate folios we added */
3139 if (index > start) {
3140 shmem_undo_range(inode,
3141 (loff_t)start << PAGE_SHIFT,
3142 ((loff_t)index << PAGE_SHIFT) - 1, true);
3143 }
3144 goto undone;
3145 }
3146
3147 /*
3148 * Here is a more important optimization than it appears:
3149 * a second SGP_FALLOC on the same large folio will clear it,
3150 * making it uptodate and un-undoable if we fail later.
3151 */
3152 index = folio_next_index(folio);
3153 /* Beware 32-bit wraparound */
3154 if (!index)
3155 index--;
3156
3157 /*
3158 * Inform shmem_writepage() how far we have reached.
3159 * No need for lock or barrier: we have the page lock.
3160 */
3161 if (!folio_test_uptodate(folio))
3162 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3163 shmem_falloc.next = index;
3164
3165 /*
3166 * If !uptodate, leave it that way so that freeable folios
3167 * can be recognized if we need to rollback on error later.
3168 * But mark it dirty so that memory pressure will swap rather
3169 * than free the folios we are allocating (and SGP_CACHE folios
3170 * might still be clean: we now need to mark those dirty too).
3171 */
3172 folio_mark_dirty(folio);
3173 folio_unlock(folio);
3174 folio_put(folio);
3175 cond_resched();
3176 }
3177
3178 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3179 i_size_write(inode, offset + len);
3180 undone:
3181 spin_lock(&inode->i_lock);
3182 inode->i_private = NULL;
3183 spin_unlock(&inode->i_lock);
3184 out:
3185 if (!error)
3186 file_modified(file);
3187 inode_unlock(inode);
3188 return error;
3189 }
3190
shmem_statfs(struct dentry * dentry,struct kstatfs * buf)3191 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3192 {
3193 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3194
3195 buf->f_type = TMPFS_MAGIC;
3196 buf->f_bsize = PAGE_SIZE;
3197 buf->f_namelen = NAME_MAX;
3198 if (sbinfo->max_blocks) {
3199 buf->f_blocks = sbinfo->max_blocks;
3200 buf->f_bavail =
3201 buf->f_bfree = sbinfo->max_blocks -
3202 percpu_counter_sum(&sbinfo->used_blocks);
3203 }
3204 if (sbinfo->max_inodes) {
3205 buf->f_files = sbinfo->max_inodes;
3206 buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
3207 }
3208 /* else leave those fields 0 like simple_statfs */
3209
3210 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3211
3212 return 0;
3213 }
3214
3215 /*
3216 * File creation. Allocate an inode, and we're done..
3217 */
3218 static int
shmem_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t dev)3219 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3220 struct dentry *dentry, umode_t mode, dev_t dev)
3221 {
3222 struct inode *inode;
3223 int error;
3224
3225 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3226 if (IS_ERR(inode))
3227 return PTR_ERR(inode);
3228
3229 error = simple_acl_create(dir, inode);
3230 if (error)
3231 goto out_iput;
3232 error = security_inode_init_security(inode, dir,
3233 &dentry->d_name,
3234 shmem_initxattrs, NULL);
3235 if (error && error != -EOPNOTSUPP)
3236 goto out_iput;
3237
3238 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3239 if (error)
3240 goto out_iput;
3241
3242 dir->i_size += BOGO_DIRENT_SIZE;
3243 dir->i_mtime = inode_set_ctime_current(dir);
3244 inode_inc_iversion(dir);
3245 d_instantiate(dentry, inode);
3246 dget(dentry); /* Extra count - pin the dentry in core */
3247 return error;
3248
3249 out_iput:
3250 iput(inode);
3251 return error;
3252 }
3253
3254 static int
shmem_tmpfile(struct mnt_idmap * idmap,struct inode * dir,struct file * file,umode_t mode)3255 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3256 struct file *file, umode_t mode)
3257 {
3258 struct inode *inode;
3259 int error;
3260
3261 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3262
3263 if (IS_ERR(inode)) {
3264 error = PTR_ERR(inode);
3265 goto err_out;
3266 }
3267
3268 error = security_inode_init_security(inode, dir,
3269 NULL,
3270 shmem_initxattrs, NULL);
3271 if (error && error != -EOPNOTSUPP)
3272 goto out_iput;
3273 error = simple_acl_create(dir, inode);
3274 if (error)
3275 goto out_iput;
3276 d_tmpfile(file, inode);
3277
3278 err_out:
3279 return finish_open_simple(file, error);
3280 out_iput:
3281 iput(inode);
3282 return error;
3283 }
3284
shmem_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)3285 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3286 struct dentry *dentry, umode_t mode)
3287 {
3288 int error;
3289
3290 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3291 if (error)
3292 return error;
3293 inc_nlink(dir);
3294 return 0;
3295 }
3296
shmem_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)3297 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3298 struct dentry *dentry, umode_t mode, bool excl)
3299 {
3300 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3301 }
3302
3303 /*
3304 * Link a file..
3305 */
shmem_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)3306 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3307 {
3308 struct inode *inode = d_inode(old_dentry);
3309 int ret = 0;
3310
3311 /*
3312 * No ordinary (disk based) filesystem counts links as inodes;
3313 * but each new link needs a new dentry, pinning lowmem, and
3314 * tmpfs dentries cannot be pruned until they are unlinked.
3315 * But if an O_TMPFILE file is linked into the tmpfs, the
3316 * first link must skip that, to get the accounting right.
3317 */
3318 if (inode->i_nlink) {
3319 ret = shmem_reserve_inode(inode->i_sb, NULL);
3320 if (ret)
3321 goto out;
3322 }
3323
3324 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3325 if (ret) {
3326 if (inode->i_nlink)
3327 shmem_free_inode(inode->i_sb, 0);
3328 goto out;
3329 }
3330
3331 dir->i_size += BOGO_DIRENT_SIZE;
3332 dir->i_mtime = inode_set_ctime_to_ts(dir,
3333 inode_set_ctime_current(inode));
3334 inode_inc_iversion(dir);
3335 inc_nlink(inode);
3336 ihold(inode); /* New dentry reference */
3337 dget(dentry); /* Extra pinning count for the created dentry */
3338 d_instantiate(dentry, inode);
3339 out:
3340 return ret;
3341 }
3342
shmem_unlink(struct inode * dir,struct dentry * dentry)3343 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3344 {
3345 struct inode *inode = d_inode(dentry);
3346
3347 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3348 shmem_free_inode(inode->i_sb, 0);
3349
3350 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3351
3352 dir->i_size -= BOGO_DIRENT_SIZE;
3353 dir->i_mtime = inode_set_ctime_to_ts(dir,
3354 inode_set_ctime_current(inode));
3355 inode_inc_iversion(dir);
3356 drop_nlink(inode);
3357 dput(dentry); /* Undo the count from "create" - this does all the work */
3358 return 0;
3359 }
3360
shmem_rmdir(struct inode * dir,struct dentry * dentry)3361 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3362 {
3363 if (!simple_empty(dentry))
3364 return -ENOTEMPTY;
3365
3366 drop_nlink(d_inode(dentry));
3367 drop_nlink(dir);
3368 return shmem_unlink(dir, dentry);
3369 }
3370
shmem_whiteout(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry)3371 static int shmem_whiteout(struct mnt_idmap *idmap,
3372 struct inode *old_dir, struct dentry *old_dentry)
3373 {
3374 struct dentry *whiteout;
3375 int error;
3376
3377 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3378 if (!whiteout)
3379 return -ENOMEM;
3380
3381 error = shmem_mknod(idmap, old_dir, whiteout,
3382 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3383 dput(whiteout);
3384 if (error)
3385 return error;
3386
3387 /*
3388 * Cheat and hash the whiteout while the old dentry is still in
3389 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3390 *
3391 * d_lookup() will consistently find one of them at this point,
3392 * not sure which one, but that isn't even important.
