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 __ro_after_init;
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 #include <linux/rcupdate_wait.h>
83
84 #include <linux/uaccess.h>
85
86 #include "internal.h"
87
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
93
94 /* Pretend that one inode + its dentry occupy this much memory */
95 #define BOGO_INODE_SIZE 1024
96
97 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
98 #define SHORT_SYMLINK_LEN 128
99
100 /*
101 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
102 * inode->i_private (with i_rwsem making sure that it has only one user at
103 * a time): we would prefer not to enlarge the shmem inode just for that.
104 */
105 struct shmem_falloc {
106 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
107 pgoff_t start; /* start of range currently being fallocated */
108 pgoff_t next; /* the next page offset to be fallocated */
109 pgoff_t nr_falloced; /* how many new pages have been fallocated */
110 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
111 };
112
113 struct shmem_options {
114 unsigned long long blocks;
115 unsigned long long inodes;
116 struct mempolicy *mpol;
117 kuid_t uid;
118 kgid_t gid;
119 umode_t mode;
120 bool full_inums;
121 int huge;
122 int seen;
123 bool noswap;
124 unsigned short quota_types;
125 struct shmem_quota_limits qlimits;
126 #define SHMEM_SEEN_BLOCKS 1
127 #define SHMEM_SEEN_INODES 2
128 #define SHMEM_SEEN_HUGE 4
129 #define SHMEM_SEEN_INUMS 8
130 #define SHMEM_SEEN_NOSWAP 16
131 #define SHMEM_SEEN_QUOTA 32
132 };
133
134 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
135 static unsigned long huge_shmem_orders_always __read_mostly;
136 static unsigned long huge_shmem_orders_madvise __read_mostly;
137 static unsigned long huge_shmem_orders_inherit __read_mostly;
138 static unsigned long huge_shmem_orders_within_size __read_mostly;
139 #endif
140
141 #ifdef CONFIG_TMPFS
shmem_default_max_blocks(void)142 static unsigned long shmem_default_max_blocks(void)
143 {
144 return totalram_pages() / 2;
145 }
146
shmem_default_max_inodes(void)147 static unsigned long shmem_default_max_inodes(void)
148 {
149 unsigned long nr_pages = totalram_pages();
150
151 return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
152 ULONG_MAX / BOGO_INODE_SIZE);
153 }
154 #endif
155
156 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
157 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
158 struct vm_area_struct *vma, vm_fault_t *fault_type);
159
SHMEM_SB(struct super_block * sb)160 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161 {
162 return sb->s_fs_info;
163 }
164
165 /*
166 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
167 * for shared memory and for shared anonymous (/dev/zero) mappings
168 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
169 * consistent with the pre-accounting of private mappings ...
170 */
shmem_acct_size(unsigned long flags,loff_t size)171 static inline int shmem_acct_size(unsigned long flags, loff_t size)
172 {
173 return (flags & VM_NORESERVE) ?
174 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
175 }
176
shmem_unacct_size(unsigned long flags,loff_t size)177 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178 {
179 if (!(flags & VM_NORESERVE))
180 vm_unacct_memory(VM_ACCT(size));
181 }
182
shmem_reacct_size(unsigned long flags,loff_t oldsize,loff_t newsize)183 static inline int shmem_reacct_size(unsigned long flags,
184 loff_t oldsize, loff_t newsize)
185 {
186 if (!(flags & VM_NORESERVE)) {
187 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
188 return security_vm_enough_memory_mm(current->mm,
189 VM_ACCT(newsize) - VM_ACCT(oldsize));
190 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
191 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 }
193 return 0;
194 }
195
196 /*
197 * ... whereas tmpfs objects are accounted incrementally as
198 * pages are allocated, in order to allow large sparse files.
199 * shmem_get_folio reports shmem_acct_blocks failure as -ENOSPC not -ENOMEM,
200 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 */
shmem_acct_blocks(unsigned long flags,long pages)202 static inline int shmem_acct_blocks(unsigned long flags, long pages)
203 {
204 if (!(flags & VM_NORESERVE))
205 return 0;
206
207 return security_vm_enough_memory_mm(current->mm,
208 pages * VM_ACCT(PAGE_SIZE));
209 }
210
shmem_unacct_blocks(unsigned long flags,long pages)211 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 {
213 if (flags & VM_NORESERVE)
214 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
215 }
216
shmem_inode_acct_blocks(struct inode * inode,long pages)217 static int shmem_inode_acct_blocks(struct inode *inode, long pages)
218 {
219 struct shmem_inode_info *info = SHMEM_I(inode);
220 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 int err = -ENOSPC;
222
223 if (shmem_acct_blocks(info->flags, pages))
224 return err;
225
226 might_sleep(); /* when quotas */
227 if (sbinfo->max_blocks) {
228 if (!percpu_counter_limited_add(&sbinfo->used_blocks,
229 sbinfo->max_blocks, pages))
230 goto unacct;
231
232 err = dquot_alloc_block_nodirty(inode, pages);
233 if (err) {
234 percpu_counter_sub(&sbinfo->used_blocks, pages);
235 goto unacct;
236 }
237 } else {
238 err = dquot_alloc_block_nodirty(inode, pages);
239 if (err)
240 goto unacct;
241 }
242
243 return 0;
244
245 unacct:
246 shmem_unacct_blocks(info->flags, pages);
247 return err;
248 }
249
shmem_inode_unacct_blocks(struct inode * inode,long pages)250 static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
251 {
252 struct shmem_inode_info *info = SHMEM_I(inode);
253 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
254
255 might_sleep(); /* when quotas */
256 dquot_free_block_nodirty(inode, pages);
257
258 if (sbinfo->max_blocks)
259 percpu_counter_sub(&sbinfo->used_blocks, pages);
260 shmem_unacct_blocks(info->flags, pages);
261 }
262
263 static const struct super_operations shmem_ops;
264 static const struct address_space_operations shmem_aops;
265 static const struct file_operations shmem_file_operations;
266 static const struct inode_operations shmem_inode_operations;
267 static const struct inode_operations shmem_dir_inode_operations;
268 static const struct inode_operations shmem_special_inode_operations;
269 static const struct vm_operations_struct shmem_vm_ops;
270 static const struct vm_operations_struct shmem_anon_vm_ops;
271 static struct file_system_type shmem_fs_type;
272
shmem_mapping(struct address_space * mapping)273 bool shmem_mapping(struct address_space *mapping)
274 {
275 return mapping->a_ops == &shmem_aops;
276 }
277 EXPORT_SYMBOL_GPL(shmem_mapping);
278
vma_is_anon_shmem(struct vm_area_struct * vma)279 bool vma_is_anon_shmem(struct vm_area_struct *vma)
280 {
281 return vma->vm_ops == &shmem_anon_vm_ops;
282 }
283
vma_is_shmem(struct vm_area_struct * vma)284 bool vma_is_shmem(struct vm_area_struct *vma)
285 {
286 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
287 }
288
289 static LIST_HEAD(shmem_swaplist);
290 static DEFINE_MUTEX(shmem_swaplist_mutex);
291
292 #ifdef CONFIG_TMPFS_QUOTA
293
shmem_enable_quotas(struct super_block * sb,unsigned short quota_types)294 static int shmem_enable_quotas(struct super_block *sb,
295 unsigned short quota_types)
296 {
297 int type, err = 0;
298
299 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
300 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
301 if (!(quota_types & (1 << type)))
302 continue;
303 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
304 DQUOT_USAGE_ENABLED |
305 DQUOT_LIMITS_ENABLED);
306 if (err)
307 goto out_err;
308 }
309 return 0;
310
311 out_err:
312 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
313 type, err);
314 for (type--; type >= 0; type--)
315 dquot_quota_off(sb, type);
316 return err;
317 }
318
shmem_disable_quotas(struct super_block * sb)319 static void shmem_disable_quotas(struct super_block *sb)
320 {
321 int type;
322
323 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
324 dquot_quota_off(sb, type);
325 }
326
shmem_get_dquots(struct inode * inode)327 static struct dquot __rcu **shmem_get_dquots(struct inode *inode)
328 {
329 return SHMEM_I(inode)->i_dquot;
330 }
331 #endif /* CONFIG_TMPFS_QUOTA */
332
333 /*
334 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
335 * produces a novel ino for the newly allocated inode.
336 *
337 * It may also be called when making a hard link to permit the space needed by
338 * each dentry. However, in that case, no new inode number is needed since that
339 * internally draws from another pool of inode numbers (currently global
340 * get_next_ino()). This case is indicated by passing NULL as inop.
341 */
342 #define SHMEM_INO_BATCH 1024
shmem_reserve_inode(struct super_block * sb,ino_t * inop)343 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
344 {
345 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
346 ino_t ino;
347
348 if (!(sb->s_flags & SB_KERNMOUNT)) {
349 raw_spin_lock(&sbinfo->stat_lock);
350 if (sbinfo->max_inodes) {
351 if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
352 raw_spin_unlock(&sbinfo->stat_lock);
353 return -ENOSPC;
354 }
355 sbinfo->free_ispace -= BOGO_INODE_SIZE;
356 }
357 if (inop) {
358 ino = sbinfo->next_ino++;
359 if (unlikely(is_zero_ino(ino)))
360 ino = sbinfo->next_ino++;
361 if (unlikely(!sbinfo->full_inums &&
362 ino > UINT_MAX)) {
363 /*
364 * Emulate get_next_ino uint wraparound for
365 * compatibility
366 */
367 if (IS_ENABLED(CONFIG_64BIT))
368 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
369 __func__, MINOR(sb->s_dev));
370 sbinfo->next_ino = 1;
371 ino = sbinfo->next_ino++;
372 }
373 *inop = ino;
374 }
375 raw_spin_unlock(&sbinfo->stat_lock);
376 } else if (inop) {
377 /*
378 * __shmem_file_setup, one of our callers, is lock-free: it
379 * doesn't hold stat_lock in shmem_reserve_inode since
380 * max_inodes is always 0, and is called from potentially
381 * unknown contexts. As such, use a per-cpu batched allocator
382 * which doesn't require the per-sb stat_lock unless we are at
383 * the batch boundary.
384 *
385 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
386 * shmem mounts are not exposed to userspace, so we don't need
387 * to worry about things like glibc compatibility.
388 */
389 ino_t *next_ino;
390
391 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
392 ino = *next_ino;
393 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
394 raw_spin_lock(&sbinfo->stat_lock);
395 ino = sbinfo->next_ino;
396 sbinfo->next_ino += SHMEM_INO_BATCH;
397 raw_spin_unlock(&sbinfo->stat_lock);
398 if (unlikely(is_zero_ino(ino)))
399 ino++;
400 }
401 *inop = ino;
402 *next_ino = ++ino;
403 put_cpu();
404 }
405
406 return 0;
407 }
408
shmem_free_inode(struct super_block * sb,size_t freed_ispace)409 static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
410 {
411 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
412 if (sbinfo->max_inodes) {
413 raw_spin_lock(&sbinfo->stat_lock);
414 sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
415 raw_spin_unlock(&sbinfo->stat_lock);
416 }
417 }
418
419 /**
420 * shmem_recalc_inode - recalculate the block usage of an inode
421 * @inode: inode to recalc
422 * @alloced: the change in number of pages allocated to inode
423 * @swapped: the change in number of pages swapped from inode
424 *
425 * We have to calculate the free blocks since the mm can drop
426 * undirtied hole pages behind our back.
427 *
428 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
429 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
430 */
shmem_recalc_inode(struct inode * inode,long alloced,long swapped)431 static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
432 {
433 struct shmem_inode_info *info = SHMEM_I(inode);
434 long freed;
435
436 spin_lock(&info->lock);
437 info->alloced += alloced;
438 info->swapped += swapped;
439 freed = info->alloced - info->swapped -
440 READ_ONCE(inode->i_mapping->nrpages);
441 /*
442 * Special case: whereas normally shmem_recalc_inode() is called
443 * after i_mapping->nrpages has already been adjusted (up or down),
444 * shmem_writepage() has to raise swapped before nrpages is lowered -
445 * to stop a racing shmem_recalc_inode() from thinking that a page has
446 * been freed. Compensate here, to avoid the need for a followup call.
447 */
448 if (swapped > 0)
449 freed += swapped;
450 if (freed > 0)
451 info->alloced -= freed;
452 spin_unlock(&info->lock);
453
454 /* The quota case may block */
455 if (freed > 0)
456 shmem_inode_unacct_blocks(inode, freed);
457 }
458
shmem_charge(struct inode * inode,long pages)459 bool shmem_charge(struct inode *inode, long pages)
460 {
461 struct address_space *mapping = inode->i_mapping;
462
463 if (shmem_inode_acct_blocks(inode, pages))
464 return false;
465
466 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
467 xa_lock_irq(&mapping->i_pages);
468 mapping->nrpages += pages;
469 xa_unlock_irq(&mapping->i_pages);
470
471 shmem_recalc_inode(inode, pages, 0);
472 return true;
473 }
474
shmem_uncharge(struct inode * inode,long pages)475 void shmem_uncharge(struct inode *inode, long pages)
476 {
477 /* pages argument is currently unused: keep it to help debugging */
478 /* nrpages adjustment done by __filemap_remove_folio() or caller */
479
480 shmem_recalc_inode(inode, 0, 0);
481 }
482
483 /*
484 * Replace item expected in xarray by a new item, while holding xa_lock.
485 */
shmem_replace_entry(struct address_space * mapping,pgoff_t index,void * expected,void * replacement)486 static int shmem_replace_entry(struct address_space *mapping,
487 pgoff_t index, void *expected, void *replacement)
488 {
489 XA_STATE(xas, &mapping->i_pages, index);
490 void *item;
491
492 VM_BUG_ON(!expected);
493 VM_BUG_ON(!replacement);
494 item = xas_load(&xas);
495 if (item != expected)
496 return -ENOENT;
497 xas_store(&xas, replacement);
498 return 0;
499 }
500
501 /*
502 * Sometimes, before we decide whether to proceed or to fail, we must check
503 * that an entry was not already brought back from swap by a racing thread.
504 *
505 * Checking folio is not enough: by the time a swapcache folio is locked, it
506 * might be reused, and again be swapcache, using the same swap as before.
507 */
shmem_confirm_swap(struct address_space * mapping,pgoff_t index,swp_entry_t swap)508 static bool shmem_confirm_swap(struct address_space *mapping,
509 pgoff_t index, swp_entry_t swap)
510 {
511 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
512 }
513
514 /*
515 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
516 *
517 * SHMEM_HUGE_NEVER:
518 * disables huge pages for the mount;
519 * SHMEM_HUGE_ALWAYS:
520 * enables huge pages for the mount;
521 * SHMEM_HUGE_WITHIN_SIZE:
522 * only allocate huge pages if the page will be fully within i_size,
523 * also respect fadvise()/madvise() hints;
524 * SHMEM_HUGE_ADVISE:
525 * only allocate huge pages if requested with fadvise()/madvise();
526 */
527
528 #define SHMEM_HUGE_NEVER 0
529 #define SHMEM_HUGE_ALWAYS 1
530 #define SHMEM_HUGE_WITHIN_SIZE 2
531 #define SHMEM_HUGE_ADVISE 3
532
533 /*
534 * Special values.
535 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
536 *
537 * SHMEM_HUGE_DENY:
538 * disables huge on shm_mnt and all mounts, for emergency use;
539 * SHMEM_HUGE_FORCE:
540 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
541 *
542 */
543 #define SHMEM_HUGE_DENY (-1)
544 #define SHMEM_HUGE_FORCE (-2)
545
546 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
547 /* ifdef here to avoid bloating shmem.o when not necessary */
548
549 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
550
__shmem_huge_global_enabled(struct inode * inode,pgoff_t index,loff_t write_end,bool shmem_huge_force,struct vm_area_struct * vma,unsigned long vm_flags)551 static bool __shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
552 loff_t write_end, bool shmem_huge_force,
553 struct vm_area_struct *vma,
554 unsigned long vm_flags)
555 {
556 struct mm_struct *mm = vma ? vma->vm_mm : NULL;
557 loff_t i_size;
558
559 if (!S_ISREG(inode->i_mode))
560 return false;
561 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
562 return false;
563 if (shmem_huge == SHMEM_HUGE_DENY)
564 return false;
565 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
566 return true;
567
568 switch (SHMEM_SB(inode->i_sb)->huge) {
569 case SHMEM_HUGE_ALWAYS:
570 return true;
571 case SHMEM_HUGE_WITHIN_SIZE:
572 index = round_up(index + 1, HPAGE_PMD_NR);
573 i_size = max(write_end, i_size_read(inode));
574 i_size = round_up(i_size, PAGE_SIZE);
575 if (i_size >> PAGE_SHIFT >= index)
576 return true;
577 fallthrough;
578 case SHMEM_HUGE_ADVISE:
579 if (mm && (vm_flags & VM_HUGEPAGE))
580 return true;
581 fallthrough;
582 default:
583 return false;
584 }
585 }
586
shmem_huge_global_enabled(struct inode * inode,pgoff_t index,loff_t write_end,bool shmem_huge_force,struct vm_area_struct * vma,unsigned long vm_flags)587 static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
588 loff_t write_end, bool shmem_huge_force,
589 struct vm_area_struct *vma, unsigned long vm_flags)
590 {
591 if (HPAGE_PMD_ORDER > MAX_PAGECACHE_ORDER)
592 return false;
593
594 return __shmem_huge_global_enabled(inode, index, write_end,
595 shmem_huge_force, vma, vm_flags);
596 }
597
598 #if defined(CONFIG_SYSFS)
shmem_parse_huge(const char * str)599 static int shmem_parse_huge(const char *str)
600 {
601 if (!strcmp(str, "never"))
602 return SHMEM_HUGE_NEVER;
603 if (!strcmp(str, "always"))
604 return SHMEM_HUGE_ALWAYS;
605 if (!strcmp(str, "within_size"))
606 return SHMEM_HUGE_WITHIN_SIZE;
607 if (!strcmp(str, "advise"))
608 return SHMEM_HUGE_ADVISE;
609 if (!strcmp(str, "deny"))
610 return SHMEM_HUGE_DENY;
611 if (!strcmp(str, "force"))
612 return SHMEM_HUGE_FORCE;
613 return -EINVAL;
614 }
615 #endif
616
617 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
shmem_format_huge(int huge)618 static const char *shmem_format_huge(int huge)
619 {
620 switch (huge) {
621 case SHMEM_HUGE_NEVER:
622 return "never";
623 case SHMEM_HUGE_ALWAYS:
624 return "always";
625 case SHMEM_HUGE_WITHIN_SIZE:
626 return "within_size";
627 case SHMEM_HUGE_ADVISE:
628 return "advise";
629 case SHMEM_HUGE_DENY:
630 return "deny";
631 case SHMEM_HUGE_FORCE:
632 return "force";
633 default:
634 VM_BUG_ON(1);
635 return "bad_val";
636 }
637 }
638 #endif
639
shmem_unused_huge_shrink(struct shmem_sb_info * sbinfo,struct shrink_control * sc,unsigned long nr_to_free)640 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
641 struct shrink_control *sc, unsigned long nr_to_free)
642 {
643 LIST_HEAD(list), *pos, *next;
644 struct inode *inode;
645 struct shmem_inode_info *info;
646 struct folio *folio;
647 unsigned long batch = sc ? sc->nr_to_scan : 128;
648 unsigned long split = 0, freed = 0;
649
650 if (list_empty(&sbinfo->shrinklist))
651 return SHRINK_STOP;
652
653 spin_lock(&sbinfo->shrinklist_lock);
654 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
655 info = list_entry(pos, struct shmem_inode_info, shrinklist);
656
657 /* pin the inode */
658 inode = igrab(&info->vfs_inode);
659
660 /* inode is about to be evicted */
661 if (!inode) {
662 list_del_init(&info->shrinklist);
663 goto next;
664 }
665
666 list_move(&info->shrinklist, &list);
667 next:
668 sbinfo->shrinklist_len--;
669 if (!--batch)
670 break;
671 }
672 spin_unlock(&sbinfo->shrinklist_lock);
673
674 list_for_each_safe(pos, next, &list) {
675 pgoff_t next, end;
676 loff_t i_size;
677 int ret;
678
679 info = list_entry(pos, struct shmem_inode_info, shrinklist);
680 inode = &info->vfs_inode;
681
682 if (nr_to_free && freed >= nr_to_free)
683 goto move_back;
684
685 i_size = i_size_read(inode);
686 folio = filemap_get_entry(inode->i_mapping, i_size / PAGE_SIZE);
687 if (!folio || xa_is_value(folio))
688 goto drop;
689
690 /* No large folio at the end of the file: nothing to split */
691 if (!folio_test_large(folio)) {
692 folio_put(folio);
693 goto drop;
694 }
695
696 /* Check if there is anything to gain from splitting */
697 next = folio_next_index(folio);
698 end = shmem_fallocend(inode, DIV_ROUND_UP(i_size, PAGE_SIZE));
699 if (end <= folio->index || end >= next) {
700 folio_put(folio);
701 goto drop;
702 }
703
704 /*
705 * Move the inode on the list back to shrinklist if we failed
706 * to lock the page at this time.
707 *
708 * Waiting for the lock may lead to deadlock in the
709 * reclaim path.
710 */
711 if (!folio_trylock(folio)) {
712 folio_put(folio);
713 goto move_back;
714 }
715
716 ret = split_folio(folio);
717 folio_unlock(folio);
718 folio_put(folio);
719
720 /* If split failed move the inode on the list back to shrinklist */
721 if (ret)
722 goto move_back;
723
724 freed += next - end;
725 split++;
726 drop:
727 list_del_init(&info->shrinklist);
728 goto put;
729 move_back:
730 /*
731 * Make sure the inode is either on the global list or deleted
732 * from any local list before iput() since it could be deleted
733 * in another thread once we put the inode (then the local list
734 * is corrupted).
735 */
736 spin_lock(&sbinfo->shrinklist_lock);
737 list_move(&info->shrinklist, &sbinfo->shrinklist);
738 sbinfo->shrinklist_len++;
739 spin_unlock(&sbinfo->shrinklist_lock);
740 put:
741 iput(inode);
742 }
743
744 return split;
745 }
746
shmem_unused_huge_scan(struct super_block * sb,struct shrink_control * sc)747 static long shmem_unused_huge_scan(struct super_block *sb,
748 struct shrink_control *sc)
749 {
750 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
751
752 if (!READ_ONCE(sbinfo->shrinklist_len))
753 return SHRINK_STOP;
754
755 return shmem_unused_huge_shrink(sbinfo, sc, 0);
756 }
757
shmem_unused_huge_count(struct super_block * sb,struct shrink_control * sc)758 static long shmem_unused_huge_count(struct super_block *sb,
759 struct shrink_control *sc)
760 {
761 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
762 return READ_ONCE(sbinfo->shrinklist_len);
763 }
764 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
765
766 #define shmem_huge SHMEM_HUGE_DENY
767
shmem_unused_huge_shrink(struct shmem_sb_info * sbinfo,struct shrink_control * sc,unsigned long nr_to_free)768 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
769 struct shrink_control *sc, unsigned long nr_to_free)
770 {
771 return 0;
772 }
773
shmem_huge_global_enabled(struct inode * inode,pgoff_t index,loff_t write_end,bool shmem_huge_force,struct vm_area_struct * vma,unsigned long vm_flags)774 static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
775 loff_t write_end, bool shmem_huge_force,
776 struct vm_area_struct *vma, unsigned long vm_flags)
777 {
778 return false;
779 }
780 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
781
782 /*
783 * Somewhat like filemap_add_folio, but error if expected item has gone.
784 */
shmem_add_to_page_cache(struct folio * folio,struct address_space * mapping,pgoff_t index,void * expected,gfp_t gfp)785 static int shmem_add_to_page_cache(struct folio *folio,
786 struct address_space *mapping,
787 pgoff_t index, void *expected, gfp_t gfp)
788 {
789 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
790 long nr = folio_nr_pages(folio);
791
792 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
793 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
794 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
795
796 folio_ref_add(folio, nr);
797 folio->mapping = mapping;
798 folio->index = index;
799
800 gfp &= GFP_RECLAIM_MASK;
801 folio_throttle_swaprate(folio, gfp);
802
803 do {
804 xas_lock_irq(&xas);
805 if (expected != xas_find_conflict(&xas)) {
806 xas_set_err(&xas, -EEXIST);
807 goto unlock;
808 }
809 if (expected && xas_find_conflict(&xas)) {
810 xas_set_err(&xas, -EEXIST);
811 goto unlock;
812 }
813 xas_store(&xas, folio);
814 if (xas_error(&xas))
815 goto unlock;
816 if (folio_test_pmd_mappable(folio))
817 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
818 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
819 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
820 mapping->nrpages += nr;
821 unlock:
822 xas_unlock_irq(&xas);
823 } while (xas_nomem(&xas, gfp));
824
825 if (xas_error(&xas)) {
826 folio->mapping = NULL;
827 folio_ref_sub(folio, nr);
828 return xas_error(&xas);
829 }
830
831 return 0;
832 }
833
834 /*
835 * Somewhat like filemap_remove_folio, but substitutes swap for @folio.
836 */
shmem_delete_from_page_cache(struct folio * folio,void * radswap)837 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
838 {
839 struct address_space *mapping = folio->mapping;
840 long nr = folio_nr_pages(folio);
841 int error;
842
843 xa_lock_irq(&mapping->i_pages);
844 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
845 folio->mapping = NULL;
846 mapping->nrpages -= nr;
847 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
848 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
849 xa_unlock_irq(&mapping->i_pages);
850 folio_put_refs(folio, nr);
851 BUG_ON(error);
852 }
853
854 /*
855 * Remove swap entry from page cache, free the swap and its page cache. Returns
856 * the number of pages being freed. 0 means entry not found in XArray (0 pages
857 * being freed).
858 */
shmem_free_swap(struct address_space * mapping,pgoff_t index,void * radswap)859 static long shmem_free_swap(struct address_space *mapping,
860 pgoff_t index, void *radswap)
861 {
862 int order = xa_get_order(&mapping->i_pages, index);
863 void *old;
864
865 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
866 if (old != radswap)
867 return 0;
868 free_swap_and_cache_nr(radix_to_swp_entry(radswap), 1 << order);
869
870 return 1 << order;
871 }
872
873 /*
874 * Determine (in bytes) how many of the shmem object's pages mapped by the
875 * given offsets are swapped out.
876 *
877 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
878 * as long as the inode doesn't go away and racy results are not a problem.
879 */
shmem_partial_swap_usage(struct address_space * mapping,pgoff_t start,pgoff_t end)880 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
881 pgoff_t start, pgoff_t end)
882 {
883 XA_STATE(xas, &mapping->i_pages, start);
884 struct page *page;
885 unsigned long swapped = 0;
886 unsigned long max = end - 1;
887
888 rcu_read_lock();
889 xas_for_each(&xas, page, max) {
890 if (xas_retry(&xas, page))
891 continue;
892 if (xa_is_value(page))
893 swapped += 1 << xas_get_order(&xas);
894 if (xas.xa_index == max)
895 break;
896 if (need_resched()) {
897 xas_pause(&xas);
898 cond_resched_rcu();
899 }
900 }
901 rcu_read_unlock();
902
903 return swapped << PAGE_SHIFT;
904 }
905
906 /*
907 * Determine (in bytes) how many of the shmem object's pages mapped by the
908 * given vma is swapped out.
909 *
910 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
911 * as long as the inode doesn't go away and racy results are not a problem.
912 */
shmem_swap_usage(struct vm_area_struct * vma)913 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
914 {
915 struct inode *inode = file_inode(vma->vm_file);
916 struct shmem_inode_info *info = SHMEM_I(inode);
917 struct address_space *mapping = inode->i_mapping;
918 unsigned long swapped;
919
920 /* Be careful as we don't hold info->lock */
921 swapped = READ_ONCE(info->swapped);
922
923 /*
924 * The easier cases are when the shmem object has nothing in swap, or
925 * the vma maps it whole. Then we can simply use the stats that we
926 * already track.
927 */
928 if (!swapped)
929 return 0;
930
931 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
932 return swapped << PAGE_SHIFT;
933
934 /* Here comes the more involved part */
935 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
936 vma->vm_pgoff + vma_pages(vma));
937 }
938
939 /*
940 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
941 */
shmem_unlock_mapping(struct address_space * mapping)942 void shmem_unlock_mapping(struct address_space *mapping)
943 {
944 struct folio_batch fbatch;
945 pgoff_t index = 0;
946
947 folio_batch_init(&fbatch);
948 /*
949 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
950 */
951 while (!mapping_unevictable(mapping) &&
952 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
953 check_move_unevictable_folios(&fbatch);
954 folio_batch_release(&fbatch);
955 cond_resched();
956 }
957 }
958
shmem_get_partial_folio(struct inode * inode,pgoff_t index)959 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
960 {
961 struct folio *folio;
962
963 /*
964 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
965 * beyond i_size, and reports fallocated folios as holes.
966 */
967 folio = filemap_get_entry(inode->i_mapping, index);
968 if (!folio)
969 return folio;
970 if (!xa_is_value(folio)) {
971 folio_lock(folio);
972 if (folio->mapping == inode->i_mapping)
973 return folio;
974 /* The folio has been swapped out */
975 folio_unlock(folio);
976 folio_put(folio);
977 }
978 /*
979 * But read a folio back from swap if any of it is within i_size
980 * (although in some cases this is just a waste of time).
981 */
982 folio = NULL;
983 shmem_get_folio(inode, index, 0, &folio, SGP_READ);
984 return folio;
985 }
986
987 /*
988 * Remove range of pages and swap entries from page cache, and free them.
989 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
990 */
shmem_undo_range(struct inode * inode,loff_t lstart,loff_t lend,bool unfalloc)991 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
992 bool unfalloc)
993 {
994 struct address_space *mapping = inode->i_mapping;
995 struct shmem_inode_info *info = SHMEM_I(inode);
996 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
997 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
998 struct folio_batch fbatch;
999 pgoff_t indices[PAGEVEC_SIZE];
1000 struct folio *folio;
1001 bool same_folio;
1002 long nr_swaps_freed = 0;
1003 pgoff_t index;
1004 int i;
1005
1006 if (lend == -1)
1007 end = -1; /* unsigned, so actually very big */
1008
1009 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
1010 info->fallocend = start;
1011
1012 folio_batch_init(&fbatch);
1013 index = start;
1014 while (index < end && find_lock_entries(mapping, &index, end - 1,
1015 &fbatch, indices)) {
1016 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1017 folio = fbatch.folios[i];
1018
1019 if (xa_is_value(folio)) {
1020 if (unfalloc)
1021 continue;
1022 nr_swaps_freed += shmem_free_swap(mapping,
1023 indices[i], folio);
1024 continue;
1025 }
1026
1027 if (!unfalloc || !folio_test_uptodate(folio))
1028 truncate_inode_folio(mapping, folio);
1029 folio_unlock(folio);
1030 }
1031 folio_batch_remove_exceptionals(&fbatch);
1032 folio_batch_release(&fbatch);
1033 cond_resched();
1034 }
1035
1036 /*
1037 * When undoing a failed fallocate, we want none of the partial folio
1038 * zeroing and splitting below, but shall want to truncate the whole
1039 * folio when !uptodate indicates that it was added by this fallocate,
1040 * even when [lstart, lend] covers only a part of the folio.
1041 */
1042 if (unfalloc)
1043 goto whole_folios;
1044
1045 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1046 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1047 if (folio) {
1048 same_folio = lend < folio_pos(folio) + folio_size(folio);
1049 folio_mark_dirty(folio);
1050 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
1051 start = folio_next_index(folio);
1052 if (same_folio)
1053 end = folio->index;
1054 }
1055 folio_unlock(folio);
1056 folio_put(folio);
1057 folio = NULL;
1058 }
1059
1060 if (!same_folio)
1061 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1062 if (folio) {
1063 folio_mark_dirty(folio);
1064 if (!truncate_inode_partial_folio(folio, lstart, lend))
1065 end = folio->index;
1066 folio_unlock(folio);
1067 folio_put(folio);
1068 }
1069
1070 whole_folios:
1071
1072 index = start;
1073 while (index < end) {
1074 cond_resched();
1075
1076 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1077 indices)) {
1078 /* If all gone or hole-punch or unfalloc, we're done */
1079 if (index == start || end != -1)
1080 break;
1081 /* But if truncating, restart to make sure all gone */
1082 index = start;
1083 continue;
1084 }
1085 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1086 folio = fbatch.folios[i];
1087
1088 if (xa_is_value(folio)) {
1089 long swaps_freed;
1090
1091 if (unfalloc)
1092 continue;
1093 swaps_freed = shmem_free_swap(mapping, indices[i], folio);
1094 if (!swaps_freed) {
1095 /* Swap was replaced by page: retry */
1096 index = indices[i];
1097 break;
1098 }
1099 nr_swaps_freed += swaps_freed;
1100 continue;
1101 }
1102
1103 folio_lock(folio);
1104
1105 if (!unfalloc || !folio_test_uptodate(folio)) {
1106 if (folio_mapping(folio) != mapping) {
1107 /* Page was replaced by swap: retry */
1108 folio_unlock(folio);
1109 index = indices[i];
1110 break;
1111 }
1112 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1113 folio);
1114
1115 if (!folio_test_large(folio)) {
1116 truncate_inode_folio(mapping, folio);
1117 } else if (truncate_inode_partial_folio(folio, lstart, lend)) {
1118 /*
1119 * If we split a page, reset the loop so
1120 * that we pick up the new sub pages.