3393 */
3394 d_rehash(whiteout);
3395 return 0;
3396 }
3397
3398 /*
3399 * The VFS layer already does all the dentry stuff for rename,
3400 * we just have to decrement the usage count for the target if
3401 * it exists so that the VFS layer correctly free's it when it
3402 * gets overwritten.
3403 */
shmem_rename2(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)3404 static int shmem_rename2(struct mnt_idmap *idmap,
3405 struct inode *old_dir, struct dentry *old_dentry,
3406 struct inode *new_dir, struct dentry *new_dentry,
3407 unsigned int flags)
3408 {
3409 struct inode *inode = d_inode(old_dentry);
3410 int they_are_dirs = S_ISDIR(inode->i_mode);
3411 int error;
3412
3413 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3414 return -EINVAL;
3415
3416 if (flags & RENAME_EXCHANGE)
3417 return simple_offset_rename_exchange(old_dir, old_dentry,
3418 new_dir, new_dentry);
3419
3420 if (!simple_empty(new_dentry))
3421 return -ENOTEMPTY;
3422
3423 if (flags & RENAME_WHITEOUT) {
3424 error = shmem_whiteout(idmap, old_dir, old_dentry);
3425 if (error)
3426 return error;
3427 }
3428
3429 simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry);
3430 error = simple_offset_add(shmem_get_offset_ctx(new_dir), old_dentry);
3431 if (error)
3432 return error;
3433
3434 if (d_really_is_positive(new_dentry)) {
3435 (void) shmem_unlink(new_dir, new_dentry);
3436 if (they_are_dirs) {
3437 drop_nlink(d_inode(new_dentry));
3438 drop_nlink(old_dir);
3439 }
3440 } else if (they_are_dirs) {
3441 drop_nlink(old_dir);
3442 inc_nlink(new_dir);
3443 }
3444
3445 old_dir->i_size -= BOGO_DIRENT_SIZE;
3446 new_dir->i_size += BOGO_DIRENT_SIZE;
3447 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
3448 inode_inc_iversion(old_dir);
3449 inode_inc_iversion(new_dir);
3450 return 0;
3451 }
3452
shmem_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)3453 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3454 struct dentry *dentry, const char *symname)
3455 {
3456 int error;
3457 int len;
3458 struct inode *inode;
3459 struct folio *folio;
3460
3461 len = strlen(symname) + 1;
3462 if (len > PAGE_SIZE)
3463 return -ENAMETOOLONG;
3464
3465 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3466 VM_NORESERVE);
3467
3468 if (IS_ERR(inode))
3469 return PTR_ERR(inode);
3470
3471 error = security_inode_init_security(inode, dir, &dentry->d_name,
3472 shmem_initxattrs, NULL);
3473 if (error && error != -EOPNOTSUPP)
3474 goto out_iput;
3475
3476 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3477 if (error)
3478 goto out_iput;
3479
3480 inode->i_size = len-1;
3481 if (len <= SHORT_SYMLINK_LEN) {
3482 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3483 if (!inode->i_link) {
3484 error = -ENOMEM;
3485 goto out_remove_offset;
3486 }
3487 inode->i_op = &shmem_short_symlink_operations;
3488 } else {
3489 inode_nohighmem(inode);
3490 error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3491 if (error)
3492 goto out_remove_offset;
3493 inode->i_mapping->a_ops = &shmem_aops;
3494 inode->i_op = &shmem_symlink_inode_operations;
3495 memcpy(folio_address(folio), symname, len);
3496 folio_mark_uptodate(folio);
3497 folio_mark_dirty(folio);
3498 folio_unlock(folio);
3499 folio_put(folio);
3500 }
3501 dir->i_size += BOGO_DIRENT_SIZE;
3502 dir->i_mtime = inode_set_ctime_current(dir);
3503 inode_inc_iversion(dir);
3504 d_instantiate(dentry, inode);
3505 dget(dentry);
3506 return 0;
3507
3508 out_remove_offset:
3509 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3510 out_iput:
3511 iput(inode);
3512 return error;
3513 }
3514
shmem_put_link(void * arg)3515 static void shmem_put_link(void *arg)
3516 {
3517 folio_mark_accessed(arg);
3518 folio_put(arg);
3519 }
3520
shmem_get_link(struct dentry * dentry,struct inode * inode,struct delayed_call * done)3521 static const char *shmem_get_link(struct dentry *dentry,
3522 struct inode *inode,
3523 struct delayed_call *done)
3524 {
3525 struct folio *folio = NULL;
3526 int error;
3527
3528 if (!dentry) {
3529 folio = filemap_get_folio(inode->i_mapping, 0);
3530 if (IS_ERR(folio))
3531 return ERR_PTR(-ECHILD);
3532 if (PageHWPoison(folio_page(folio, 0)) ||
3533 !folio_test_uptodate(folio)) {
3534 folio_put(folio);
3535 return ERR_PTR(-ECHILD);
3536 }
3537 } else {
3538 error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3539 if (error)
3540 return ERR_PTR(error);
3541 if (!folio)
3542 return ERR_PTR(-ECHILD);
3543 if (PageHWPoison(folio_page(folio, 0))) {
3544 folio_unlock(folio);
3545 folio_put(folio);
3546 return ERR_PTR(-ECHILD);
3547 }
3548 folio_unlock(folio);
3549 }
3550 set_delayed_call(done, shmem_put_link, folio);
3551 return folio_address(folio);
3552 }
3553
3554 #ifdef CONFIG_TMPFS_XATTR
3555
shmem_fileattr_get(struct dentry * dentry,struct fileattr * fa)3556 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3557 {
3558 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3559
3560 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3561
3562 return 0;
3563 }
3564
shmem_fileattr_set(struct mnt_idmap * idmap,struct dentry * dentry,struct fileattr * fa)3565 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3566 struct dentry *dentry, struct fileattr *fa)
3567 {
3568 struct inode *inode = d_inode(dentry);
3569 struct shmem_inode_info *info = SHMEM_I(inode);
3570
3571 if (fileattr_has_fsx(fa))
3572 return -EOPNOTSUPP;
3573 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3574 return -EOPNOTSUPP;
3575
3576 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3577 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3578
3579 shmem_set_inode_flags(inode, info->fsflags);
3580 inode_set_ctime_current(inode);
3581 inode_inc_iversion(inode);
3582 return 0;
3583 }
3584
3585 /*
3586 * Superblocks without xattr inode operations may get some security.* xattr
3587 * support from the LSM "for free". As soon as we have any other xattrs
3588 * like ACLs, we also need to implement the security.* handlers at
3589 * filesystem level, though.
3590 */
3591
3592 /*
3593 * Callback for security_inode_init_security() for acquiring xattrs.
3594 */
shmem_initxattrs(struct inode * inode,const struct xattr * xattr_array,void * fs_info)3595 static int shmem_initxattrs(struct inode *inode,
3596 const struct xattr *xattr_array,
3597 void *fs_info)
3598 {
3599 struct shmem_inode_info *info = SHMEM_I(inode);
3600 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3601 const struct xattr *xattr;
3602 struct simple_xattr *new_xattr;
3603 size_t ispace = 0;
3604 size_t len;
3605
3606 if (sbinfo->max_inodes) {
3607 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3608 ispace += simple_xattr_space(xattr->name,
3609 xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
3610 }
3611 if (ispace) {
3612 raw_spin_lock(&sbinfo->stat_lock);
3613 if (sbinfo->free_ispace < ispace)
3614 ispace = 0;
3615 else
3616 sbinfo->free_ispace -= ispace;
3617 raw_spin_unlock(&sbinfo->stat_lock);
3618 if (!ispace)
3619 return -ENOSPC;
3620 }
3621 }
3622
3623 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3624 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3625 if (!new_xattr)
3626 break;
3627
3628 len = strlen(xattr->name) + 1;
3629 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3630 GFP_KERNEL_ACCOUNT);
3631 if (!new_xattr->name) {
3632 kvfree(new_xattr);
3633 break;
3634 }
3635
3636 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3637 XATTR_SECURITY_PREFIX_LEN);
3638 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3639 xattr->name, len);
3640
3641 simple_xattr_add(&info->xattrs, new_xattr);
3642 }
3643
3644 if (xattr->name != NULL) {
3645 if (ispace) {
3646 raw_spin_lock(&sbinfo->stat_lock);
3647 sbinfo->free_ispace += ispace;
3648 raw_spin_unlock(&sbinfo->stat_lock);
3649 }
3650 simple_xattrs_free(&info->xattrs, NULL);
3651 return -ENOMEM;
3652 }
3653
3654 return 0;
3655 }
3656
shmem_xattr_handler_get(const struct xattr_handler * handler,struct dentry * unused,struct inode * inode,const char * name,void * buffer,size_t size)3657 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3658 struct dentry *unused, struct inode *inode,
3659 const char *name, void *buffer, size_t size)
3660 {
3661 struct shmem_inode_info *info = SHMEM_I(inode);
3662
3663 name = xattr_full_name(handler, name);
3664 return simple_xattr_get(&info->xattrs, name, buffer, size);
3665 }
3666
shmem_xattr_handler_set(const struct xattr_handler * handler,struct mnt_idmap * idmap,struct dentry * unused,struct inode * inode,const char * name,const void * value,size_t size,int flags)3667 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3668 struct mnt_idmap *idmap,
3669 struct dentry *unused, struct inode *inode,
3670 const char *name, const void *value,
3671 size_t size, int flags)
3672 {
3673 struct shmem_inode_info *info = SHMEM_I(inode);
3674 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3675 struct simple_xattr *old_xattr;
3676 size_t ispace = 0;
3677
3678 name = xattr_full_name(handler, name);
3679 if (value && sbinfo->max_inodes) {
3680 ispace = simple_xattr_space(name, size);
3681 raw_spin_lock(&sbinfo->stat_lock);
3682 if (sbinfo->free_ispace < ispace)
3683 ispace = 0;
3684 else
3685 sbinfo->free_ispace -= ispace;
3686 raw_spin_unlock(&sbinfo->stat_lock);
3687 if (!ispace)
3688 return -ENOSPC;
3689 }
3690
3691 old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
3692 if (!IS_ERR(old_xattr)) {
3693 ispace = 0;
3694 if (old_xattr && sbinfo->max_inodes)
3695 ispace = simple_xattr_space(old_xattr->name,
3696 old_xattr->size);
3697 simple_xattr_free(old_xattr);
3698 old_xattr = NULL;
3699 inode_set_ctime_current(inode);
3700 inode_inc_iversion(inode);
3701 }
3702 if (ispace) {
3703 raw_spin_lock(&sbinfo->stat_lock);
3704 sbinfo->free_ispace += ispace;
3705 raw_spin_unlock(&sbinfo->stat_lock);
3706 }
3707 return PTR_ERR(old_xattr);
3708 }
3709
3710 static const struct xattr_handler shmem_security_xattr_handler = {
3711 .prefix = XATTR_SECURITY_PREFIX,
3712 .get = shmem_xattr_handler_get,
3713 .set = shmem_xattr_handler_set,
3714 };
3715
3716 static const struct xattr_handler shmem_trusted_xattr_handler = {
3717 .prefix = XATTR_TRUSTED_PREFIX,
3718 .get = shmem_xattr_handler_get,
3719 .set = shmem_xattr_handler_set,
3720 };
3721
3722 static const struct xattr_handler shmem_user_xattr_handler = {
3723 .prefix = XATTR_USER_PREFIX,
3724 .get = shmem_xattr_handler_get,
3725 .set = shmem_xattr_handler_set,
3726 };
3727
3728 static const struct xattr_handler *shmem_xattr_handlers[] = {
3729 &shmem_security_xattr_handler,
3730 &shmem_trusted_xattr_handler,
3731 &shmem_user_xattr_handler,
3732 NULL
3733 };
3734
shmem_listxattr(struct dentry * dentry,char * buffer,size_t size)3735 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3736 {
3737 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3738 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3739 }
3740 #endif /* CONFIG_TMPFS_XATTR */
3741
3742 static const struct inode_operations shmem_short_symlink_operations = {
3743 .getattr = shmem_getattr,
3744 .setattr = shmem_setattr,
3745 .get_link = simple_get_link,
3746 #ifdef CONFIG_TMPFS_XATTR
3747 .listxattr = shmem_listxattr,
3748 #endif
3749 };
3750
3751 static const struct inode_operations shmem_symlink_inode_operations = {
3752 .getattr = shmem_getattr,
3753 .setattr = shmem_setattr,
3754 .get_link = shmem_get_link,
3755 #ifdef CONFIG_TMPFS_XATTR
3756 .listxattr = shmem_listxattr,
3757 #endif
3758 };
3759
shmem_get_parent(struct dentry * child)3760 static struct dentry *shmem_get_parent(struct dentry *child)
3761 {
3762 return ERR_PTR(-ESTALE);
3763 }
3764
shmem_match(struct inode * ino,void * vfh)3765 static int shmem_match(struct inode *ino, void *vfh)
3766 {
3767 __u32 *fh = vfh;
3768 __u64 inum = fh[2];
3769 inum = (inum << 32) | fh[1];
3770 return ino->i_ino == inum && fh[0] == ino->i_generation;
3771 }
3772
3773 /* Find any alias of inode, but prefer a hashed alias */
shmem_find_alias(struct inode * inode)3774 static struct dentry *shmem_find_alias(struct inode *inode)
3775 {
3776 struct dentry *alias = d_find_alias(inode);
3777
3778 return alias ?: d_find_any_alias(inode);
3779 }
3780
3781
shmem_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)3782 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3783 struct fid *fid, int fh_len, int fh_type)
3784 {
3785 struct inode *inode;
3786 struct dentry *dentry = NULL;
3787 u64 inum;
3788
3789 if (fh_len < 3)
3790 return NULL;
3791
3792 inum = fid->raw[2];
3793 inum = (inum << 32) | fid->raw[1];
3794
3795 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3796 shmem_match, fid->raw);
3797 if (inode) {
3798 dentry = shmem_find_alias(inode);
3799 iput(inode);
3800 }
3801
3802 return dentry;
3803 }
3804
shmem_encode_fh(struct inode * inode,__u32 * fh,int * len,struct inode * parent)3805 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3806 struct inode *parent)
3807 {
3808 if (*len < 3) {
3809 *len = 3;
3810 return FILEID_INVALID;
3811 }
3812
3813 if (inode_unhashed(inode)) {
3814 /* Unfortunately insert_inode_hash is not idempotent,
3815 * so as we hash inodes here rather than at creation
3816 * time, we need a lock to ensure we only try
3817 * to do it once
3818 */
3819 static DEFINE_SPINLOCK(lock);
3820 spin_lock(&lock);
3821 if (inode_unhashed(inode))
3822 __insert_inode_hash(inode,
3823 inode->i_ino + inode->i_generation);
3824 spin_unlock(&lock);
3825 }
3826
3827 fh[0] = inode->i_generation;
3828 fh[1] = inode->i_ino;
3829 fh[2] = ((__u64)inode->i_ino) >> 32;
3830
3831 *len = 3;
3832 return 1;
3833 }
3834