1121 * Otherwise the THP was entirely
1122 * dropped or the target range was
1123 * zeroed, so just continue the loop as
1124 * is.
1125 */
1126 if (!folio_test_large(folio)) {
1127 folio_unlock(folio);
1128 index = start;
1129 break;
1130 }
1131 }
1132 }
1133 folio_unlock(folio);
1134 }
1135 folio_batch_remove_exceptionals(&fbatch);
1136 folio_batch_release(&fbatch);
1137 }
1138
1139 shmem_recalc_inode(inode, 0, -nr_swaps_freed);
1140 }
1141
shmem_truncate_range(struct inode * inode,loff_t lstart,loff_t lend)1142 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1143 {
1144 shmem_undo_range(inode, lstart, lend, false);
1145 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1146 inode_inc_iversion(inode);
1147 }
1148 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1149
shmem_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)1150 static int shmem_getattr(struct mnt_idmap *idmap,
1151 const struct path *path, struct kstat *stat,
1152 u32 request_mask, unsigned int query_flags)
1153 {
1154 struct inode *inode = path->dentry->d_inode;
1155 struct shmem_inode_info *info = SHMEM_I(inode);
1156
1157 if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1158 shmem_recalc_inode(inode, 0, 0);
1159
1160 if (info->fsflags & FS_APPEND_FL)
1161 stat->attributes |= STATX_ATTR_APPEND;
1162 if (info->fsflags & FS_IMMUTABLE_FL)
1163 stat->attributes |= STATX_ATTR_IMMUTABLE;
1164 if (info->fsflags & FS_NODUMP_FL)
1165 stat->attributes |= STATX_ATTR_NODUMP;
1166 stat->attributes_mask |= (STATX_ATTR_APPEND |
1167 STATX_ATTR_IMMUTABLE |
1168 STATX_ATTR_NODUMP);
1169 generic_fillattr(idmap, request_mask, inode, stat);
1170
1171 if (shmem_huge_global_enabled(inode, 0, 0, false, NULL, 0))
1172 stat->blksize = HPAGE_PMD_SIZE;
1173
1174 if (request_mask & STATX_BTIME) {
1175 stat->result_mask |= STATX_BTIME;
1176 stat->btime.tv_sec = info->i_crtime.tv_sec;
1177 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1178 }
1179
1180 return 0;
1181 }
1182
shmem_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)1183 static int shmem_setattr(struct mnt_idmap *idmap,
1184 struct dentry *dentry, struct iattr *attr)
1185 {
1186 struct inode *inode = d_inode(dentry);
1187 struct shmem_inode_info *info = SHMEM_I(inode);
1188 int error;
1189 bool update_mtime = false;
1190 bool update_ctime = true;
1191
1192 error = setattr_prepare(idmap, dentry, attr);
1193 if (error)
1194 return error;
1195
1196 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1197 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1198 return -EPERM;
1199 }
1200 }
1201
1202 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1203 loff_t oldsize = inode->i_size;
1204 loff_t newsize = attr->ia_size;
1205
1206 /* protected by i_rwsem */
1207 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1208 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1209 return -EPERM;
1210
1211 if (newsize != oldsize) {
1212 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1213 oldsize, newsize);
1214 if (error)
1215 return error;
1216 i_size_write(inode, newsize);
1217 update_mtime = true;
1218 } else {
1219 update_ctime = false;
1220 }
1221 if (newsize <= oldsize) {
1222 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1223 if (oldsize > holebegin)
1224 unmap_mapping_range(inode->i_mapping,
1225 holebegin, 0, 1);
1226 if (info->alloced)
1227 shmem_truncate_range(inode,
1228 newsize, (loff_t)-1);
1229 /* unmap again to remove racily COWed private pages */
1230 if (oldsize > holebegin)
1231 unmap_mapping_range(inode->i_mapping,
1232 holebegin, 0, 1);
1233 }
1234 }
1235
1236 if (is_quota_modification(idmap, inode, attr)) {
1237 error = dquot_initialize(inode);
1238 if (error)
1239 return error;
1240 }
1241
1242 /* Transfer quota accounting */
1243 if (i_uid_needs_update(idmap, attr, inode) ||
1244 i_gid_needs_update(idmap, attr, inode)) {
1245 error = dquot_transfer(idmap, inode, attr);
1246 if (error)
1247 return error;
1248 }
1249
1250 setattr_copy(idmap, inode, attr);
1251 if (attr->ia_valid & ATTR_MODE)
1252 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1253 if (!error && update_ctime) {
1254 inode_set_ctime_current(inode);
1255 if (update_mtime)
1256 inode_set_mtime_to_ts(inode, inode_get_ctime(inode));
1257 inode_inc_iversion(inode);
1258 }
1259 return error;
1260 }
1261
shmem_evict_inode(struct inode * inode)1262 static void shmem_evict_inode(struct inode *inode)
1263 {
1264 struct shmem_inode_info *info = SHMEM_I(inode);
1265 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1266 size_t freed = 0;
1267
1268 if (shmem_mapping(inode->i_mapping)) {
1269 shmem_unacct_size(info->flags, inode->i_size);
1270 inode->i_size = 0;
1271 mapping_set_exiting(inode->i_mapping);
1272 shmem_truncate_range(inode, 0, (loff_t)-1);
1273 if (!list_empty(&info->shrinklist)) {
1274 spin_lock(&sbinfo->shrinklist_lock);
1275 if (!list_empty(&info->shrinklist)) {
1276 list_del_init(&info->shrinklist);
1277 sbinfo->shrinklist_len--;
1278 }
1279 spin_unlock(&sbinfo->shrinklist_lock);
1280 }
1281 while (!list_empty(&info->swaplist)) {
1282 /* Wait while shmem_unuse() is scanning this inode... */
1283 wait_var_event(&info->stop_eviction,
1284 !atomic_read(&info->stop_eviction));
1285 mutex_lock(&shmem_swaplist_mutex);
1286 /* ...but beware of the race if we peeked too early */
1287 if (!atomic_read(&info->stop_eviction))
1288 list_del_init(&info->swaplist);
1289 mutex_unlock(&shmem_swaplist_mutex);
1290 }
1291 }
1292
1293 simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL);
1294 shmem_free_inode(inode->i_sb, freed);
1295 WARN_ON(inode->i_blocks);
1296 clear_inode(inode);
1297 #ifdef CONFIG_TMPFS_QUOTA
1298 dquot_free_inode(inode);
1299 dquot_drop(inode);
1300 #endif
1301 }
1302
shmem_find_swap_entries(struct address_space * mapping,pgoff_t start,struct folio_batch * fbatch,pgoff_t * indices,unsigned int type)1303 static int shmem_find_swap_entries(struct address_space *mapping,
1304 pgoff_t start, struct folio_batch *fbatch,
1305 pgoff_t *indices, unsigned int type)
1306 {
1307 XA_STATE(xas, &mapping->i_pages, start);
1308 struct folio *folio;
1309 swp_entry_t entry;
1310
1311 rcu_read_lock();
1312 xas_for_each(&xas, folio, ULONG_MAX) {
1313 if (xas_retry(&xas, folio))
1314 continue;
1315
1316 if (!xa_is_value(folio))
1317 continue;
1318
1319 entry = radix_to_swp_entry(folio);
1320 /*
1321 * swapin error entries can be found in the mapping. But they're
1322 * deliberately ignored here as we've done everything we can do.
1323 */
1324 if (swp_type(entry) != type)
1325 continue;
1326
1327 indices[folio_batch_count(fbatch)] = xas.xa_index;
1328 if (!folio_batch_add(fbatch, folio))
1329 break;
1330
1331 if (need_resched()) {
1332 xas_pause(&xas);
1333 cond_resched_rcu();
1334 }
1335 }
1336 rcu_read_unlock();
1337
1338 return xas.xa_index;
1339 }
1340
1341 /*
1342 * Move the swapped pages for an inode to page cache. Returns the count
1343 * of pages swapped in, or the error in case of failure.
1344 */
shmem_unuse_swap_entries(struct inode * inode,struct folio_batch * fbatch,pgoff_t * indices)1345 static int shmem_unuse_swap_entries(struct inode *inode,
1346 struct folio_batch *fbatch, pgoff_t *indices)
1347 {
1348 int i = 0;
1349 int ret = 0;
1350 int error = 0;
1351 struct address_space *mapping = inode->i_mapping;
1352
1353 for (i = 0; i < folio_batch_count(fbatch); i++) {
1354 struct folio *folio = fbatch->folios[i];
1355
1356 if (!xa_is_value(folio))
1357 continue;
1358 error = shmem_swapin_folio(inode, indices[i], &folio, SGP_CACHE,
1359 mapping_gfp_mask(mapping), NULL, NULL);
1360 if (error == 0) {
1361 folio_unlock(folio);
1362 folio_put(folio);
1363 ret++;
1364 }
1365 if (error == -ENOMEM)
1366 break;
1367 error = 0;
1368 }
1369 return error ? error : ret;
1370 }
1371
1372 /*
1373 * If swap found in inode, free it and move page from swapcache to filecache.
1374 */
shmem_unuse_inode(struct inode * inode,unsigned int type)1375 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1376 {
1377 struct address_space *mapping = inode->i_mapping;
1378 pgoff_t start = 0;
1379 struct folio_batch fbatch;
1380 pgoff_t indices[PAGEVEC_SIZE];
1381 int ret = 0;
1382
1383 do {
1384 folio_batch_init(&fbatch);
1385 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1386 if (folio_batch_count(&fbatch) == 0) {
1387 ret = 0;
1388 break;
1389 }
1390
1391 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1392 if (ret < 0)
1393 break;
1394
1395 start = indices[folio_batch_count(&fbatch) - 1];
1396 } while (true);
1397
1398 return ret;
1399 }
1400
1401 /*
1402 * Read all the shared memory data that resides in the swap
1403 * device 'type' back into memory, so the swap device can be
1404 * unused.
1405 */
shmem_unuse(unsigned int type)1406 int shmem_unuse(unsigned int type)
1407 {
1408 struct shmem_inode_info *info, *next;
1409 int error = 0;
1410
1411 if (list_empty(&shmem_swaplist))
1412 return 0;
1413
1414 mutex_lock(&shmem_swaplist_mutex);
1415 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1416 if (!info->swapped) {
1417 list_del_init(&info->swaplist);
1418 continue;
1419 }
1420 /*
1421 * Drop the swaplist mutex while searching the inode for swap;
1422 * but before doing so, make sure shmem_evict_inode() will not
1423 * remove placeholder inode from swaplist, nor let it be freed
1424 * (igrab() would protect from unlink, but not from unmount).
1425 */
1426 atomic_inc(&info->stop_eviction);
1427 mutex_unlock(&shmem_swaplist_mutex);
1428
1429 error = shmem_unuse_inode(&info->vfs_inode, type);
1430 cond_resched();
1431
1432 mutex_lock(&shmem_swaplist_mutex);
1433 next = list_next_entry(info, swaplist);
1434 if (!info->swapped)
1435 list_del_init(&info->swaplist);
1436 if (atomic_dec_and_test(&info->stop_eviction))
1437 wake_up_var(&info->stop_eviction);
1438 if (error)
1439 break;
1440 }
1441 mutex_unlock(&shmem_swaplist_mutex);
1442
1443 return error;
1444 }
1445
1446 /*
1447 * Move the page from the page cache to the swap cache.
1448 */
shmem_writepage(struct page * page,struct writeback_control * wbc)1449 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1450 {
1451 struct folio *folio = page_folio(page);
1452 struct address_space *mapping = folio->mapping;
1453 struct inode *inode = mapping->host;
1454 struct shmem_inode_info *info = SHMEM_I(inode);
1455 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1456 swp_entry_t swap;
1457 pgoff_t index;
1458 int nr_pages;
1459 bool split = false;
1460
1461 /*
1462 * Our capabilities prevent regular writeback or sync from ever calling
1463 * shmem_writepage; but a stacking filesystem might use ->writepage of
1464 * its underlying filesystem, in which case tmpfs should write out to
1465 * swap only in response to memory pressure, and not for the writeback
1466 * threads or sync.
1467 */
1468 if (WARN_ON_ONCE(!wbc->for_reclaim))
1469 goto redirty;
1470
1471 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1472 goto redirty;
1473
1474 if (!total_swap_pages)
1475 goto redirty;
1476
1477 /*
1478 * If CONFIG_THP_SWAP is not enabled, the large folio should be
1479 * split when swapping.
1480 *
1481 * And shrinkage of pages beyond i_size does not split swap, so
1482 * swapout of a large folio crossing i_size needs to split too
1483 * (unless fallocate has been used to preallocate beyond EOF).
1484 */
1485 if (folio_test_large(folio)) {
1486 index = shmem_fallocend(inode,
1487 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE));
1488 if ((index > folio->index && index < folio_next_index(folio)) ||
1489 !IS_ENABLED(CONFIG_THP_SWAP))
1490 split = true;
1491 }
1492
1493 if (split) {
1494 try_split:
1495 /* Ensure the subpages are still dirty */
1496 folio_test_set_dirty(folio);
1497 if (split_huge_page_to_list_to_order(page, wbc->list, 0))
1498 goto redirty;
1499 folio = page_folio(page);
1500 folio_clear_dirty(folio);
1501 }
1502
1503 index = folio->index;
1504 nr_pages = folio_nr_pages(folio);
1505
1506 /*
1507 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1508 * value into swapfile.c, the only way we can correctly account for a
1509 * fallocated folio arriving here is now to initialize it and write it.
1510 *
1511 * That's okay for a folio already fallocated earlier, but if we have
1512 * not yet completed the fallocation, then (a) we want to keep track
1513 * of this folio in case we have to undo it, and (b) it may not be a
1514 * good idea to continue anyway, once we're pushing into swap. So
1515 * reactivate the folio, and let shmem_fallocate() quit when too many.
1516 */
1517 if (!folio_test_uptodate(folio)) {
1518 if (inode->i_private) {
1519 struct shmem_falloc *shmem_falloc;
1520 spin_lock(&inode->i_lock);
1521 shmem_falloc = inode->i_private;
1522 if (shmem_falloc &&
1523 !shmem_falloc->waitq &&
1524 index >= shmem_falloc->start &&
1525 index < shmem_falloc->next)
1526 shmem_falloc->nr_unswapped++;
1527 else
1528 shmem_falloc = NULL;
1529 spin_unlock(&inode->i_lock);
1530 if (shmem_falloc)
1531 goto redirty;
1532 }
1533 folio_zero_range(folio, 0, folio_size(folio));
1534 flush_dcache_folio(folio);
1535 folio_mark_uptodate(folio);
1536 }
1537
1538 swap = folio_alloc_swap(folio);
1539 if (!swap.val) {
1540 if (nr_pages > 1)
1541 goto try_split;
1542
1543 goto redirty;
1544 }
1545
1546 /*
1547 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1548 * if it's not already there. Do it now before the folio is
1549 * moved to swap cache, when its pagelock no longer protects
1550 * the inode from eviction. But don't unlock the mutex until
1551 * we've incremented swapped, because shmem_unuse_inode() will
1552 * prune a !swapped inode from the swaplist under this mutex.
1553 */
1554 mutex_lock(&shmem_swaplist_mutex);
1555 if (list_empty(&info->swaplist))
1556 list_add(&info->swaplist, &shmem_swaplist);
1557
1558 if (add_to_swap_cache(folio, swap,
1559 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1560 NULL) == 0) {
1561 shmem_recalc_inode(inode, 0, nr_pages);
1562 swap_shmem_alloc(swap, nr_pages);
1563 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1564
1565 mutex_unlock(&shmem_swaplist_mutex);
1566 BUG_ON(folio_mapped(folio));
1567 return swap_writepage(&folio->page, wbc);
1568 }
1569
1570 mutex_unlock(&shmem_swaplist_mutex);
1571 put_swap_folio(folio, swap);
1572 redirty:
1573 folio_mark_dirty(folio);
1574 if (wbc->for_reclaim)
1575 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1576 folio_unlock(folio);
1577 return 0;
1578 }
1579
1580 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
shmem_show_mpol(struct seq_file * seq,struct mempolicy * mpol)1581 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1582 {
1583 char buffer[64];
1584
1585 if (!mpol || mpol->mode == MPOL_DEFAULT)
1586 return; /* show nothing */
1587
1588 mpol_to_str(buffer, sizeof(buffer), mpol);
1589
1590 seq_printf(seq, ",mpol=%s", buffer);
1591 }
1592
shmem_get_sbmpol(struct shmem_sb_info * sbinfo)1593 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1594 {
1595 struct mempolicy *mpol = NULL;
1596 if (sbinfo->mpol) {
1597 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1598 mpol = sbinfo->mpol;
1599 mpol_get(mpol);
1600 raw_spin_unlock(&sbinfo->stat_lock);
1601 }
1602 return mpol;
1603 }
1604 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
shmem_show_mpol(struct seq_file * seq,struct mempolicy * mpol)1605 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1606 {
1607 }
shmem_get_sbmpol(struct shmem_sb_info * sbinfo)1608 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1609 {
1610 return NULL;
1611 }
1612 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1613
1614 static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
1615 pgoff_t index, unsigned int order, pgoff_t *ilx);
1616
shmem_swapin_cluster(swp_entry_t swap,gfp_t gfp,struct shmem_inode_info * info,pgoff_t index)1617 static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp,
1618 struct shmem_inode_info *info, pgoff_t index)
1619 {
1620 struct mempolicy *mpol;
1621 pgoff_t ilx;
1622 struct folio *folio;
1623
1624 mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
1625 folio = swap_cluster_readahead(swap, gfp, mpol, ilx);
1626 mpol_cond_put(mpol);
1627
1628 return folio;
1629 }
1630
1631 /*
1632 * Make sure huge_gfp is always more limited than limit_gfp.