3835 static const struct export_operations shmem_export_ops = {
3836 .get_parent = shmem_get_parent,
3837 .encode_fh = shmem_encode_fh,
3838 .fh_to_dentry = shmem_fh_to_dentry,
3839 };
3840
3841 enum shmem_param {
3842 Opt_gid,
3843 Opt_huge,
3844 Opt_mode,
3845 Opt_mpol,
3846 Opt_nr_blocks,
3847 Opt_nr_inodes,
3848 Opt_size,
3849 Opt_uid,
3850 Opt_inode32,
3851 Opt_inode64,
3852 Opt_noswap,
3853 Opt_quota,
3854 Opt_usrquota,
3855 Opt_grpquota,
3856 Opt_usrquota_block_hardlimit,
3857 Opt_usrquota_inode_hardlimit,
3858 Opt_grpquota_block_hardlimit,
3859 Opt_grpquota_inode_hardlimit,
3860 };
3861
3862 static const struct constant_table shmem_param_enums_huge[] = {
3863 {"never", SHMEM_HUGE_NEVER },
3864 {"always", SHMEM_HUGE_ALWAYS },
3865 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3866 {"advise", SHMEM_HUGE_ADVISE },
3867 {}
3868 };
3869
3870 const struct fs_parameter_spec shmem_fs_parameters[] = {
3871 fsparam_u32 ("gid", Opt_gid),
3872 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3873 fsparam_u32oct("mode", Opt_mode),
3874 fsparam_string("mpol", Opt_mpol),
3875 fsparam_string("nr_blocks", Opt_nr_blocks),
3876 fsparam_string("nr_inodes", Opt_nr_inodes),
3877 fsparam_string("size", Opt_size),
3878 fsparam_u32 ("uid", Opt_uid),
3879 fsparam_flag ("inode32", Opt_inode32),
3880 fsparam_flag ("inode64", Opt_inode64),
3881 fsparam_flag ("noswap", Opt_noswap),
3882 #ifdef CONFIG_TMPFS_QUOTA
3883 fsparam_flag ("quota", Opt_quota),
3884 fsparam_flag ("usrquota", Opt_usrquota),
3885 fsparam_flag ("grpquota", Opt_grpquota),
3886 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
3887 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
3888 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
3889 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
3890 #endif
3891 {}
3892 };
3893
shmem_parse_one(struct fs_context * fc,struct fs_parameter * param)3894 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3895 {
3896 struct shmem_options *ctx = fc->fs_private;
3897 struct fs_parse_result result;
3898 unsigned long long size;
3899 char *rest;
3900 int opt;
3901 kuid_t kuid;
3902 kgid_t kgid;
3903
3904 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3905 if (opt < 0)
3906 return opt;
3907
3908 switch (opt) {
3909 case Opt_size:
3910 size = memparse(param->string, &rest);
3911 if (*rest == '%') {
3912 size <<= PAGE_SHIFT;
3913 size *= totalram_pages();
3914 do_div(size, 100);
3915 rest++;
3916 }
3917 if (*rest)
3918 goto bad_value;
3919 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3920 ctx->seen |= SHMEM_SEEN_BLOCKS;
3921 break;
3922 case Opt_nr_blocks:
3923 ctx->blocks = memparse(param->string, &rest);
3924 if (*rest || ctx->blocks > LONG_MAX)
3925 goto bad_value;
3926 ctx->seen |= SHMEM_SEEN_BLOCKS;
3927 break;
3928 case Opt_nr_inodes:
3929 ctx->inodes = memparse(param->string, &rest);
3930 if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
3931 goto bad_value;
3932 ctx->seen |= SHMEM_SEEN_INODES;
3933 break;
3934 case Opt_mode:
3935 ctx->mode = result.uint_32 & 07777;
3936 break;
3937 case Opt_uid:
3938 kuid = make_kuid(current_user_ns(), result.uint_32);
3939 if (!uid_valid(kuid))
3940 goto bad_value;
3941
3942 /*
3943 * The requested uid must be representable in the
3944 * filesystem's idmapping.
3945 */
3946 if (!kuid_has_mapping(fc->user_ns, kuid))
3947 goto bad_value;
3948
3949 ctx->uid = kuid;
3950 break;
3951 case Opt_gid:
3952 kgid = make_kgid(current_user_ns(), result.uint_32);
3953 if (!gid_valid(kgid))
3954 goto bad_value;
3955
3956 /*
3957 * The requested gid must be representable in the
3958 * filesystem's idmapping.
3959 */
3960 if (!kgid_has_mapping(fc->user_ns, kgid))
3961 goto bad_value;
3962
3963 ctx->gid = kgid;
3964 break;
3965 case Opt_huge:
3966 ctx->huge = result.uint_32;
3967 if (ctx->huge != SHMEM_HUGE_NEVER &&
3968 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3969 has_transparent_hugepage()))
3970 goto unsupported_parameter;
3971 ctx->seen |= SHMEM_SEEN_HUGE;
3972 break;
3973 case Opt_mpol:
3974 if (IS_ENABLED(CONFIG_NUMA)) {
3975 mpol_put(ctx->mpol);
3976 ctx->mpol = NULL;
3977 if (mpol_parse_str(param->string, &ctx->mpol))
3978 goto bad_value;
3979 break;
3980 }
3981 goto unsupported_parameter;
3982 case Opt_inode32:
3983 ctx->full_inums = false;
3984 ctx->seen |= SHMEM_SEEN_INUMS;
3985 break;
3986 case Opt_inode64:
3987 if (sizeof(ino_t) < 8) {
3988 return invalfc(fc,
3989 "Cannot use inode64 with <64bit inums in kernel\n");
3990 }
3991 ctx->full_inums = true;
3992 ctx->seen |= SHMEM_SEEN_INUMS;
3993 break;
3994 case Opt_noswap:
3995 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
3996 return invalfc(fc,
3997 "Turning off swap in unprivileged tmpfs mounts unsupported");
3998 }
3999 ctx->noswap = true;
4000 ctx->seen |= SHMEM_SEEN_NOSWAP;
4001 break;
4002 case Opt_quota:
4003 if (fc->user_ns != &init_user_ns)
4004 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4005 ctx->seen |= SHMEM_SEEN_QUOTA;
4006 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
4007 break;
4008 case Opt_usrquota:
4009 if (fc->user_ns != &init_user_ns)
4010 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4011 ctx->seen |= SHMEM_SEEN_QUOTA;
4012 ctx->quota_types |= QTYPE_MASK_USR;
4013 break;
4014 case Opt_grpquota:
4015 if (fc->user_ns != &init_user_ns)
4016 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4017 ctx->seen |= SHMEM_SEEN_QUOTA;
4018 ctx->quota_types |= QTYPE_MASK_GRP;
4019 break;
4020 case Opt_usrquota_block_hardlimit:
4021 size = memparse(param->string, &rest);
4022 if (*rest || !size)
4023 goto bad_value;
4024 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4025 return invalfc(fc,
4026 "User quota block hardlimit too large.");
4027 ctx->qlimits.usrquota_bhardlimit = size;
4028 break;
4029 case Opt_grpquota_block_hardlimit:
4030 size = memparse(param->string, &rest);
4031 if (*rest || !size)
4032 goto bad_value;
4033 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4034 return invalfc(fc,
4035 "Group quota block hardlimit too large.");
4036 ctx->qlimits.grpquota_bhardlimit = size;
4037 break;
4038 case Opt_usrquota_inode_hardlimit:
4039 size = memparse(param->string, &rest);
4040 if (*rest || !size)
4041 goto bad_value;
4042 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4043 return invalfc(fc,
4044 "User quota inode hardlimit too large.");
4045 ctx->qlimits.usrquota_ihardlimit = size;
4046 break;
4047 case Opt_grpquota_inode_hardlimit:
4048 size = memparse(param->string, &rest);
4049 if (*rest || !size)
4050 goto bad_value;
4051 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4052 return invalfc(fc,
4053 "Group quota inode hardlimit too large.");
4054 ctx->qlimits.grpquota_ihardlimit = size;
4055 break;
4056 }
4057 return 0;
4058
4059 unsupported_parameter:
4060 return invalfc(fc, "Unsupported parameter '%s'", param->key);
4061 bad_value:
4062 return invalfc(fc, "Bad value for '%s'", param->key);
4063 }
4064
shmem_parse_options(struct fs_context * fc,void * data)4065 static int shmem_parse_options(struct fs_context *fc, void *data)
4066 {
4067 char *options = data;
4068
4069 if (options) {
4070 int err = security_sb_eat_lsm_opts(options, &fc->security);
4071 if (err)
4072 return err;
4073 }
4074
4075 while (options != NULL) {
4076 char *this_char = options;
4077 for (;;) {
4078 /*
4079 * NUL-terminate this option: unfortunately,
4080 * mount options form a comma-separated list,
4081 * but mpol's nodelist may also contain commas.