1633 * Some of the flags set permissions, while others set limitations.
1634 */
limit_gfp_mask(gfp_t huge_gfp,gfp_t limit_gfp)1635 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1636 {
1637 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1638 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1639 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1640 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1641
1642 /* Allow allocations only from the originally specified zones. */
1643 result |= zoneflags;
1644
1645 /*
1646 * Minimize the result gfp by taking the union with the deny flags,
1647 * and the intersection of the allow flags.
1648 */
1649 result |= (limit_gfp & denyflags);
1650 result |= (huge_gfp & limit_gfp) & allowflags;
1651
1652 return result;
1653 }
1654
1655 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
shmem_allowable_huge_orders(struct inode * inode,struct vm_area_struct * vma,pgoff_t index,loff_t write_end,bool shmem_huge_force)1656 unsigned long shmem_allowable_huge_orders(struct inode *inode,
1657 struct vm_area_struct *vma, pgoff_t index,
1658 loff_t write_end, bool shmem_huge_force)
1659 {
1660 unsigned long mask = READ_ONCE(huge_shmem_orders_always);
1661 unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
1662 unsigned long vm_flags = vma ? vma->vm_flags : 0;
1663 bool global_huge;
1664 loff_t i_size;
1665 int order;
1666
1667 if (vma && ((vm_flags & VM_NOHUGEPAGE) ||
1668 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
1669 return 0;
1670
1671 /* If the hardware/firmware marked hugepage support disabled. */
1672 if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED))
1673 return 0;
1674
1675 global_huge = shmem_huge_global_enabled(inode, index, write_end,
1676 shmem_huge_force, vma, vm_flags);
1677 if (!vma || !vma_is_anon_shmem(vma)) {
1678 /*
1679 * For tmpfs, we now only support PMD sized THP if huge page
1680 * is enabled, otherwise fallback to order 0.
1681 */
1682 return global_huge ? BIT(HPAGE_PMD_ORDER) : 0;
1683 }
1684
1685 /*
1686 * Following the 'deny' semantics of the top level, force the huge
1687 * option off from all mounts.
1688 */
1689 if (shmem_huge == SHMEM_HUGE_DENY)
1690 return 0;
1691
1692 /*
1693 * Only allow inherit orders if the top-level value is 'force', which
1694 * means non-PMD sized THP can not override 'huge' mount option now.
1695 */
1696 if (shmem_huge == SHMEM_HUGE_FORCE)
1697 return READ_ONCE(huge_shmem_orders_inherit);
1698
1699 /* Allow mTHP that will be fully within i_size. */
1700 order = highest_order(within_size_orders);
1701 while (within_size_orders) {
1702 index = round_up(index + 1, order);
1703 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1704 if (i_size >> PAGE_SHIFT >= index) {
1705 mask |= within_size_orders;
1706 break;
1707 }
1708
1709 order = next_order(&within_size_orders, order);
1710 }
1711
1712 if (vm_flags & VM_HUGEPAGE)
1713 mask |= READ_ONCE(huge_shmem_orders_madvise);
1714
1715 if (global_huge)
1716 mask |= READ_ONCE(huge_shmem_orders_inherit);
1717
1718 return THP_ORDERS_ALL_FILE_DEFAULT & mask;
1719 }
1720
shmem_suitable_orders(struct inode * inode,struct vm_fault * vmf,struct address_space * mapping,pgoff_t index,unsigned long orders)1721 static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
1722 struct address_space *mapping, pgoff_t index,
1723 unsigned long orders)
1724 {
1725 struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
1726 pgoff_t aligned_index;
1727 unsigned long pages;
1728 int order;
1729
1730 if (vma) {
1731 orders = thp_vma_suitable_orders(vma, vmf->address, orders);
1732 if (!orders)
1733 return 0;
1734 }
1735
1736 /* Find the highest order that can add into the page cache */
1737 order = highest_order(orders);
1738 while (orders) {
1739 pages = 1UL << order;
1740 aligned_index = round_down(index, pages);
1741 /*
1742 * Check for conflict before waiting on a huge allocation.
1743 * Conflict might be that a huge page has just been allocated
1744 * and added to page cache by a racing thread, or that there
1745 * is already at least one small page in the huge extent.
1746 * Be careful to retry when appropriate, but not forever!
1747 * Elsewhere -EEXIST would be the right code, but not here.
1748 */
1749 if (!xa_find(&mapping->i_pages, &aligned_index,
1750 aligned_index + pages - 1, XA_PRESENT))
1751 break;
1752 order = next_order(&orders, order);
1753 }
1754
1755 return orders;
1756 }
1757 #else
shmem_suitable_orders(struct inode * inode,struct vm_fault * vmf,struct address_space * mapping,pgoff_t index,unsigned long orders)1758 static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
1759 struct address_space *mapping, pgoff_t index,
1760 unsigned long orders)
1761 {
1762 return 0;
1763 }
1764 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1765
shmem_alloc_folio(gfp_t gfp,int order,struct shmem_inode_info * info,pgoff_t index)1766 static struct folio *shmem_alloc_folio(gfp_t gfp, int order,
1767 struct shmem_inode_info *info, pgoff_t index)
1768 {
1769 struct mempolicy *mpol;
1770 pgoff_t ilx;
1771 struct folio *folio;
1772
1773 mpol = shmem_get_pgoff_policy(info, index, order, &ilx);
1774 folio = folio_alloc_mpol(gfp, order, mpol, ilx, numa_node_id());
1775 mpol_cond_put(mpol);
1776
1777 return folio;
1778 }
1779
shmem_alloc_and_add_folio(struct vm_fault * vmf,gfp_t gfp,struct inode * inode,pgoff_t index,struct mm_struct * fault_mm,unsigned long orders)1780 static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
1781 gfp_t gfp, struct inode *inode, pgoff_t index,
1782 struct mm_struct *fault_mm, unsigned long orders)
1783 {
1784 struct address_space *mapping = inode->i_mapping;
1785 struct shmem_inode_info *info = SHMEM_I(inode);
1786 unsigned long suitable_orders = 0;
1787 struct folio *folio = NULL;
1788 long pages;
1789 int error, order;
1790
1791 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1792 orders = 0;
1793
1794 if (orders > 0) {
1795 suitable_orders = shmem_suitable_orders(inode, vmf,
1796 mapping, index, orders);
1797
1798 order = highest_order(suitable_orders);
1799 while (suitable_orders) {
1800 pages = 1UL << order;
1801 index = round_down(index, pages);
1802 folio = shmem_alloc_folio(gfp, order, info, index);
1803 if (folio)
1804 goto allocated;
1805
1806 if (pages == HPAGE_PMD_NR)
1807 count_vm_event(THP_FILE_FALLBACK);
1808 count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK);
1809 order = next_order(&suitable_orders, order);
1810 }
1811 } else {
1812 pages = 1;
1813 folio = shmem_alloc_folio(gfp, 0, info, index);
1814 }
1815 if (!folio)
1816 return ERR_PTR(-ENOMEM);
1817
1818 allocated:
1819 __folio_set_locked(folio);
1820 __folio_set_swapbacked(folio);
1821
1822 gfp &= GFP_RECLAIM_MASK;
1823 error = mem_cgroup_charge(folio, fault_mm, gfp);
1824 if (error) {
1825 if (xa_find(&mapping->i_pages, &index,
1826 index + pages - 1, XA_PRESENT)) {
1827 error = -EEXIST;
1828 } else if (pages > 1) {
1829 if (pages == HPAGE_PMD_NR) {
1830 count_vm_event(THP_FILE_FALLBACK);
1831 count_vm_event(THP_FILE_FALLBACK_CHARGE);
1832 }
1833 count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK);
1834 count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE);
1835 }
1836 goto unlock;
1837 }
1838
1839 error = shmem_add_to_page_cache(folio, mapping, index, NULL, gfp);
1840 if (error)
1841 goto unlock;
1842
1843 error = shmem_inode_acct_blocks(inode, pages);
1844 if (error) {
1845 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1846 long freed;
1847 /*
1848 * Try to reclaim some space by splitting a few
1849 * large folios beyond i_size on the filesystem.
1850 */
1851 shmem_unused_huge_shrink(sbinfo, NULL, pages);
1852 /*
1853 * And do a shmem_recalc_inode() to account for freed pages:
1854 * except our folio is there in cache, so not quite balanced.
1855 */
1856 spin_lock(&info->lock);
1857 freed = pages + info->alloced - info->swapped -
1858 READ_ONCE(mapping->nrpages);
1859 if (freed > 0)
1860 info->alloced -= freed;
1861 spin_unlock(&info->lock);
1862 if (freed > 0)
1863 shmem_inode_unacct_blocks(inode, freed);
1864 error = shmem_inode_acct_blocks(inode, pages);
1865 if (error) {
1866 filemap_remove_folio(folio);
1867 goto unlock;
1868 }
1869 }
1870
1871 shmem_recalc_inode(inode, pages, 0);
1872 folio_add_lru(folio);
1873 return folio;
1874
1875 unlock:
1876 folio_unlock(folio);
1877 folio_put(folio);
1878 return ERR_PTR(error);
1879 }
1880
1881 /*
1882 * When a page is moved from swapcache to shmem filecache (either by the
1883 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1884 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1885 * ignorance of the mapping it belongs to. If that mapping has special
1886 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1887 * we may need to copy to a suitable page before moving to filecache.
1888 *
1889 * In a future release, this may well be extended to respect cpuset and
1890 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1891 * but for now it is a simple matter of zone.
1892 */
shmem_should_replace_folio(struct folio * folio,gfp_t gfp)1893 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1894 {
1895 return folio_zonenum(folio) > gfp_zone(gfp);
1896 }
1897
shmem_replace_folio(struct folio ** foliop,gfp_t gfp,struct shmem_inode_info * info,pgoff_t index,struct vm_area_struct * vma)1898 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1899 struct shmem_inode_info *info, pgoff_t index,
1900 struct vm_area_struct *vma)
1901 {
1902 struct folio *new, *old = *foliop;
1903 swp_entry_t entry = old->swap;
1904 struct address_space *swap_mapping = swap_address_space(entry);
1905 pgoff_t swap_index = swap_cache_index(entry);
1906 XA_STATE(xas, &swap_mapping->i_pages, swap_index);
1907 int nr_pages = folio_nr_pages(old);
1908 int error = 0, i;
1909
1910 /*
1911 * We have arrived here because our zones are constrained, so don't
1912 * limit chance of success by further cpuset and node constraints.
1913 */
1914 gfp &= ~GFP_CONSTRAINT_MASK;
1915 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1916 if (nr_pages > 1) {
1917 gfp_t huge_gfp = vma_thp_gfp_mask(vma);
1918
1919 gfp = limit_gfp_mask(huge_gfp, gfp);
1920 }
1921 #endif
1922
1923 new = shmem_alloc_folio(gfp, folio_order(old), info, index);
1924 if (!new)
1925 return -ENOMEM;
1926
1927 folio_ref_add(new, nr_pages);
1928 folio_copy(new, old);
1929 flush_dcache_folio(new);
1930
1931 __folio_set_locked(new);
1932 __folio_set_swapbacked(new);
1933 folio_mark_uptodate(new);
1934 new->swap = entry;
1935 folio_set_swapcache(new);
1936
1937 /* Swap cache still stores N entries instead of a high-order entry */
1938 xa_lock_irq(&swap_mapping->i_pages);
1939 for (i = 0; i < nr_pages; i++) {
1940 void *item = xas_load(&xas);
1941
1942 if (item != old) {
1943 error = -ENOENT;
1944 break;
1945 }
1946
1947 xas_store(&xas, new);
1948 xas_next(&xas);
1949 }
1950 if (!error) {
1951 mem_cgroup_replace_folio(old, new);
1952 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, nr_pages);
1953 __lruvec_stat_mod_folio(new, NR_SHMEM, nr_pages);
1954 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -nr_pages);
1955 __lruvec_stat_mod_folio(old, NR_SHMEM, -nr_pages);
1956 }
1957 xa_unlock_irq(&swap_mapping->i_pages);
1958
1959 if (unlikely(error)) {
1960 /*
1961 * Is this possible? I think not, now that our callers
1962 * check both the swapcache flag and folio->private
1963 * after getting the folio lock; but be defensive.
1964 * Reverse old to newpage for clear and free.
1965 */
1966 old = new;
1967 } else {
1968 folio_add_lru(new);
1969 *foliop = new;
1970 }
1971
1972 folio_clear_swapcache(old);
1973 old->private = NULL;
1974
1975 folio_unlock(old);
1976 /*
1977 * The old folio are removed from swap cache, drop the 'nr_pages'
1978 * reference, as well as one temporary reference getting from swap
1979 * cache.
1980 */
1981 folio_put_refs(old, nr_pages + 1);
1982 return error;
1983 }
1984
shmem_set_folio_swapin_error(struct inode * inode,pgoff_t index,struct folio * folio,swp_entry_t swap)1985 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1986 struct folio *folio, swp_entry_t swap)
1987 {
1988 struct address_space *mapping = inode->i_mapping;
1989 swp_entry_t swapin_error;
1990 void *old;
1991 int nr_pages;
1992
1993 swapin_error = make_poisoned_swp_entry();
1994 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1995 swp_to_radix_entry(swap),
1996 swp_to_radix_entry(swapin_error), 0);
1997 if (old != swp_to_radix_entry(swap))
1998 return;
1999
2000 nr_pages = folio_nr_pages(folio);
2001 folio_wait_writeback(folio);
2002 delete_from_swap_cache(folio);
2003 /*
2004 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
2005 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
2006 * in shmem_evict_inode().
2007 */
2008 shmem_recalc_inode(inode, -nr_pages, -nr_pages);
2009 swap_free_nr(swap, nr_pages);
2010 }
2011
shmem_split_large_entry(struct inode * inode,pgoff_t index,swp_entry_t swap,gfp_t gfp)2012 static int shmem_split_large_entry(struct inode *inode, pgoff_t index,
2013 swp_entry_t swap, gfp_t gfp)
2014 {
2015 struct address_space *mapping = inode->i_mapping;
2016 XA_STATE_ORDER(xas, &mapping->i_pages, index, 0);
2017 void *alloced_shadow = NULL;
2018 int alloced_order = 0, i;
2019
2020 /* Convert user data gfp flags to xarray node gfp flags */
2021 gfp &= GFP_RECLAIM_MASK;
2022
2023 for (;;) {
2024 int order = -1, split_order = 0;
2025 void *old = NULL;
2026
2027 xas_lock_irq(&xas);
2028 old = xas_load(&xas);
2029 if (!xa_is_value(old) || swp_to_radix_entry(swap) != old) {
2030 xas_set_err(&xas, -EEXIST);
2031 goto unlock;
2032 }
2033
2034 order = xas_get_order(&xas);
2035
2036 /* Swap entry may have changed before we re-acquire the lock */
2037 if (alloced_order &&
2038 (old != alloced_shadow || order != alloced_order)) {
2039 xas_destroy(&xas);
2040 alloced_order = 0;
2041 }
2042
2043 /* Try to split large swap entry in pagecache */
2044 if (order > 0) {
2045 if (!alloced_order) {
2046 split_order = order;
2047 goto unlock;
2048 }
2049 xas_split(&xas, old, order);
2050
2051 /*
2052 * Re-set the swap entry after splitting, and the swap
2053 * offset of the original large entry must be continuous.
2054 */
2055 for (i = 0; i < 1 << order; i++) {
2056 pgoff_t aligned_index = round_down(index, 1 << order);
2057 swp_entry_t tmp;
2058
2059 tmp = swp_entry(swp_type(swap), swp_offset(swap) + i);
2060 __xa_store(&mapping->i_pages, aligned_index + i,
2061 swp_to_radix_entry(tmp), 0);
2062 }
2063 }
2064
2065 unlock:
2066 xas_unlock_irq(&xas);
2067
2068 /* split needed, alloc here and retry. */
2069 if (split_order) {
2070 xas_split_alloc(&xas, old, split_order, gfp);
2071 if (xas_error(&xas))
2072 goto error;
2073 alloced_shadow = old;
2074 alloced_order = split_order;
2075 xas_reset(&xas);
2076 continue;
2077 }
2078
2079 if (!xas_nomem(&xas, gfp))
2080 break;
2081 }
2082
2083 error:
2084 if (xas_error(&xas))
2085 return xas_error(&xas);
2086
2087 return alloced_order;
2088 }
2089
2090 /*
2091 * Swap in the folio pointed to by *foliop.
2092 * Caller has to make sure that *foliop contains a valid swapped folio.
2093 * Returns 0 and the folio in foliop if success. On failure, returns the
2094 * error code and NULL in *foliop.
2095 */
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)2096 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
2097 struct folio **foliop, enum sgp_type sgp,
2098 gfp_t gfp, struct vm_area_struct *vma,
2099 vm_fault_t *fault_type)
2100 {
2101 struct address_space *mapping = inode->i_mapping;
2102 struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;
2103 struct shmem_inode_info *info = SHMEM_I(inode);
2104 struct swap_info_struct *si;
2105 struct folio *folio = NULL;
2106 swp_entry_t swap;
2107 int error, nr_pages;
2108
2109 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
2110 swap = radix_to_swp_entry(*foliop);
2111 *foliop = NULL;
2112
2113 if (is_poisoned_swp_entry(swap))
2114 return -EIO;
2115
2116 si = get_swap_device(swap);
2117 if (!si) {
2118 if (!shmem_confirm_swap(mapping, index, swap))
2119 return -EEXIST;
2120 else
2121 return -EINVAL;
2122 }
2123
2124 /* Look it up and read it in.. */
2125 folio = swap_cache_get_folio(swap, NULL, 0);
2126 if (!folio) {
2127 int split_order;
2128
2129 /* Or update major stats only when swapin succeeds?? */
2130 if (fault_type) {
2131 *fault_type |= VM_FAULT_MAJOR;
2132 count_vm_event(PGMAJFAULT);
2133 count_memcg_event_mm(fault_mm, PGMAJFAULT);
2134 }
2135
2136 /*
2137 * Now swap device can only swap in order 0 folio, then we
2138 * should split the large swap entry stored in the pagecache
2139 * if necessary.
2140 */
2141 split_order = shmem_split_large_entry(inode, index, swap, gfp);
2142 if (split_order < 0) {
2143 error = split_order;
2144 goto failed;
2145 }
2146
2147 /*
2148 * If the large swap entry has already been split, it is
2149 * necessary to recalculate the new swap entry based on
2150 * the old order alignment.
2151 */
2152 if (split_order > 0) {
2153 pgoff_t offset = index - round_down(index, 1 << split_order);
2154
2155 swap = swp_entry(swp_type(swap), swp_offset(swap) + offset);
2156 }
2157
2158 /* Here we actually start the io */
2159 folio = shmem_swapin_cluster(swap, gfp, info, index);
2160 if (!folio) {
2161 error = -ENOMEM;
2162 goto failed;
2163 }
2164 }
2165
2166 /* We have to do this with folio locked to prevent races */
2167 folio_lock(folio);
2168 if (!folio_test_swapcache(folio) ||
2169 folio->swap.val != swap.val ||
2170 !shmem_confirm_swap(mapping, index, swap)) {
2171 error = -EEXIST;
2172 goto unlock;
2173 }
2174 if (!folio_test_uptodate(folio)) {
2175 error = -EIO;
2176 goto failed;
2177 }
2178 folio_wait_writeback(folio);
2179 nr_pages = folio_nr_pages(folio);
2180
2181 /*
2182 * Some architectures may have to restore extra metadata to the
2183 * folio after reading from swap.
2184 */
2185 arch_swap_restore(folio_swap(swap, folio), folio);
2186
2187 if (shmem_should_replace_folio(folio, gfp)) {
2188 error = shmem_replace_folio(&folio, gfp, info, index, vma);
2189 if (error)
2190 goto failed;
2191 }
2192
2193 error = shmem_add_to_page_cache(folio, mapping,
2194 round_down(index, nr_pages),
2195 swp_to_radix_entry(swap), gfp);
2196 if (error)
2197 goto failed;
2198
2199 shmem_recalc_inode(inode, 0, -nr_pages);
2200
2201 if (sgp == SGP_WRITE)
2202 folio_mark_accessed(folio);
2203
2204 delete_from_swap_cache(folio);
2205 folio_mark_dirty(folio);
2206 swap_free_nr(swap, nr_pages);
2207 put_swap_device(si);
2208
2209 *foliop = folio;
2210 return 0;
2211 failed:
2212 if (!shmem_confirm_swap(mapping, index, swap))
2213 error = -EEXIST;
2214 if (error == -EIO)
2215 shmem_set_folio_swapin_error(inode, index, folio, swap);
2216 unlock:
2217 if (folio) {
2218 folio_unlock(folio);
2219 folio_put(folio);
2220 }
2221 put_swap_device(si);
2222
2223 return error;
2224 }
2225
2226 /*
2227 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
2228 *
2229 * If we allocate a new one we do not mark it dirty. That's up to the
2230 * vm. If we swap it in we mark it dirty since we also free the swap
2231 * entry since a page cannot live in both the swap and page cache.
2232 *
2233 * vmf and fault_type are only supplied by shmem_fault: otherwise they are NULL.
2234 */
shmem_get_folio_gfp(struct inode * inode,pgoff_t index,loff_t write_end,struct folio ** foliop,enum sgp_type sgp,gfp_t gfp,struct vm_fault * vmf,vm_fault_t * fault_type)2235 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
2236 loff_t write_end, struct folio **foliop, enum sgp_type sgp,
2237 gfp_t gfp, struct vm_fault *vmf, vm_fault_t *fault_type)
2238 {
2239 struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
2240 struct mm_struct *fault_mm;
2241 struct folio *folio;
2242 int error;
2243 bool alloced;
2244 unsigned long orders = 0;
2245
2246 if (WARN_ON_ONCE(!shmem_mapping(inode->i_mapping)))
2247 return -EINVAL;
2248
2249 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
2250 return -EFBIG;
2251 repeat:
2252 if (sgp <= SGP_CACHE &&
2253 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode))
2254 return -EINVAL;
2255
2256 alloced = false;
2257 fault_mm = vma ? vma->vm_mm : NULL;
2258
2259 folio = filemap_get_entry(inode->i_mapping, index);
2260 if (folio && vma && userfaultfd_minor(vma)) {
2261 if (!xa_is_value(folio))
2262 folio_put(folio);
2263 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
2264 return 0;
2265 }
2266
2267 if (xa_is_value(folio)) {
2268 error = shmem_swapin_folio(inode, index, &folio,
2269 sgp, gfp, vma, fault_type);
2270 if (error == -EEXIST)
2271 goto repeat;
2272
2273 *foliop = folio;
2274 return error;
2275 }
2276
2277 if (folio) {
2278 folio_lock(folio);
2279
2280 /* Has the folio been truncated or swapped out? */
2281 if (unlikely(folio->mapping != inode->i_mapping)) {
2282 folio_unlock(folio);
2283 folio_put(folio);
2284 goto repeat;
2285 }
2286 if (sgp == SGP_WRITE)
2287 folio_mark_accessed(folio);
2288 if (folio_test_uptodate(folio))
2289 goto out;
2290 /* fallocated folio */
2291 if (sgp != SGP_READ)
2292 goto clear;
2293 folio_unlock(folio);
2294 folio_put(folio);
2295 }
2296
2297 /*
2298 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
2299 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
2300 */
2301 *foliop = NULL;
2302 if (sgp == SGP_READ)
2303 return 0;
2304 if (sgp == SGP_NOALLOC)
2305 return -ENOENT;
2306
2307 /*
2308 * Fast cache lookup and swap lookup did not find it: allocate.
2309 */
2310
2311 if (vma && userfaultfd_missing(vma)) {
2312 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2313 return 0;
2314 }
2315
2316 /* Find hugepage orders that are allowed for anonymous shmem and tmpfs. */
2317 orders = shmem_allowable_huge_orders(inode, vma, index, write_end, false);
2318 if (orders > 0) {
2319 gfp_t huge_gfp;
2320
2321 huge_gfp = vma_thp_gfp_mask(vma);
2322 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2323 folio = shmem_alloc_and_add_folio(vmf, huge_gfp,
2324 inode, index, fault_mm, orders);
2325 if (!IS_ERR(folio)) {
2326 if (folio_test_pmd_mappable(folio))
2327 count_vm_event(THP_FILE_ALLOC);
2328 count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC);
2329 goto alloced;
2330 }
2331 if (PTR_ERR(folio) == -EEXIST)
2332 goto repeat;
2333 }
2334
2335 folio = shmem_alloc_and_add_folio(vmf, gfp, inode, index, fault_mm, 0);
2336 if (IS_ERR(folio)) {
2337 error = PTR_ERR(folio);
2338 if (error == -EEXIST)
2339 goto repeat;
2340 folio = NULL;
2341 goto unlock;
2342 }
2343
2344 alloced:
2345 alloced = true;
2346 if (folio_test_large(folio) &&
2347 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2348 folio_next_index(folio)) {
2349 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2350 struct shmem_inode_info *info = SHMEM_I(inode);
2351 /*
2352 * Part of the large folio is beyond i_size: subject
2353 * to shrink under memory pressure.
2354 */
2355 spin_lock(&sbinfo->shrinklist_lock);
2356 /*
2357 * _careful to defend against unlocked access to
2358 * ->shrink_list in shmem_unused_huge_shrink()
2359 */
2360 if (list_empty_careful(&info->shrinklist)) {
2361 list_add_tail(&info->shrinklist,
2362 &sbinfo->shrinklist);
2363 sbinfo->shrinklist_len++;
2364 }
2365 spin_unlock(&sbinfo->shrinklist_lock);
2366 }
2367
2368 if (sgp == SGP_WRITE)
2369 folio_set_referenced(folio);
2370 /*
2371 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2372 */
2373 if (sgp == SGP_FALLOC)
2374 sgp = SGP_WRITE;
2375 clear:
2376 /*
2377 * Let SGP_WRITE caller clear ends if write does not fill folio;
2378 * but SGP_FALLOC on a folio fallocated earlier must initialize
2379 * it now, lest undo on failure cancel our earlier guarantee.
2380 */
2381 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2382 long i, n = folio_nr_pages(folio);
2383
2384 for (i = 0; i < n; i++)
2385 clear_highpage(folio_page(folio, i));
2386 flush_dcache_folio(folio);
2387 folio_mark_uptodate(folio);
2388 }
2389
2390 /* Perhaps the file has been truncated since we checked */
2391 if (sgp <= SGP_CACHE &&
2392 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2393 error = -EINVAL;
2394 goto unlock;
2395 }
2396 out:
2397 *foliop = folio;
2398 return 0;
2399
2400 /*
2401 * Error recovery.
2402 */
2403 unlock:
2404 if (alloced)
2405 filemap_remove_folio(folio);
2406 shmem_recalc_inode(inode, 0, 0);
2407 if (folio) {
2408 folio_unlock(folio);
2409 folio_put(folio);
2410 }
2411 return error;
2412 }
2413
2414 /**
2415 * shmem_get_folio - find, and lock a shmem folio.
2416 * @inode: inode to search
2417 * @index: the page index.
2418 * @write_end: end of a write, could extend inode size
2419 * @foliop: pointer to the folio if found
2420 * @sgp: SGP_* flags to control behavior
2421 *
2422 * Looks up the page cache entry at @inode & @index. If a folio is
2423 * present, it is returned locked with an increased refcount.
2424 *
2425 * If the caller modifies data in the folio, it must call folio_mark_dirty()
2426 * before unlocking the folio to ensure that the folio is not reclaimed.
2427 * There is no need to reserve space before calling folio_mark_dirty().
2428 *
2429 * When no folio is found, the behavior depends on @sgp:
2430 * - for SGP_READ, *@foliop is %NULL and 0 is returned
2431 * - for SGP_NOALLOC, *@foliop is %NULL and -ENOENT is returned
2432 * - for all other flags a new folio is allocated, inserted into the
2433 * page cache and returned locked in @foliop.
2434 *
2435 * Context: May sleep.
2436 * Return: 0 if successful, else a negative error code.
2437 */
shmem_get_folio(struct inode * inode,pgoff_t index,loff_t write_end,struct folio ** foliop,enum sgp_type sgp)2438 int shmem_get_folio(struct inode *inode, pgoff_t index, loff_t write_end,
2439 struct folio **foliop, enum sgp_type sgp)
2440 {
2441 return shmem_get_folio_gfp(inode, index, write_end, foliop, sgp,
2442 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
2443 }
2444 EXPORT_SYMBOL_GPL(shmem_get_folio);
2445
2446 /*
2447 * This is like autoremove_wake_function, but it removes the wait queue
2448 * entry unconditionally - even if something else had already woken the
2449 * target.
2450 */
synchronous_wake_function(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)2451 static int synchronous_wake_function(wait_queue_entry_t *wait,
2452 unsigned int mode, int sync, void *key)
2453 {
2454 int ret = default_wake_function(wait, mode, sync, key);
2455 list_del_init(&wait->entry);
2456 return ret;
2457 }
2458
2459 /*
2460 * Trinity finds that probing a hole which tmpfs is punching can
2461 * prevent the hole-punch from ever completing: which in turn
2462 * locks writers out with its hold on i_rwsem. So refrain from
2463 * faulting pages into the hole while it's being punched. Although
2464 * shmem_undo_range() does remove the additions, it may be unable to
2465 * keep up, as each new page needs its own unmap_mapping_range() call,
2466 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2467 *
2468 * It does not matter if we sometimes reach this check just before the
2469 * hole-punch begins, so that one fault then races with the punch:
2470 * we just need to make racing faults a rare case.
2471 *
2472 * The implementation below would be much simpler if we just used a
2473 * standard mutex or completion: but we cannot take i_rwsem in fault,
2474 * and bloating every shmem inode for this unlikely case would be sad.
2475 */
shmem_falloc_wait(struct vm_fault * vmf,struct inode * inode)2476 static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode)
2477 {
2478 struct shmem_falloc *shmem_falloc;
2479 struct file *fpin = NULL;
2480 vm_fault_t ret = 0;
2481
2482 spin_lock(&inode->i_lock);
2483 shmem_falloc = inode->i_private;
2484 if (shmem_falloc &&
2485 shmem_falloc->waitq &&
2486 vmf->pgoff >= shmem_falloc->start &&
2487 vmf->pgoff < shmem_falloc->next) {
2488 wait_queue_head_t *shmem_falloc_waitq;
2489 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2490
2491 ret = VM_FAULT_NOPAGE;
2492 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2493 shmem_falloc_waitq = shmem_falloc->waitq;
2494 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2495 TASK_UNINTERRUPTIBLE);
2496 spin_unlock(&inode->i_lock);
2497 schedule();
2498
2499 /*
2500 * shmem_falloc_waitq points into the shmem_fallocate()
2501 * stack of the hole-punching task: shmem_falloc_waitq
2502 * is usually invalid by the time we reach here, but
2503 * finish_wait() does not dereference it in that case;
2504 * though i_lock needed lest racing with wake_up_all().
2505 */
2506 spin_lock(&inode->i_lock);
2507 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2508 }
2509 spin_unlock(&inode->i_lock);
2510 if (fpin) {
2511 fput(fpin);
2512 ret = VM_FAULT_RETRY;
2513 }
2514 return ret;
2515 }
2516
shmem_fault(struct vm_fault * vmf)2517 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2518 {
2519 struct inode *inode = file_inode(vmf->vma->vm_file);
2520 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2521 struct folio *folio = NULL;
2522 vm_fault_t ret = 0;
2523 int err;
2524
2525 /*
2526 * Trinity finds that probing a hole which tmpfs is punching can
2527 * prevent the hole-punch from ever completing: noted in i_private.