4082 */
4083 options = strchr(options, ',');
4084 if (options == NULL)
4085 break;
4086 options++;
4087 if (!isdigit(*options)) {
4088 options[-1] = '\0';
4089 break;
4090 }
4091 }
4092 if (*this_char) {
4093 char *value = strchr(this_char, '=');
4094 size_t len = 0;
4095 int err;
4096
4097 if (value) {
4098 *value++ = '\0';
4099 len = strlen(value);
4100 }
4101 err = vfs_parse_fs_string(fc, this_char, value, len);
4102 if (err < 0)
4103 return err;
4104 }
4105 }
4106 return 0;
4107 }
4108
4109 /*
4110 * Reconfigure a shmem filesystem.
4111 */
shmem_reconfigure(struct fs_context * fc)4112 static int shmem_reconfigure(struct fs_context *fc)
4113 {
4114 struct shmem_options *ctx = fc->fs_private;
4115 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
4116 unsigned long used_isp;
4117 struct mempolicy *mpol = NULL;
4118 const char *err;
4119
4120 raw_spin_lock(&sbinfo->stat_lock);
4121 used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
4122
4123 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4124 if (!sbinfo->max_blocks) {
4125 err = "Cannot retroactively limit size";
4126 goto out;
4127 }
4128 if (percpu_counter_compare(&sbinfo->used_blocks,
4129 ctx->blocks) > 0) {
4130 err = "Too small a size for current use";
4131 goto out;
4132 }
4133 }
4134 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4135 if (!sbinfo->max_inodes) {
4136 err = "Cannot retroactively limit inodes";
4137 goto out;
4138 }
4139 if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
4140 err = "Too few inodes for current use";
4141 goto out;
4142 }
4143 }
4144
4145 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4146 sbinfo->next_ino > UINT_MAX) {
4147 err = "Current inum too high to switch to 32-bit inums";
4148 goto out;
4149 }
4150 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4151 err = "Cannot disable swap on remount";
4152 goto out;
4153 }
4154 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4155 err = "Cannot enable swap on remount if it was disabled on first mount";
4156 goto out;
4157 }
4158
4159 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4160 !sb_any_quota_loaded(fc->root->d_sb)) {
4161 err = "Cannot enable quota on remount";
4162 goto out;
4163 }
4164
4165 #ifdef CONFIG_TMPFS_QUOTA
4166 #define CHANGED_LIMIT(name) \
4167 (ctx->qlimits.name## hardlimit && \
4168 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4169
4170 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4171 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4172 err = "Cannot change global quota limit on remount";
4173 goto out;
4174 }
4175 #endif /* CONFIG_TMPFS_QUOTA */
4176
4177 if (ctx->seen & SHMEM_SEEN_HUGE)
4178 sbinfo->huge = ctx->huge;
4179 if (ctx->seen & SHMEM_SEEN_INUMS)
4180 sbinfo->full_inums = ctx->full_inums;
4181 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4182 sbinfo->max_blocks = ctx->blocks;
4183 if (ctx->seen & SHMEM_SEEN_INODES) {
4184 sbinfo->max_inodes = ctx->inodes;
4185 sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
4186 }
4187
4188 /*
4189 * Preserve previous mempolicy unless mpol remount option was specified.
4190 */
4191 if (ctx->mpol) {
4192 mpol = sbinfo->mpol;
4193 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4194 ctx->mpol = NULL;
4195 }
4196
4197 if (ctx->noswap)
4198 sbinfo->noswap = true;
4199
4200 raw_spin_unlock(&sbinfo->stat_lock);
4201 mpol_put(mpol);
4202 return 0;
4203 out:
4204 raw_spin_unlock(&sbinfo->stat_lock);
4205 return invalfc(fc, "%s", err);
4206 }
4207
shmem_show_options(struct seq_file * seq,struct dentry * root)4208 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
4209 {
4210 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
4211 struct mempolicy *mpol;
4212
4213 if (sbinfo->max_blocks != shmem_default_max_blocks())
4214 seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks));
4215 if (sbinfo->max_inodes != shmem_default_max_inodes())
4216 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
4217 if (sbinfo->mode != (0777 | S_ISVTX))
4218 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
4219 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4220 seq_printf(seq, ",uid=%u",
4221 from_kuid_munged(&init_user_ns, sbinfo->uid));
4222 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4223 seq_printf(seq, ",gid=%u",
4224 from_kgid_munged(&init_user_ns, sbinfo->gid));
4225
4226 /*
4227 * Showing inode{64,32} might be useful even if it's the system default,
4228 * since then people don't have to resort to checking both here and
4229 * /proc/config.gz to confirm 64-bit inums were successfully applied
4230 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4231 *
4232 * We hide it when inode64 isn't the default and we are using 32-bit
4233 * inodes, since that probably just means the feature isn't even under
4234 * consideration.
4235 *
4236 * As such:
4237 *
4238 * +-----------------+-----------------+
4239 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4240 * +------------------+-----------------+-----------------+
4241 * | full_inums=true | show | show |
4242 * | full_inums=false | show | hide |
4243 * +------------------+-----------------+-----------------+
4244 *
4245 */
4246 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4247 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
4248 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4249 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4250 if (sbinfo->huge)
4251 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4252 #endif
4253 mpol = shmem_get_sbmpol(sbinfo);
4254 shmem_show_mpol(seq, mpol);
4255 mpol_put(mpol);
4256 if (sbinfo->noswap)
4257 seq_printf(seq, ",noswap");
4258 return 0;
4259 }
4260
4261 #endif /* CONFIG_TMPFS */
4262
shmem_put_super(struct super_block * sb)4263 static void shmem_put_super(struct super_block *sb)
4264 {
4265 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4266
4267 #ifdef CONFIG_TMPFS_QUOTA
4268 shmem_disable_quotas(sb);
4269 #endif
4270 free_percpu(sbinfo->ino_batch);
4271 percpu_counter_destroy(&sbinfo->used_blocks);
4272 mpol_put(sbinfo->mpol);
4273 kfree(sbinfo);
4274 sb->s_fs_info = NULL;
4275 }
4276
shmem_fill_super(struct super_block * sb,struct fs_context * fc)4277 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
4278 {
4279 struct shmem_options *ctx = fc->fs_private;
4280 struct inode *inode;
4281 struct shmem_sb_info *sbinfo;
4282 int error = -ENOMEM;
4283
4284 /* Round up to L1_CACHE_BYTES to resist false sharing */
4285 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
4286 L1_CACHE_BYTES), GFP_KERNEL);
4287 if (!sbinfo)
4288 return error;
4289
4290 sb->s_fs_info = sbinfo;
4291
4292 #ifdef CONFIG_TMPFS
4293 /*
4294 * Per default we only allow half of the physical ram per
4295 * tmpfs instance, limiting inodes to one per page of lowmem;
4296 * but the internal instance is left unlimited.
4297 */
4298 if (!(sb->s_flags & SB_KERNMOUNT)) {
4299 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4300 ctx->blocks = shmem_default_max_blocks();
4301 if (!(ctx->seen & SHMEM_SEEN_INODES))
4302 ctx->inodes = shmem_default_max_inodes();
4303 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4304 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
4305 sbinfo->noswap = ctx->noswap;
4306 } else {
4307 sb->s_flags |= SB_NOUSER;
4308 }
4309 sb->s_export_op = &shmem_export_ops;
4310 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
4311 #else
4312 sb->s_flags |= SB_NOUSER;
4313 #endif
4314 sbinfo->max_blocks = ctx->blocks;
4315 sbinfo->max_inodes = ctx->inodes;
4316 sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
4317 if (sb->s_flags & SB_KERNMOUNT) {
4318 sbinfo->ino_batch = alloc_percpu(ino_t);
4319 if (!sbinfo->ino_batch)
4320 goto failed;
4321 }
4322 sbinfo->uid = ctx->uid;
4323 sbinfo->gid = ctx->gid;
4324 sbinfo->full_inums = ctx->full_inums;
4325 sbinfo->mode = ctx->mode;
4326 sbinfo->huge = ctx->huge;
4327 sbinfo->mpol = ctx->mpol;
4328 ctx->mpol = NULL;
4329
4330 raw_spin_lock_init(&sbinfo->stat_lock);
4331 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
4332 goto failed;
4333 spin_lock_init(&sbinfo->shrinklist_lock);
4334 INIT_LIST_HEAD(&sbinfo->shrinklist);
4335
4336 sb->s_maxbytes = MAX_LFS_FILESIZE;
4337 sb->s_blocksize = PAGE_SIZE;
4338 sb->s_blocksize_bits = PAGE_SHIFT;
4339 sb->s_magic = TMPFS_MAGIC;
4340 sb->s_op = &shmem_ops;
4341 sb->s_time_gran = 1;
4342 #ifdef CONFIG_TMPFS_XATTR
4343 sb->s_xattr = shmem_xattr_handlers;
4344 #endif
4345 #ifdef CONFIG_TMPFS_POSIX_ACL
4346 sb->s_flags |= SB_POSIXACL;
4347 #endif
4348 uuid_gen(&sb->s_uuid);
4349
4350 #ifdef CONFIG_TMPFS_QUOTA
4351 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4352 sb->dq_op = &shmem_quota_operations;
4353 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4354 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4355
4356 /* Copy the default limits from ctx into sbinfo */
4357 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4358 sizeof(struct shmem_quota_limits));
4359
4360 if (shmem_enable_quotas(sb, ctx->quota_types))
4361 goto failed;
4362 }
4363 #endif /* CONFIG_TMPFS_QUOTA */
4364
4365 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
4366 VM_NORESERVE);
4367 if (IS_ERR(inode)) {
4368 error = PTR_ERR(inode);
4369 goto failed;
4370 }
4371 inode->i_uid = sbinfo->uid;
4372 inode->i_gid = sbinfo->gid;
4373 sb->s_root = d_make_root(inode);
4374 if (!sb->s_root)
4375 goto failed;
4376 return 0;
4377
4378 failed:
4379 shmem_put_super(sb);
4380 return error;
4381 }
4382
shmem_get_tree(struct fs_context * fc)4383 static int shmem_get_tree(struct fs_context *fc)
4384 {
4385 return get_tree_nodev(fc, shmem_fill_super);
4386 }
4387
shmem_free_fc(struct fs_context * fc)4388 static void shmem_free_fc(struct fs_context *fc)
4389 {
4390 struct shmem_options *ctx = fc->fs_private;
4391
4392 if (ctx) {
4393 mpol_put(ctx->mpol);
4394 kfree(ctx);
4395 }
4396 }
4397
4398 static const struct fs_context_operations shmem_fs_context_ops = {
4399 .free = shmem_free_fc,
4400 .get_tree = shmem_get_tree,
4401 #ifdef CONFIG_TMPFS
4402 .parse_monolithic = shmem_parse_options,
4403 .parse_param = shmem_parse_one,
4404 .reconfigure = shmem_reconfigure,
4405 #endif
4406 };
4407
4408 static struct kmem_cache *shmem_inode_cachep;
4409
shmem_alloc_inode(struct super_block * sb)4410 static struct inode *shmem_alloc_inode(struct super_block *sb)
4411 {
4412 struct shmem_inode_info *info;
4413 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4414 if (!info)
4415 return NULL;
4416 return &info->vfs_inode;
4417 }
4418
shmem_free_in_core_inode(struct inode * inode)4419 static void shmem_free_in_core_inode(struct inode *inode)
4420 {
4421 if (S_ISLNK(inode->i_mode))
4422 kfree(inode->i_link);
4423 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4424 }
4425
shmem_destroy_inode(struct inode * inode)4426 static void shmem_destroy_inode(struct inode *inode)
4427 {
4428 if (S_ISREG(inode->i_mode))
4429 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4430 if (S_ISDIR(inode->i_mode))
4431 simple_offset_destroy(shmem_get_offset_ctx(inode));
4432 }
4433
shmem_init_inode(void * foo)4434 static void shmem_init_inode(void *foo)
4435 {
4436 struct shmem_inode_info *info = foo;
4437 inode_init_once(&info->vfs_inode);
4438 }
4439
shmem_init_inodecache(void)4440 static void shmem_init_inodecache(void)
4441 {
4442 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4443 sizeof(struct shmem_inode_info),
4444 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4445 }
4446
shmem_destroy_inodecache(void)4447 static void shmem_destroy_inodecache(void)
4448 {
4449 kmem_cache_destroy(shmem_inode_cachep);
4450 }
4451
4452 /* Keep the page in page cache instead of truncating it */
shmem_error_remove_page(struct address_space * mapping,struct page * page)4453 static int shmem_error_remove_page(struct address_space *mapping,
4454 struct page *page)
4455 {
4456 return 0;
4457 }
4458
4459 const struct address_space_operations shmem_aops = {
4460 .writepage = shmem_writepage,
4461 .dirty_folio = noop_dirty_folio,
4462 #ifdef CONFIG_TMPFS
4463 .write_begin = shmem_write_begin,
4464 .write_end = shmem_write_end,
4465 #endif
4466 #ifdef CONFIG_MIGRATION
4467 .migrate_folio = migrate_folio,
4468 #endif
4469 .error_remove_page = shmem_error_remove_page,
4470 };
4471 EXPORT_SYMBOL(shmem_aops);
4472
4473 static const struct file_operations shmem_file_operations = {
4474 .mmap = shmem_mmap,
4475 .open = shmem_file_open,
4476 .get_unmapped_area = shmem_get_unmapped_area,
4477 #ifdef CONFIG_TMPFS
4478 .llseek = shmem_file_llseek,
4479 .read_iter = shmem_file_read_iter,
4480 .write_iter = shmem_file_write_iter,
4481 .fsync = noop_fsync,
4482 .splice_read = shmem_file_splice_read,
4483 .splice_write = iter_file_splice_write,
4484 .fallocate = shmem_fallocate,
4485 #endif
4486 };
4487
4488 static const struct inode_operations shmem_inode_operations = {
4489 .getattr = shmem_getattr,
4490 .setattr = shmem_setattr,
4491 #ifdef CONFIG_TMPFS_XATTR
4492 .listxattr = shmem_listxattr,
4493 .set_acl = simple_set_acl,
4494 .fileattr_get = shmem_fileattr_get,
4495 .fileattr_set = shmem_fileattr_set,
4496 #endif