2528 */
2529 if (unlikely(inode->i_private)) {
2530 ret = shmem_falloc_wait(vmf, inode);
2531 if (ret)
2532 return ret;
2533 }
2534
2535 WARN_ON_ONCE(vmf->page != NULL);
2536 err = shmem_get_folio_gfp(inode, vmf->pgoff, 0, &folio, SGP_CACHE,
2537 gfp, vmf, &ret);
2538 if (err)
2539 return vmf_error(err);
2540 if (folio) {
2541 vmf->page = folio_file_page(folio, vmf->pgoff);
2542 ret |= VM_FAULT_LOCKED;
2543 }
2544 return ret;
2545 }
2546
shmem_get_unmapped_area(struct file * file,unsigned long uaddr,unsigned long len,unsigned long pgoff,unsigned long flags)2547 unsigned long shmem_get_unmapped_area(struct file *file,
2548 unsigned long uaddr, unsigned long len,
2549 unsigned long pgoff, unsigned long flags)
2550 {
2551 unsigned long addr;
2552 unsigned long offset;
2553 unsigned long inflated_len;
2554 unsigned long inflated_addr;
2555 unsigned long inflated_offset;
2556 unsigned long hpage_size;
2557
2558 if (len > TASK_SIZE)
2559 return -ENOMEM;
2560
2561 addr = mm_get_unmapped_area(current->mm, file, uaddr, len, pgoff,
2562 flags);
2563
2564 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2565 return addr;
2566 if (IS_ERR_VALUE(addr))
2567 return addr;
2568 if (addr & ~PAGE_MASK)
2569 return addr;
2570 if (addr > TASK_SIZE - len)
2571 return addr;
2572
2573 if (shmem_huge == SHMEM_HUGE_DENY)
2574 return addr;
2575 if (flags & MAP_FIXED)
2576 return addr;
2577 /*
2578 * Our priority is to support MAP_SHARED mapped hugely;
2579 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2580 * But if caller specified an address hint and we allocated area there
2581 * successfully, respect that as before.
2582 */
2583 if (uaddr == addr)
2584 return addr;
2585
2586 hpage_size = HPAGE_PMD_SIZE;
2587 if (shmem_huge != SHMEM_HUGE_FORCE) {
2588 struct super_block *sb;
2589 unsigned long __maybe_unused hpage_orders;
2590 int order = 0;
2591
2592 if (file) {
2593 VM_BUG_ON(file->f_op != &shmem_file_operations);
2594 sb = file_inode(file)->i_sb;
2595 } else {
2596 /*
2597 * Called directly from mm/mmap.c, or drivers/char/mem.c
2598 * for "/dev/zero", to create a shared anonymous object.
2599 */
2600 if (IS_ERR(shm_mnt))
2601 return addr;
2602 sb = shm_mnt->mnt_sb;
2603
2604 /*
2605 * Find the highest mTHP order used for anonymous shmem to
2606 * provide a suitable alignment address.
2607 */
2608 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2609 hpage_orders = READ_ONCE(huge_shmem_orders_always);
2610 hpage_orders |= READ_ONCE(huge_shmem_orders_within_size);
2611 hpage_orders |= READ_ONCE(huge_shmem_orders_madvise);
2612 if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER)
2613 hpage_orders |= READ_ONCE(huge_shmem_orders_inherit);
2614
2615 if (hpage_orders > 0) {
2616 order = highest_order(hpage_orders);
2617 hpage_size = PAGE_SIZE << order;
2618 }
2619 #endif
2620 }
2621 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER && !order)
2622 return addr;
2623 }
2624
2625 if (len < hpage_size)
2626 return addr;
2627
2628 offset = (pgoff << PAGE_SHIFT) & (hpage_size - 1);
2629 if (offset && offset + len < 2 * hpage_size)
2630 return addr;
2631 if ((addr & (hpage_size - 1)) == offset)
2632 return addr;
2633
2634 inflated_len = len + hpage_size - PAGE_SIZE;
2635 if (inflated_len > TASK_SIZE)
2636 return addr;
2637 if (inflated_len < len)
2638 return addr;
2639
2640 inflated_addr = mm_get_unmapped_area(current->mm, NULL, uaddr,
2641 inflated_len, 0, flags);
2642 if (IS_ERR_VALUE(inflated_addr))
2643 return addr;
2644 if (inflated_addr & ~PAGE_MASK)
2645 return addr;
2646
2647 inflated_offset = inflated_addr & (hpage_size - 1);
2648 inflated_addr += offset - inflated_offset;
2649 if (inflated_offset > offset)
2650 inflated_addr += hpage_size;
2651
2652 if (inflated_addr > TASK_SIZE - len)
2653 return addr;
2654 return inflated_addr;
2655 }
2656
2657 #ifdef CONFIG_NUMA
shmem_set_policy(struct vm_area_struct * vma,struct mempolicy * mpol)2658 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2659 {
2660 struct inode *inode = file_inode(vma->vm_file);
2661 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2662 }
2663
shmem_get_policy(struct vm_area_struct * vma,unsigned long addr,pgoff_t * ilx)2664 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2665 unsigned long addr, pgoff_t *ilx)
2666 {
2667 struct inode *inode = file_inode(vma->vm_file);
2668 pgoff_t index;
2669
2670 /*
2671 * Bias interleave by inode number to distribute better across nodes;
2672 * but this interface is independent of which page order is used, so
2673 * supplies only that bias, letting caller apply the offset (adjusted
2674 * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()).
2675 */
2676 *ilx = inode->i_ino;
2677 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2678 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2679 }
2680
shmem_get_pgoff_policy(struct shmem_inode_info * info,pgoff_t index,unsigned int order,pgoff_t * ilx)2681 static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2682 pgoff_t index, unsigned int order, pgoff_t *ilx)
2683 {
2684 struct mempolicy *mpol;
2685
2686 /* Bias interleave by inode number to distribute better across nodes */
2687 *ilx = info->vfs_inode.i_ino + (index >> order);
2688
2689 mpol = mpol_shared_policy_lookup(&info->policy, index);
2690 return mpol ? mpol : get_task_policy(current);
2691 }
2692 #else
shmem_get_pgoff_policy(struct shmem_inode_info * info,pgoff_t index,unsigned int order,pgoff_t * ilx)2693 static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2694 pgoff_t index, unsigned int order, pgoff_t *ilx)
2695 {
2696 *ilx = 0;
2697 return NULL;
2698 }
2699 #endif /* CONFIG_NUMA */
2700
shmem_lock(struct file * file,int lock,struct ucounts * ucounts)2701 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2702 {
2703 struct inode *inode = file_inode(file);
2704 struct shmem_inode_info *info = SHMEM_I(inode);
2705 int retval = -ENOMEM;
2706
2707 /*
2708 * What serializes the accesses to info->flags?
2709 * ipc_lock_object() when called from shmctl_do_lock(),
2710 * no serialization needed when called from shm_destroy().
2711 */
2712 if (lock && !(info->flags & VM_LOCKED)) {
2713 if (!user_shm_lock(inode->i_size, ucounts))
2714 goto out_nomem;
2715 info->flags |= VM_LOCKED;
2716 mapping_set_unevictable(file->f_mapping);
2717 }
2718 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2719 user_shm_unlock(inode->i_size, ucounts);
2720 info->flags &= ~VM_LOCKED;
2721 mapping_clear_unevictable(file->f_mapping);
2722 }
2723 retval = 0;
2724
2725 out_nomem:
2726 return retval;
2727 }
2728
shmem_mmap(struct file * file,struct vm_area_struct * vma)2729 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2730 {
2731 struct inode *inode = file_inode(file);
2732 struct shmem_inode_info *info = SHMEM_I(inode);
2733 int ret;
2734
2735 ret = seal_check_write(info->seals, vma);
2736 if (ret)
2737 return ret;
2738
2739 /* arm64 - allow memory tagging on RAM-based files */
2740 vm_flags_set(vma, VM_MTE_ALLOWED);
2741
2742 file_accessed(file);
2743 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2744 if (inode->i_nlink)
2745 vma->vm_ops = &shmem_vm_ops;
2746 else
2747 vma->vm_ops = &shmem_anon_vm_ops;
2748 return 0;
2749 }
2750
shmem_file_open(struct inode * inode,struct file * file)2751 static int shmem_file_open(struct inode *inode, struct file *file)
2752 {
2753 file->f_mode |= FMODE_CAN_ODIRECT;
2754 return generic_file_open(inode, file);
2755 }
2756
2757 #ifdef CONFIG_TMPFS_XATTR
2758 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2759
2760 /*
2761 * chattr's fsflags are unrelated to extended attributes,
2762 * but tmpfs has chosen to enable them under the same config option.
2763 */
shmem_set_inode_flags(struct inode * inode,unsigned int fsflags)2764 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2765 {
2766 unsigned int i_flags = 0;
2767
2768 if (fsflags & FS_NOATIME_FL)
2769 i_flags |= S_NOATIME;
2770 if (fsflags & FS_APPEND_FL)
2771 i_flags |= S_APPEND;
2772 if (fsflags & FS_IMMUTABLE_FL)
2773 i_flags |= S_IMMUTABLE;
2774 /*
2775 * But FS_NODUMP_FL does not require any action in i_flags.
2776 */
2777 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2778 }
2779 #else
shmem_set_inode_flags(struct inode * inode,unsigned int fsflags)2780 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2781 {
2782 }
2783 #define shmem_initxattrs NULL
2784 #endif
2785
shmem_get_offset_ctx(struct inode * inode)2786 static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2787 {
2788 return &SHMEM_I(inode)->dir_offsets;
2789 }
2790
__shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)2791 static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2792 struct super_block *sb,
2793 struct inode *dir, umode_t mode,
2794 dev_t dev, unsigned long flags)
2795 {
2796 struct inode *inode;
2797 struct shmem_inode_info *info;
2798 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2799 ino_t ino;
2800 int err;
2801
2802 err = shmem_reserve_inode(sb, &ino);
2803 if (err)
2804 return ERR_PTR(err);
2805
2806 inode = new_inode(sb);
2807 if (!inode) {
2808 shmem_free_inode(sb, 0);
2809 return ERR_PTR(-ENOSPC);
2810 }
2811
2812 inode->i_ino = ino;
2813 inode_init_owner(idmap, inode, dir, mode);
2814 inode->i_blocks = 0;
2815 simple_inode_init_ts(inode);
2816 inode->i_generation = get_random_u32();
2817 info = SHMEM_I(inode);
2818 memset(info, 0, (char *)inode - (char *)info);
2819 spin_lock_init(&info->lock);
2820 atomic_set(&info->stop_eviction, 0);
2821 info->seals = F_SEAL_SEAL;
2822 info->flags = flags & VM_NORESERVE;
2823 info->i_crtime = inode_get_mtime(inode);
2824 info->fsflags = (dir == NULL) ? 0 :
2825 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2826 if (info->fsflags)
2827 shmem_set_inode_flags(inode, info->fsflags);
2828 INIT_LIST_HEAD(&info->shrinklist);
2829 INIT_LIST_HEAD(&info->swaplist);
2830 simple_xattrs_init(&info->xattrs);
2831 cache_no_acl(inode);
2832 if (sbinfo->noswap)
2833 mapping_set_unevictable(inode->i_mapping);
2834 mapping_set_large_folios(inode->i_mapping);
2835
2836 switch (mode & S_IFMT) {
2837 default:
2838 inode->i_op = &shmem_special_inode_operations;
2839 init_special_inode(inode, mode, dev);
2840 break;
2841 case S_IFREG:
2842 inode->i_mapping->a_ops = &shmem_aops;
2843 inode->i_op = &shmem_inode_operations;
2844 inode->i_fop = &shmem_file_operations;
2845 mpol_shared_policy_init(&info->policy,
2846 shmem_get_sbmpol(sbinfo));
2847 break;
2848 case S_IFDIR:
2849 inc_nlink(inode);
2850 /* Some things misbehave if size == 0 on a directory */
2851 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2852 inode->i_op = &shmem_dir_inode_operations;
2853 inode->i_fop = &simple_offset_dir_operations;
2854 simple_offset_init(shmem_get_offset_ctx(inode));
2855 break;
2856 case S_IFLNK:
2857 /*
2858 * Must not load anything in the rbtree,
2859 * mpol_free_shared_policy will not be called.
2860 */
2861 mpol_shared_policy_init(&info->policy, NULL);
2862 break;
2863 }
2864
2865 lockdep_annotate_inode_mutex_key(inode);
2866 return inode;
2867 }
2868
2869 #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)2870 static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2871 struct super_block *sb, struct inode *dir,
2872 umode_t mode, dev_t dev, unsigned long flags)
2873 {
2874 int err;
2875 struct inode *inode;
2876
2877 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2878 if (IS_ERR(inode))
2879 return inode;
2880
2881 err = dquot_initialize(inode);
2882 if (err)
2883 goto errout;
2884
2885 err = dquot_alloc_inode(inode);
2886 if (err) {
2887 dquot_drop(inode);
2888 goto errout;
2889 }
2890 return inode;
2891
2892 errout:
2893 inode->i_flags |= S_NOQUOTA;
2894 iput(inode);
2895 return ERR_PTR(err);
2896 }
2897 #else
shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)2898 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2899 struct super_block *sb, struct inode *dir,
2900 umode_t mode, dev_t dev, unsigned long flags)
2901 {
2902 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2903 }
2904 #endif /* CONFIG_TMPFS_QUOTA */
2905
2906 #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)2907 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2908 struct vm_area_struct *dst_vma,
2909 unsigned long dst_addr,
2910 unsigned long src_addr,
2911 uffd_flags_t flags,
2912 struct folio **foliop)
2913 {
2914 struct inode *inode = file_inode(dst_vma->vm_file);
2915 struct shmem_inode_info *info = SHMEM_I(inode);
2916 struct address_space *mapping = inode->i_mapping;
2917 gfp_t gfp = mapping_gfp_mask(mapping);
2918 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2919 void *page_kaddr;
2920 struct folio *folio;
2921 int ret;
2922 pgoff_t max_off;
2923
2924 if (shmem_inode_acct_blocks(inode, 1)) {
2925 /*
2926 * We may have got a page, returned -ENOENT triggering a retry,
2927 * and now we find ourselves with -ENOMEM. Release the page, to
2928 * avoid a BUG_ON in our caller.
2929 */
2930 if (unlikely(*foliop)) {
2931 folio_put(*foliop);
2932 *foliop = NULL;
2933 }
2934 return -ENOMEM;
2935 }
2936
2937 if (!*foliop) {
2938 ret = -ENOMEM;
2939 folio = shmem_alloc_folio(gfp, 0, info, pgoff);
2940 if (!folio)
2941 goto out_unacct_blocks;
2942
2943 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2944 page_kaddr = kmap_local_folio(folio, 0);
2945 /*
2946 * The read mmap_lock is held here. Despite the
2947 * mmap_lock being read recursive a deadlock is still
2948 * possible if a writer has taken a lock. For example:
2949 *
2950 * process A thread 1 takes read lock on own mmap_lock
2951 * process A thread 2 calls mmap, blocks taking write lock
2952 * process B thread 1 takes page fault, read lock on own mmap lock
2953 * process B thread 2 calls mmap, blocks taking write lock
2954 * process A thread 1 blocks taking read lock on process B
2955 * process B thread 1 blocks taking read lock on process A
2956 *
2957 * Disable page faults to prevent potential deadlock
2958 * and retry the copy outside the mmap_lock.