4497 };
4498
4499 static const struct inode_operations shmem_dir_inode_operations = {
4500 #ifdef CONFIG_TMPFS
4501 .getattr = shmem_getattr,
4502 .create = shmem_create,
4503 .lookup = simple_lookup,
4504 .link = shmem_link,
4505 .unlink = shmem_unlink,
4506 .symlink = shmem_symlink,
4507 .mkdir = shmem_mkdir,
4508 .rmdir = shmem_rmdir,
4509 .mknod = shmem_mknod,
4510 .rename = shmem_rename2,
4511 .tmpfile = shmem_tmpfile,
4512 .get_offset_ctx = shmem_get_offset_ctx,
4513 #endif
4514 #ifdef CONFIG_TMPFS_XATTR
4515 .listxattr = shmem_listxattr,
4516 .fileattr_get = shmem_fileattr_get,
4517 .fileattr_set = shmem_fileattr_set,
4518 #endif
4519 #ifdef CONFIG_TMPFS_POSIX_ACL
4520 .setattr = shmem_setattr,
4521 .set_acl = simple_set_acl,
4522 #endif
4523 };
4524
4525 static const struct inode_operations shmem_special_inode_operations = {
4526 .getattr = shmem_getattr,
4527 #ifdef CONFIG_TMPFS_XATTR
4528 .listxattr = shmem_listxattr,
4529 #endif
4530 #ifdef CONFIG_TMPFS_POSIX_ACL
4531 .setattr = shmem_setattr,
4532 .set_acl = simple_set_acl,
4533 #endif
4534 };
4535
4536 static const struct super_operations shmem_ops = {
4537 .alloc_inode = shmem_alloc_inode,
4538 .free_inode = shmem_free_in_core_inode,
4539 .destroy_inode = shmem_destroy_inode,
4540 #ifdef CONFIG_TMPFS
4541 .statfs = shmem_statfs,
4542 .show_options = shmem_show_options,
4543 #endif
4544 #ifdef CONFIG_TMPFS_QUOTA
4545 .get_dquots = shmem_get_dquots,
4546 #endif
4547 .evict_inode = shmem_evict_inode,
4548 .drop_inode = generic_delete_inode,
4549 .put_super = shmem_put_super,
4550 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4551 .nr_cached_objects = shmem_unused_huge_count,
4552 .free_cached_objects = shmem_unused_huge_scan,
4553 #endif
4554 };
4555
4556 static const struct vm_operations_struct shmem_vm_ops = {
4557 .fault = shmem_fault,
4558 .map_pages = filemap_map_pages,
4559 #ifdef CONFIG_NUMA
4560 .set_policy = shmem_set_policy,
4561 .get_policy = shmem_get_policy,
4562 #endif
4563 };
4564
4565 static const struct vm_operations_struct shmem_anon_vm_ops = {
4566 .fault = shmem_fault,
4567 .map_pages = filemap_map_pages,
4568 #ifdef CONFIG_NUMA
4569 .set_policy = shmem_set_policy,
4570 .get_policy = shmem_get_policy,
4571 #endif
4572 };
4573
shmem_init_fs_context(struct fs_context * fc)4574 int shmem_init_fs_context(struct fs_context *fc)
4575 {
4576 struct shmem_options *ctx;
4577
4578 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4579 if (!ctx)
4580 return -ENOMEM;
4581
4582 ctx->mode = 0777 | S_ISVTX;
4583 ctx->uid = current_fsuid();
4584 ctx->gid = current_fsgid();
4585
4586 fc->fs_private = ctx;
4587 fc->ops = &shmem_fs_context_ops;
4588 return 0;
4589 }
4590
4591 static struct file_system_type shmem_fs_type = {
4592 .owner = THIS_MODULE,
4593 .name = "tmpfs",
4594 .init_fs_context = shmem_init_fs_context,
4595 #ifdef CONFIG_TMPFS
4596 .parameters = shmem_fs_parameters,
4597 #endif
4598 .kill_sb = kill_litter_super,
4599 #ifdef CONFIG_SHMEM
4600 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4601 #else
4602 .fs_flags = FS_USERNS_MOUNT,
4603 #endif
4604 };
4605
shmem_init(void)4606 void __init shmem_init(void)
4607 {
4608 int error;
4609
4610 shmem_init_inodecache();
4611
4612 #ifdef CONFIG_TMPFS_QUOTA
4613 error = register_quota_format(&shmem_quota_format);
4614 if (error < 0) {
4615 pr_err("Could not register quota format\n");
4616 goto out3;
4617 }
4618 #endif
4619
4620 error = register_filesystem(&shmem_fs_type);
4621 if (error) {
4622 pr_err("Could not register tmpfs\n");
4623 goto out2;
4624 }
4625
4626 shm_mnt = kern_mount(&shmem_fs_type);
4627 if (IS_ERR(shm_mnt)) {
4628 error = PTR_ERR(shm_mnt);
4629 pr_err("Could not kern_mount tmpfs\n");
4630 goto out1;
4631 }
4632
4633 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4634 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4635 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4636 else
4637 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4638 #endif
4639 return;
4640
4641 out1:
4642 unregister_filesystem(&shmem_fs_type);
4643 out2:
4644 #ifdef CONFIG_TMPFS_QUOTA
4645 unregister_quota_format(&shmem_quota_format);
4646 out3:
4647 #endif
4648 shmem_destroy_inodecache();
4649 shm_mnt = ERR_PTR(error);
4650 }
4651
4652 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
shmem_enabled_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)4653 static ssize_t shmem_enabled_show(struct kobject *kobj,
4654 struct kobj_attribute *attr, char *buf)
4655 {
4656 static const int values[] = {
4657 SHMEM_HUGE_ALWAYS,
4658 SHMEM_HUGE_WITHIN_SIZE,
4659 SHMEM_HUGE_ADVISE,
4660 SHMEM_HUGE_NEVER,
4661 SHMEM_HUGE_DENY,
4662 SHMEM_HUGE_FORCE,
4663 };
4664 int len = 0;
4665 int i;
4666
4667 for (i = 0; i < ARRAY_SIZE(values); i++) {
4668 len += sysfs_emit_at(buf, len,
4669 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4670 i ? " " : "",
4671 shmem_format_huge(values[i]));
4672 }
4673
4674 len += sysfs_emit_at(buf, len, "\n");
4675
4676 return len;
4677 }
4678
shmem_enabled_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)4679 static ssize_t shmem_enabled_store(struct kobject *kobj,
4680 struct kobj_attribute *attr, const char *buf, size_t count)
4681 {
4682 char tmp[16];
4683 int huge;
4684
4685 if (count + 1 > sizeof(tmp))
4686 return -EINVAL;
4687 memcpy(tmp, buf, count);
4688 tmp[count] = '\0';
4689 if (count && tmp[count - 1] == '\n')
4690 tmp[count - 1] = '\0';
4691
4692 huge = shmem_parse_huge(tmp);
4693 if (huge == -EINVAL)
4694 return -EINVAL;
4695 if (!has_transparent_hugepage() &&
4696 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4697 return -EINVAL;
4698
4699 shmem_huge = huge;
4700 if (shmem_huge > SHMEM_HUGE_DENY)
4701 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4702 return count;
4703 }
4704
4705 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4706 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4707
4708 #else /* !CONFIG_SHMEM */
4709
4710 /*
4711 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4712 *
4713 * This is intended for small system where the benefits of the full
4714 * shmem code (swap-backed and resource-limited) are outweighed by
4715 * their complexity. On systems without swap this code should be
4716 * effectively equivalent, but much lighter weight.