2959 */
2960 pagefault_disable();
2961 ret = copy_from_user(page_kaddr,
2962 (const void __user *)src_addr,
2963 PAGE_SIZE);
2964 pagefault_enable();
2965 kunmap_local(page_kaddr);
2966
2967 /* fallback to copy_from_user outside mmap_lock */
2968 if (unlikely(ret)) {
2969 *foliop = folio;
2970 ret = -ENOENT;
2971 /* don't free the page */
2972 goto out_unacct_blocks;
2973 }
2974
2975 flush_dcache_folio(folio);
2976 } else { /* ZEROPAGE */
2977 clear_user_highpage(&folio->page, dst_addr);
2978 }
2979 } else {
2980 folio = *foliop;
2981 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2982 *foliop = NULL;
2983 }
2984
2985 VM_BUG_ON(folio_test_locked(folio));
2986 VM_BUG_ON(folio_test_swapbacked(folio));
2987 __folio_set_locked(folio);
2988 __folio_set_swapbacked(folio);
2989 __folio_mark_uptodate(folio);
2990
2991 ret = -EFAULT;
2992 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2993 if (unlikely(pgoff >= max_off))
2994 goto out_release;
2995
2996 ret = mem_cgroup_charge(folio, dst_vma->vm_mm, gfp);
2997 if (ret)
2998 goto out_release;
2999 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp);
3000 if (ret)
3001 goto out_release;
3002
3003 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
3004 &folio->page, true, flags);
3005 if (ret)
3006 goto out_delete_from_cache;
3007
3008 shmem_recalc_inode(inode, 1, 0);
3009 folio_unlock(folio);
3010 return 0;
3011 out_delete_from_cache:
3012 filemap_remove_folio(folio);
3013 out_release:
3014 folio_unlock(folio);
3015 folio_put(folio);
3016 out_unacct_blocks:
3017 shmem_inode_unacct_blocks(inode, 1);
3018 return ret;
3019 }
3020 #endif /* CONFIG_USERFAULTFD */
3021
3022 #ifdef CONFIG_TMPFS
3023 static const struct inode_operations shmem_symlink_inode_operations;
3024 static const struct inode_operations shmem_short_symlink_operations;
3025
3026 static int
shmem_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop,void ** fsdata)3027 shmem_write_begin(struct file *file, struct address_space *mapping,
3028 loff_t pos, unsigned len,
3029 struct folio **foliop, void **fsdata)
3030 {
3031 struct inode *inode = mapping->host;
3032 struct shmem_inode_info *info = SHMEM_I(inode);
3033 pgoff_t index = pos >> PAGE_SHIFT;
3034 struct folio *folio;
3035 int ret = 0;
3036
3037 /* i_rwsem is held by caller */
3038 if (unlikely(info->seals & (F_SEAL_GROW |
3039 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
3040 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
3041 return -EPERM;
3042 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
3043 return -EPERM;
3044 }
3045
3046 ret = shmem_get_folio(inode, index, pos + len, &folio, SGP_WRITE);
3047 if (ret)
3048 return ret;
3049
3050 if (folio_test_hwpoison(folio) ||
3051 (folio_test_large(folio) && folio_test_has_hwpoisoned(folio))) {
3052 folio_unlock(folio);
3053 folio_put(folio);
3054 return -EIO;
3055 }
3056
3057 *foliop = folio;
3058 return 0;
3059 }
3060
3061 static int
shmem_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio,void * fsdata)3062 shmem_write_end(struct file *file, struct address_space *mapping,
3063 loff_t pos, unsigned len, unsigned copied,
3064 struct folio *folio, void *fsdata)
3065 {
3066 struct inode *inode = mapping->host;
3067
3068 if (pos + copied > inode->i_size)
3069 i_size_write(inode, pos + copied);
3070
3071 if (!folio_test_uptodate(folio)) {
3072 if (copied < folio_size(folio)) {
3073 size_t from = offset_in_folio(folio, pos);
3074 folio_zero_segments(folio, 0, from,
3075 from + copied, folio_size(folio));
3076 }
3077 folio_mark_uptodate(folio);
3078 }
3079 folio_mark_dirty(folio);
3080 folio_unlock(folio);
3081 folio_put(folio);
3082
3083 return copied;
3084 }
3085
shmem_file_read_iter(struct kiocb * iocb,struct iov_iter * to)3086 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3087 {
3088 struct file *file = iocb->ki_filp;
3089 struct inode *inode = file_inode(file);
3090 struct address_space *mapping = inode->i_mapping;
3091 pgoff_t index;
3092 unsigned long offset;
3093 int error = 0;
3094 ssize_t retval = 0;
3095 loff_t *ppos = &iocb->ki_pos;
3096
3097 index = *ppos >> PAGE_SHIFT;
3098 offset = *ppos & ~PAGE_MASK;
3099
3100 for (;;) {
3101 struct folio *folio = NULL;
3102 struct page *page = NULL;
3103 pgoff_t end_index;
3104 unsigned long nr, ret;
3105 loff_t i_size = i_size_read(inode);
3106
3107 end_index = i_size >> PAGE_SHIFT;
3108 if (index > end_index)
3109 break;
3110 if (index == end_index) {
3111 nr = i_size & ~PAGE_MASK;
3112 if (nr <= offset)
3113 break;
3114 }
3115
3116 error = shmem_get_folio(inode, index, 0, &folio, SGP_READ);
3117 if (error) {
3118 if (error == -EINVAL)
3119 error = 0;
3120 break;
3121 }
3122 if (folio) {
3123 folio_unlock(folio);
3124
3125 page = folio_file_page(folio, index);
3126 if (PageHWPoison(page)) {
3127 folio_put(folio);
3128 error = -EIO;
3129 break;
3130 }
3131 }
3132
3133 /*
3134 * We must evaluate after, since reads (unlike writes)
3135 * are called without i_rwsem protection against truncate
3136 */
3137 nr = PAGE_SIZE;
3138 i_size = i_size_read(inode);
3139 end_index = i_size >> PAGE_SHIFT;
3140 if (index == end_index) {
3141 nr = i_size & ~PAGE_MASK;
3142 if (nr <= offset) {
3143 if (folio)
3144 folio_put(folio);
3145 break;
3146 }
3147 }
3148 nr -= offset;
3149
3150 if (folio) {
3151 /*
3152 * If users can be writing to this page using arbitrary
3153 * virtual addresses, take care about potential aliasing
3154 * before reading the page on the kernel side.
3155 */
3156 if (mapping_writably_mapped(mapping))
3157 flush_dcache_page(page);
3158 /*
3159 * Mark the page accessed if we read the beginning.
3160 */
3161 if (!offset)
3162 folio_mark_accessed(folio);
3163 /*
3164 * Ok, we have the page, and it's up-to-date, so
3165 * now we can copy it to user space...
3166 */
3167 ret = copy_page_to_iter(page, offset, nr, to);
3168 folio_put(folio);
3169
3170 } else if (user_backed_iter(to)) {
3171 /*
3172 * Copy to user tends to be so well optimized, but
3173 * clear_user() not so much, that it is noticeably
3174 * faster to copy the zero page instead of clearing.
3175 */
3176 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
3177 } else {
3178 /*
3179 * But submitting the same page twice in a row to
3180 * splice() - or others? - can result in confusion:
3181 * so don't attempt that optimization on pipes etc.
3182 */
3183 ret = iov_iter_zero(nr, to);
3184 }
3185
3186 retval += ret;
3187 offset += ret;
3188 index += offset >> PAGE_SHIFT;
3189 offset &= ~PAGE_MASK;
3190
3191 if (!iov_iter_count(to))
3192 break;
3193 if (ret < nr) {
3194 error = -EFAULT;
3195 break;
3196 }
3197 cond_resched();
3198 }
3199
3200 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
3201 file_accessed(file);
3202 return retval ? retval : error;
3203 }
3204
shmem_file_write_iter(struct kiocb * iocb,struct iov_iter * from)3205 static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3206 {
3207 struct file *file = iocb->ki_filp;
3208 struct inode *inode = file->f_mapping->host;
3209 ssize_t ret;
3210
3211 inode_lock(inode);
3212 ret = generic_write_checks(iocb, from);
3213 if (ret <= 0)
3214 goto unlock;
3215 ret = file_remove_privs(file);
3216 if (ret)
3217 goto unlock;
3218 ret = file_update_time(file);
3219 if (ret)
3220 goto unlock;
3221 ret = generic_perform_write(iocb, from);
3222 unlock:
3223 inode_unlock(inode);
3224 return ret;
3225 }
3226
zero_pipe_buf_get(struct pipe_inode_info * pipe,struct pipe_buffer * buf)3227 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
3228 struct pipe_buffer *buf)
3229 {
3230 return true;
3231 }
3232
zero_pipe_buf_release(struct pipe_inode_info * pipe,struct pipe_buffer * buf)3233 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
3234 struct pipe_buffer *buf)
3235 {
3236 }
3237
zero_pipe_buf_try_steal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)3238 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
3239 struct pipe_buffer *buf)
3240 {
3241 return false;
3242 }
3243
3244 static const struct pipe_buf_operations zero_pipe_buf_ops = {
3245 .release = zero_pipe_buf_release,
3246 .try_steal = zero_pipe_buf_try_steal,
3247 .get = zero_pipe_buf_get,
3248 };
3249
splice_zeropage_into_pipe(struct pipe_inode_info * pipe,loff_t fpos,size_t size)3250 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
3251 loff_t fpos, size_t size)
3252 {
3253 size_t offset = fpos & ~PAGE_MASK;
3254
3255 size = min_t(size_t, size, PAGE_SIZE - offset);
3256
3257 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
3258 struct pipe_buffer *buf = pipe_head_buf(pipe);
3259
3260 *buf = (struct pipe_buffer) {
3261 .ops = &zero_pipe_buf_ops,
3262 .page = ZERO_PAGE(0),
3263 .offset = offset,
3264 .len = size,
3265 };
3266 pipe->head++;
3267 }
3268
3269 return size;
3270 }
3271
shmem_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)3272 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
3273 struct pipe_inode_info *pipe,
3274 size_t len, unsigned int flags)
3275 {
3276 struct inode *inode = file_inode(in);
3277 struct address_space *mapping = inode->i_mapping;
3278 struct folio *folio = NULL;
3279 size_t total_spliced = 0, used, npages, n, part;
3280 loff_t isize;
3281 int error = 0;
3282
3283 /* Work out how much data we can actually add into the pipe */
3284 used = pipe_occupancy(pipe->head, pipe->tail);
3285 npages = max_t(ssize_t, pipe->max_usage - used, 0);
3286 len = min_t(size_t, len, npages * PAGE_SIZE);
3287
3288 do {
3289 if (*ppos >= i_size_read(inode))
3290 break;
3291
3292 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, 0, &folio,
3293 SGP_READ);
3294 if (error) {
3295 if (error == -EINVAL)
3296 error = 0;
3297 break;
3298 }
3299 if (folio) {
3300 folio_unlock(folio);
3301
3302 if (folio_test_hwpoison(folio) ||
3303 (folio_test_large(folio) &&
3304 folio_test_has_hwpoisoned(folio))) {
3305 error = -EIO;
3306 break;
3307 }
3308 }
3309
3310 /*
3311 * i_size must be checked after we know the pages are Uptodate.
3312 *
3313 * Checking i_size after the check allows us to calculate
3314 * the correct value for "nr", which means the zero-filled
3315 * part of the page is not copied back to userspace (unless
3316 * another truncate extends the file - this is desired though).
3317 */
3318 isize = i_size_read(inode);
3319 if (unlikely(*ppos >= isize))
3320 break;
3321 part = min_t(loff_t, isize - *ppos, len);
3322
3323 if (folio) {
3324 /*
3325 * If users can be writing to this page using arbitrary
3326 * virtual addresses, take care about potential aliasing
3327 * before reading the page on the kernel side.
3328 */
3329 if (mapping_writably_mapped(mapping))
3330 flush_dcache_folio(folio);
3331 folio_mark_accessed(folio);
3332 /*
3333 * Ok, we have the page, and it's up-to-date, so we can
3334 * now splice it into the pipe.
3335 */
3336 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
3337 folio_put(folio);
3338 folio = NULL;
3339 } else {
3340 n = splice_zeropage_into_pipe(pipe, *ppos, part);
3341 }
3342
3343 if (!n)
3344 break;
3345 len -= n;
3346 total_spliced += n;
3347 *ppos += n;
3348 in->f_ra.prev_pos = *ppos;
3349 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
3350 break;
3351
3352 cond_resched();
3353 } while (len);
3354
3355 if (folio)
3356 folio_put(folio);
3357
3358 file_accessed(in);
3359 return total_spliced ? total_spliced : error;
3360 }
3361
shmem_file_llseek(struct file * file,loff_t offset,int whence)3362 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
3363 {
3364 struct address_space *mapping = file->f_mapping;
3365 struct inode *inode = mapping->host;
3366
3367 if (whence != SEEK_DATA && whence != SEEK_HOLE)
3368 return generic_file_llseek_size(file, offset, whence,
3369 MAX_LFS_FILESIZE, i_size_read(inode));
3370 if (offset < 0)
3371 return -ENXIO;
3372
3373 inode_lock(inode);
3374 /* We're holding i_rwsem so we can access i_size directly */
3375 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
3376 if (offset >= 0)
3377 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
3378 inode_unlock(inode);
3379 return offset;
3380 }
3381
shmem_fallocate(struct file * file,int mode,loff_t offset,loff_t len)3382 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3383 loff_t len)
3384 {
3385 struct inode *inode = file_inode(file);
3386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3387 struct shmem_inode_info *info = SHMEM_I(inode);
3388 struct shmem_falloc shmem_falloc;
3389 pgoff_t start, index, end, undo_fallocend;
3390 int error;
3391
3392 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3393 return -EOPNOTSUPP;
3394
3395 inode_lock(inode);
3396
3397 if (mode & FALLOC_FL_PUNCH_HOLE) {
3398 struct address_space *mapping = file->f_mapping;
3399 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3400 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
3401 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
3402
3403 /* protected by i_rwsem */
3404 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
3405 error = -EPERM;
3406 goto out;
3407 }
3408
3409 shmem_falloc.waitq = &shmem_falloc_waitq;
3410 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
3411 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3412 spin_lock(&inode->i_lock);
3413 inode->i_private = &shmem_falloc;
3414 spin_unlock(&inode->i_lock);
3415
3416 if ((u64)unmap_end > (u64)unmap_start)
3417 unmap_mapping_range(mapping, unmap_start,
3418 1 + unmap_end - unmap_start, 0);
3419 shmem_truncate_range(inode, offset, offset + len - 1);
3420 /* No need to unmap again: hole-punching leaves COWed pages */
3421
3422 spin_lock(&inode->i_lock);
3423 inode->i_private = NULL;
3424 wake_up_all(&shmem_falloc_waitq);
3425 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
3426 spin_unlock(&inode->i_lock);
3427 error = 0;
3428 goto out;
3429 }
3430
3431 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3432 error = inode_newsize_ok(inode, offset + len);
3433 if (error)
3434 goto out;
3435
3436 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3437 error = -EPERM;
3438 goto out;
3439 }
3440
3441 start = offset >> PAGE_SHIFT;
3442 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
3443 /* Try to avoid a swapstorm if len is impossible to satisfy */
3444 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3445 error = -ENOSPC;
3446 goto out;
3447 }
3448
3449 shmem_falloc.waitq = NULL;
3450 shmem_falloc.start = start;
3451 shmem_falloc.next = start;
3452 shmem_falloc.nr_falloced = 0;
3453 shmem_falloc.nr_unswapped = 0;
3454 spin_lock(&inode->i_lock);
3455 inode->i_private = &shmem_falloc;
3456 spin_unlock(&inode->i_lock);
3457
3458 /*
3459 * info->fallocend is only relevant when huge pages might be
3460 * involved: to prevent split_huge_page() freeing fallocated
3461 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3462 */
3463 undo_fallocend = info->fallocend;
3464 if (info->fallocend < end)
3465 info->fallocend = end;
3466
3467 for (index = start; index < end; ) {
3468 struct folio *folio;
3469
3470 /*
3471 * Check for fatal signal so that we abort early in OOM
3472 * situations. We don't want to abort in case of non-fatal
3473 * signals as large fallocate can take noticeable time and
3474 * e.g. periodic timers may result in fallocate constantly
3475 * restarting.
3476 */
3477 if (fatal_signal_pending(current))
3478 error = -EINTR;
3479 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3480 error = -ENOMEM;
3481 else
3482 error = shmem_get_folio(inode, index, offset + len,
3483 &folio, SGP_FALLOC);
3484 if (error) {
3485 info->fallocend = undo_fallocend;
3486 /* Remove the !uptodate folios we added */
3487 if (index > start) {
3488 shmem_undo_range(inode,
3489 (loff_t)start << PAGE_SHIFT,
3490 ((loff_t)index << PAGE_SHIFT) - 1, true);
3491 }
3492 goto undone;
3493 }
3494
3495 /*
3496 * Here is a more important optimization than it appears:
3497 * a second SGP_FALLOC on the same large folio will clear it,
3498 * making it uptodate and un-undoable if we fail later.
3499 */
3500 index = folio_next_index(folio);
3501 /* Beware 32-bit wraparound */
3502 if (!index)
3503 index--;
3504
3505 /*
3506 * Inform shmem_writepage() how far we have reached.
3507 * No need for lock or barrier: we have the page lock.
3508 */
3509 if (!folio_test_uptodate(folio))
3510 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3511 shmem_falloc.next = index;
3512
3513 /*
3514 * If !uptodate, leave it that way so that freeable folios
3515 * can be recognized if we need to rollback on error later.
3516 * But mark it dirty so that memory pressure will swap rather
3517 * than free the folios we are allocating (and SGP_CACHE folios
3518 * might still be clean: we now need to mark those dirty too).
3519 */
3520 folio_mark_dirty(folio);
3521 folio_unlock(folio);
3522 folio_put(folio);
3523 cond_resched();
3524 }
3525
3526 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3527 i_size_write(inode, offset + len);
3528 undone:
3529 spin_lock(&inode->i_lock);
3530 inode->i_private = NULL;
3531 spin_unlock(&inode->i_lock);
3532 out:
3533 if (!error)
3534 file_modified(file);
3535 inode_unlock(inode);
3536 return error;
3537 }
3538
shmem_statfs(struct dentry * dentry,struct kstatfs * buf)3539 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3540 {
3541 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3542
3543 buf->f_type = TMPFS_MAGIC;
3544 buf->f_bsize = PAGE_SIZE;
3545 buf->f_namelen = NAME_MAX;
3546 if (sbinfo->max_blocks) {
3547 buf->f_blocks = sbinfo->max_blocks;
3548 buf->f_bavail =
3549 buf->f_bfree = sbinfo->max_blocks -
3550 percpu_counter_sum(&sbinfo->used_blocks);
3551 }
3552 if (sbinfo->max_inodes) {
3553 buf->f_files = sbinfo->max_inodes;
3554 buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
3555 }
3556 /* else leave those fields 0 like simple_statfs */
3557
3558 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3559
3560 return 0;
3561 }
3562
3563 /*
3564 * File creation. Allocate an inode, and we're done..