4717 */
4718
4719 static struct file_system_type shmem_fs_type = {
4720 .name = "tmpfs",
4721 .init_fs_context = ramfs_init_fs_context,
4722 .parameters = ramfs_fs_parameters,
4723 .kill_sb = ramfs_kill_sb,
4724 .fs_flags = FS_USERNS_MOUNT,
4725 };
4726
shmem_init(void)4727 void __init shmem_init(void)
4728 {
4729 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4730
4731 shm_mnt = kern_mount(&shmem_fs_type);
4732 BUG_ON(IS_ERR(shm_mnt));
4733 }
4734
shmem_unuse(unsigned int type)4735 int shmem_unuse(unsigned int type)
4736 {
4737 return 0;
4738 }
4739
shmem_lock(struct file * file,int lock,struct ucounts * ucounts)4740 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4741 {
4742 return 0;
4743 }
4744
shmem_unlock_mapping(struct address_space * mapping)4745 void shmem_unlock_mapping(struct address_space *mapping)
4746 {
4747 }
4748
4749 #ifdef CONFIG_MMU
shmem_get_unmapped_area(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)4750 unsigned long shmem_get_unmapped_area(struct file *file,
4751 unsigned long addr, unsigned long len,
4752 unsigned long pgoff, unsigned long flags)
4753 {
4754 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4755 }
4756 #endif
4757
shmem_truncate_range(struct inode * inode,loff_t lstart,loff_t lend)4758 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4759 {
4760 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4761 }
4762 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4763
4764 #define shmem_vm_ops generic_file_vm_ops
4765 #define shmem_anon_vm_ops generic_file_vm_ops
4766 #define shmem_file_operations ramfs_file_operations
4767 #define shmem_acct_size(flags, size) 0
4768 #define shmem_unacct_size(flags, size) do {} while (0)
4769
shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)4770 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb, struct inode *dir,
4771 umode_t mode, dev_t dev, unsigned long flags)
4772 {
4773 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
4774 return inode ? inode : ERR_PTR(-ENOSPC);
4775 }
4776
4777 #endif /* CONFIG_SHMEM */
4778
4779 /* common code */
4780
__shmem_file_setup(struct vfsmount * mnt,const char * name,loff_t size,unsigned long flags,unsigned int i_flags)4781 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4782 unsigned long flags, unsigned int i_flags)
4783 {
4784 struct inode *inode;
4785 struct file *res;
4786
4787 if (IS_ERR(mnt))
4788 return ERR_CAST(mnt);
4789
4790 if (size < 0 || size > MAX_LFS_FILESIZE)
4791 return ERR_PTR(-EINVAL);
4792
4793 if (shmem_acct_size(flags, size))
4794 return ERR_PTR(-ENOMEM);
4795
4796 if (is_idmapped_mnt(mnt))
4797 return ERR_PTR(-EINVAL);
4798
4799 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4800 S_IFREG | S_IRWXUGO, 0, flags);
4801
4802 if (IS_ERR(inode)) {
4803 shmem_unacct_size(flags, size);
4804 return ERR_CAST(inode);
4805 }
4806 inode->i_flags |= i_flags;
4807 inode->i_size = size;
4808 clear_nlink(inode); /* It is unlinked */
4809 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4810 if (!IS_ERR(res))
4811 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4812 &shmem_file_operations);
4813 if (IS_ERR(res))
4814 iput(inode);
4815 return res;
4816 }
4817
4818 /**
4819 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4820 * kernel internal. There will be NO LSM permission checks against the
4821 * underlying inode. So users of this interface must do LSM checks at a
4822 * higher layer. The users are the big_key and shm implementations. LSM
4823 * checks are provided at the key or shm level rather than the inode.
4824 * @name: name for dentry (to be seen in /proc/<pid>/maps
4825 * @size: size to be set for the file
4826 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4827 */
shmem_kernel_file_setup(const char * name,loff_t size,unsigned long flags)4828 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4829 {
4830 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4831 }
4832
4833 /**
4834 * shmem_file_setup - get an unlinked file living in tmpfs
4835 * @name: name for dentry (to be seen in /proc/<pid>/maps
4836 * @size: size to be set for the file
4837 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4838 */
shmem_file_setup(const char * name,loff_t size,unsigned long flags)4839 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4840 {
4841 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4842 }
4843 EXPORT_SYMBOL_GPL(shmem_file_setup);
4844
4845 /**
4846 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4847 * @mnt: the tmpfs mount where the file will be created
4848 * @name: name for dentry (to be seen in /proc/<pid>/maps
4849 * @size: size to be set for the file
4850 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4851 */
shmem_file_setup_with_mnt(struct vfsmount * mnt,const char * name,loff_t size,unsigned long flags)4852 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4853 loff_t size, unsigned long flags)
4854 {
4855 return __shmem_file_setup(mnt, name, size, flags, 0);
4856 }
4857 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4858
4859 /**
4860 * shmem_zero_setup - setup a shared anonymous mapping
4861 * @vma: the vma to be mmapped is prepared by do_mmap
4862 */
shmem_zero_setup(struct vm_area_struct * vma)4863 int shmem_zero_setup(struct vm_area_struct *vma)
4864 {
4865 struct file *file;
4866 loff_t size = vma->vm_end - vma->vm_start;
4867
4868 /*
4869 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4870 * between XFS directory reading and selinux: since this file is only
4871 * accessible to the user through its mapping, use S_PRIVATE flag to
4872 * bypass file security, in the same way as shmem_kernel_file_setup().
4873 */
4874 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4875 if (IS_ERR(file))
4876 return PTR_ERR(file);
4877
4878 if (vma->vm_file)
4879 fput(vma->vm_file);
4880 vma->vm_file = file;
4881 vma->vm_ops = &shmem_anon_vm_ops;
4882
4883 return 0;
4884 }
4885
4886 /**
4887 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4888 * @mapping: the folio's address_space
4889 * @index: the folio index
4890 * @gfp: the page allocator flags to use if allocating
4891 *
4892 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4893 * with any new page allocations done using the specified allocation flags.
4894 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4895 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4896 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4897 *
4898 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4899 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4900 */
shmem_read_folio_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)4901 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4902 pgoff_t index, gfp_t gfp)
4903 {
4904 #ifdef CONFIG_SHMEM
4905 struct inode *inode = mapping->host;
4906 struct folio *folio;
4907 int error;
4908
4909 BUG_ON(!shmem_mapping(mapping));
4910 error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4911 gfp, NULL, NULL, NULL);
4912 if (error)
4913 return ERR_PTR(error);
4914
4915 folio_unlock(folio);
4916 return folio;
4917 #else
4918 /*
4919 * The tiny !SHMEM case uses ramfs without swap
4920 */
4921 return mapping_read_folio_gfp(mapping, index, gfp);
4922 #endif
4923 }
4924 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4925
shmem_read_mapping_page_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)4926 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4927 pgoff_t index, gfp_t gfp)
4928 {
4929 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4930 struct page *page;
4931
4932 if (IS_ERR(folio))
4933 return &folio->page;
4934
4935 page = folio_file_page(folio, index);
4936 if (PageHWPoison(page)) {
4937 folio_put(folio);
4938 return ERR_PTR(-EIO);
4939 }
4940
4941 return page;
4942 }
4943 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4944