3565 */
3566 static int
shmem_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t dev)3567 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3568 struct dentry *dentry, umode_t mode, dev_t dev)
3569 {
3570 struct inode *inode;
3571 int error;
3572
3573 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3574 if (IS_ERR(inode))
3575 return PTR_ERR(inode);
3576
3577 error = simple_acl_create(dir, inode);
3578 if (error)
3579 goto out_iput;
3580 error = security_inode_init_security(inode, dir, &dentry->d_name,
3581 shmem_initxattrs, NULL);
3582 if (error && error != -EOPNOTSUPP)
3583 goto out_iput;
3584
3585 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3586 if (error)
3587 goto out_iput;
3588
3589 dir->i_size += BOGO_DIRENT_SIZE;
3590 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3591 inode_inc_iversion(dir);
3592 d_instantiate(dentry, inode);
3593 dget(dentry); /* Extra count - pin the dentry in core */
3594 return error;
3595
3596 out_iput:
3597 iput(inode);
3598 return error;
3599 }
3600
3601 static int
shmem_tmpfile(struct mnt_idmap * idmap,struct inode * dir,struct file * file,umode_t mode)3602 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3603 struct file *file, umode_t mode)
3604 {
3605 struct inode *inode;
3606 int error;
3607
3608 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3609 if (IS_ERR(inode)) {
3610 error = PTR_ERR(inode);
3611 goto err_out;
3612 }
3613 error = security_inode_init_security(inode, dir, NULL,
3614 shmem_initxattrs, NULL);
3615 if (error && error != -EOPNOTSUPP)
3616 goto out_iput;
3617 error = simple_acl_create(dir, inode);
3618 if (error)
3619 goto out_iput;
3620 d_tmpfile(file, inode);
3621
3622 err_out:
3623 return finish_open_simple(file, error);
3624 out_iput:
3625 iput(inode);
3626 return error;
3627 }
3628
shmem_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)3629 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3630 struct dentry *dentry, umode_t mode)
3631 {
3632 int error;
3633
3634 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3635 if (error)
3636 return error;
3637 inc_nlink(dir);
3638 return 0;
3639 }
3640
shmem_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)3641 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3642 struct dentry *dentry, umode_t mode, bool excl)
3643 {
3644 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3645 }
3646
3647 /*
3648 * Link a file..
3649 */
shmem_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)3650 static int shmem_link(struct dentry *old_dentry, struct inode *dir,
3651 struct dentry *dentry)
3652 {
3653 struct inode *inode = d_inode(old_dentry);
3654 int ret = 0;
3655
3656 /*
3657 * No ordinary (disk based) filesystem counts links as inodes;
3658 * but each new link needs a new dentry, pinning lowmem, and
3659 * tmpfs dentries cannot be pruned until they are unlinked.
3660 * But if an O_TMPFILE file is linked into the tmpfs, the
3661 * first link must skip that, to get the accounting right.
3662 */
3663 if (inode->i_nlink) {
3664 ret = shmem_reserve_inode(inode->i_sb, NULL);
3665 if (ret)
3666 goto out;
3667 }
3668
3669 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3670 if (ret) {
3671 if (inode->i_nlink)
3672 shmem_free_inode(inode->i_sb, 0);
3673 goto out;
3674 }
3675
3676 dir->i_size += BOGO_DIRENT_SIZE;
3677 inode_set_mtime_to_ts(dir,
3678 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
3679 inode_inc_iversion(dir);
3680 inc_nlink(inode);
3681 ihold(inode); /* New dentry reference */
3682 dget(dentry); /* Extra pinning count for the created dentry */
3683 d_instantiate(dentry, inode);
3684 out:
3685 return ret;
3686 }
3687
shmem_unlink(struct inode * dir,struct dentry * dentry)3688 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3689 {
3690 struct inode *inode = d_inode(dentry);
3691
3692 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3693 shmem_free_inode(inode->i_sb, 0);
3694
3695 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3696
3697 dir->i_size -= BOGO_DIRENT_SIZE;
3698 inode_set_mtime_to_ts(dir,
3699 inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
3700 inode_inc_iversion(dir);
3701 drop_nlink(inode);
3702 dput(dentry); /* Undo the count from "create" - does all the work */
3703 return 0;
3704 }
3705
shmem_rmdir(struct inode * dir,struct dentry * dentry)3706 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3707 {
3708 if (!simple_offset_empty(dentry))
3709 return -ENOTEMPTY;
3710
3711 drop_nlink(d_inode(dentry));
3712 drop_nlink(dir);
3713 return shmem_unlink(dir, dentry);
3714 }
3715
shmem_whiteout(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry)3716 static int shmem_whiteout(struct mnt_idmap *idmap,
3717 struct inode *old_dir, struct dentry *old_dentry)
3718 {
3719 struct dentry *whiteout;
3720 int error;
3721
3722 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3723 if (!whiteout)
3724 return -ENOMEM;
3725
3726 error = shmem_mknod(idmap, old_dir, whiteout,
3727 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3728 dput(whiteout);
3729 if (error)
3730 return error;
3731
3732 /*
3733 * Cheat and hash the whiteout while the old dentry is still in
3734 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3735 *
3736 * d_lookup() will consistently find one of them at this point,
3737 * not sure which one, but that isn't even important.
3738 */
3739 d_rehash(whiteout);
3740 return 0;
3741 }
3742
3743 /*
3744 * The VFS layer already does all the dentry stuff for rename,
3745 * we just have to decrement the usage count for the target if
3746 * it exists so that the VFS layer correctly free's it when it
3747 * gets overwritten.
3748 */
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)3749 static int shmem_rename2(struct mnt_idmap *idmap,
3750 struct inode *old_dir, struct dentry *old_dentry,
3751 struct inode *new_dir, struct dentry *new_dentry,
3752 unsigned int flags)
3753 {
3754 struct inode *inode = d_inode(old_dentry);
3755 int they_are_dirs = S_ISDIR(inode->i_mode);
3756 int error;
3757
3758 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3759 return -EINVAL;
3760
3761 if (flags & RENAME_EXCHANGE)
3762 return simple_offset_rename_exchange(old_dir, old_dentry,
3763 new_dir, new_dentry);
3764
3765 if (!simple_offset_empty(new_dentry))
3766 return -ENOTEMPTY;
3767
3768 if (flags & RENAME_WHITEOUT) {
3769 error = shmem_whiteout(idmap, old_dir, old_dentry);
3770 if (error)
3771 return error;
3772 }
3773
3774 error = simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry);
3775 if (error)
3776 return error;
3777
3778 if (d_really_is_positive(new_dentry)) {
3779 (void) shmem_unlink(new_dir, new_dentry);
3780 if (they_are_dirs) {
3781 drop_nlink(d_inode(new_dentry));
3782 drop_nlink(old_dir);
3783 }
3784 } else if (they_are_dirs) {
3785 drop_nlink(old_dir);
3786 inc_nlink(new_dir);
3787 }
3788
3789 old_dir->i_size -= BOGO_DIRENT_SIZE;
3790 new_dir->i_size += BOGO_DIRENT_SIZE;
3791 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
3792 inode_inc_iversion(old_dir);
3793 inode_inc_iversion(new_dir);
3794 return 0;
3795 }
3796
shmem_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)3797 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3798 struct dentry *dentry, const char *symname)
3799 {
3800 int error;
3801 int len;
3802 struct inode *inode;
3803 struct folio *folio;
3804
3805 len = strlen(symname) + 1;
3806 if (len > PAGE_SIZE)
3807 return -ENAMETOOLONG;
3808
3809 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3810 VM_NORESERVE);
3811 if (IS_ERR(inode))
3812 return PTR_ERR(inode);
3813
3814 error = security_inode_init_security(inode, dir, &dentry->d_name,
3815 shmem_initxattrs, NULL);
3816 if (error && error != -EOPNOTSUPP)
3817 goto out_iput;
3818
3819 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3820 if (error)
3821 goto out_iput;
3822
3823 inode->i_size = len-1;
3824 if (len <= SHORT_SYMLINK_LEN) {
3825 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3826 if (!inode->i_link) {
3827 error = -ENOMEM;
3828 goto out_remove_offset;
3829 }
3830 inode->i_op = &shmem_short_symlink_operations;
3831 } else {
3832 inode_nohighmem(inode);
3833 inode->i_mapping->a_ops = &shmem_aops;
3834 error = shmem_get_folio(inode, 0, 0, &folio, SGP_WRITE);
3835 if (error)
3836 goto out_remove_offset;
3837 inode->i_op = &shmem_symlink_inode_operations;
3838 memcpy(folio_address(folio), symname, len);
3839 folio_mark_uptodate(folio);
3840 folio_mark_dirty(folio);
3841 folio_unlock(folio);
3842 folio_put(folio);
3843 }
3844 dir->i_size += BOGO_DIRENT_SIZE;
3845 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3846 inode_inc_iversion(dir);
3847 d_instantiate(dentry, inode);
3848 dget(dentry);
3849 return 0;
3850
3851 out_remove_offset:
3852 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3853 out_iput:
3854 iput(inode);
3855 return error;
3856 }
3857
shmem_put_link(void * arg)3858 static void shmem_put_link(void *arg)
3859 {
3860 folio_mark_accessed(arg);
3861 folio_put(arg);
3862 }
3863
shmem_get_link(struct dentry * dentry,struct inode * inode,struct delayed_call * done)3864 static const char *shmem_get_link(struct dentry *dentry, struct inode *inode,
3865 struct delayed_call *done)
3866 {
3867 struct folio *folio = NULL;
3868 int error;
3869
3870 if (!dentry) {
3871 folio = filemap_get_folio(inode->i_mapping, 0);
3872 if (IS_ERR(folio))
3873 return ERR_PTR(-ECHILD);
3874 if (PageHWPoison(folio_page(folio, 0)) ||
3875 !folio_test_uptodate(folio)) {
3876 folio_put(folio);
3877 return ERR_PTR(-ECHILD);
3878 }
3879 } else {
3880 error = shmem_get_folio(inode, 0, 0, &folio, SGP_READ);
3881 if (error)
3882 return ERR_PTR(error);
3883 if (!folio)
3884 return ERR_PTR(-ECHILD);
3885 if (PageHWPoison(folio_page(folio, 0))) {
3886 folio_unlock(folio);
3887 folio_put(folio);
3888 return ERR_PTR(-ECHILD);
3889 }
3890 folio_unlock(folio);
3891 }
3892 set_delayed_call(done, shmem_put_link, folio);
3893 return folio_address(folio);
3894 }
3895
3896 #ifdef CONFIG_TMPFS_XATTR
3897
shmem_fileattr_get(struct dentry * dentry,struct fileattr * fa)3898 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3899 {
3900 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3901
3902 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3903
3904 return 0;
3905 }
3906
shmem_fileattr_set(struct mnt_idmap * idmap,struct dentry * dentry,struct fileattr * fa)3907 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3908 struct dentry *dentry, struct fileattr *fa)
3909 {
3910 struct inode *inode = d_inode(dentry);
3911 struct shmem_inode_info *info = SHMEM_I(inode);
3912
3913 if (fileattr_has_fsx(fa))
3914 return -EOPNOTSUPP;
3915 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3916 return -EOPNOTSUPP;
3917
3918 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3919 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3920
3921 shmem_set_inode_flags(inode, info->fsflags);
3922 inode_set_ctime_current(inode);
3923 inode_inc_iversion(inode);
3924 return 0;
3925 }
3926
3927 /*
3928 * Superblocks without xattr inode operations may get some security.* xattr
3929 * support from the LSM "for free". As soon as we have any other xattrs
3930 * like ACLs, we also need to implement the security.* handlers at
3931 * filesystem level, though.
3932 */
3933
3934 /*
3935 * Callback for security_inode_init_security() for acquiring xattrs.
3936 */
shmem_initxattrs(struct inode * inode,const struct xattr * xattr_array,void * fs_info)3937 static int shmem_initxattrs(struct inode *inode,
3938 const struct xattr *xattr_array, void *fs_info)
3939 {
3940 struct shmem_inode_info *info = SHMEM_I(inode);
3941 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3942 const struct xattr *xattr;
3943 struct simple_xattr *new_xattr;
3944 size_t ispace = 0;
3945 size_t len;
3946
3947 if (sbinfo->max_inodes) {
3948 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3949 ispace += simple_xattr_space(xattr->name,
3950 xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
3951 }
3952 if (ispace) {
3953 raw_spin_lock(&sbinfo->stat_lock);
3954 if (sbinfo->free_ispace < ispace)
3955 ispace = 0;
3956 else
3957 sbinfo->free_ispace -= ispace;
3958 raw_spin_unlock(&sbinfo->stat_lock);
3959 if (!ispace)
3960 return -ENOSPC;
3961 }
3962 }
3963
3964 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3965 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3966 if (!new_xattr)
3967 break;
3968
3969 len = strlen(xattr->name) + 1;
3970 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3971 GFP_KERNEL_ACCOUNT);
3972 if (!new_xattr->name) {
3973 kvfree(new_xattr);
3974 break;
3975 }
3976
3977 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3978 XATTR_SECURITY_PREFIX_LEN);
3979 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3980 xattr->name, len);
3981
3982 simple_xattr_add(&info->xattrs, new_xattr);
3983 }
3984
3985 if (xattr->name != NULL) {
3986 if (ispace) {
3987 raw_spin_lock(&sbinfo->stat_lock);
3988 sbinfo->free_ispace += ispace;
3989 raw_spin_unlock(&sbinfo->stat_lock);
3990 }
3991 simple_xattrs_free(&info->xattrs, NULL);
3992 return -ENOMEM;
3993 }
3994
3995 return 0;
3996 }
3997
shmem_xattr_handler_get(const struct xattr_handler * handler,struct dentry * unused,struct inode * inode,const char * name,void * buffer,size_t size)3998 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3999 struct dentry *unused, struct inode *inode,
4000 const char *name, void *buffer, size_t size)
4001 {
4002 struct shmem_inode_info *info = SHMEM_I(inode);
4003
4004 name = xattr_full_name(handler, name);
4005 return simple_xattr_get(&info->xattrs, name, buffer, size);
4006 }
4007
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)4008 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
4009 struct mnt_idmap *idmap,
4010 struct dentry *unused, struct inode *inode,
4011 const char *name, const void *value,
4012 size_t size, int flags)
4013 {
4014 struct shmem_inode_info *info = SHMEM_I(inode);
4015 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4016 struct simple_xattr *old_xattr;
4017 size_t ispace = 0;
4018
4019 name = xattr_full_name(handler, name);
4020 if (value && sbinfo->max_inodes) {
4021 ispace = simple_xattr_space(name, size);
4022 raw_spin_lock(&sbinfo->stat_lock);
4023 if (sbinfo->free_ispace < ispace)
4024 ispace = 0;
4025 else
4026 sbinfo->free_ispace -= ispace;
4027 raw_spin_unlock(&sbinfo->stat_lock);
4028 if (!ispace)
4029 return -ENOSPC;
4030 }
4031
4032 old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
4033 if (!IS_ERR(old_xattr)) {
4034 ispace = 0;
4035 if (old_xattr && sbinfo->max_inodes)
4036 ispace = simple_xattr_space(old_xattr->name,
4037 old_xattr->size);
4038 simple_xattr_free(old_xattr);
4039 old_xattr = NULL;
4040 inode_set_ctime_current(inode);
4041 inode_inc_iversion(inode);
4042 }
4043 if (ispace) {
4044 raw_spin_lock(&sbinfo->stat_lock);
4045 sbinfo->free_ispace += ispace;
4046 raw_spin_unlock(&sbinfo->stat_lock);
4047 }
4048 return PTR_ERR(old_xattr);
4049 }
4050
4051 static const struct xattr_handler shmem_security_xattr_handler = {
4052 .prefix = XATTR_SECURITY_PREFIX,
4053 .get = shmem_xattr_handler_get,
4054 .set = shmem_xattr_handler_set,
4055 };
4056
4057 static const struct xattr_handler shmem_trusted_xattr_handler = {
4058 .prefix = XATTR_TRUSTED_PREFIX,
4059 .get = shmem_xattr_handler_get,
4060 .set = shmem_xattr_handler_set,
4061 };
4062
4063 static const struct xattr_handler shmem_user_xattr_handler = {
4064 .prefix = XATTR_USER_PREFIX,
4065 .get = shmem_xattr_handler_get,
4066 .set = shmem_xattr_handler_set,
4067 };
4068
4069 static const struct xattr_handler * const shmem_xattr_handlers[] = {
4070 &shmem_security_xattr_handler,
4071 &shmem_trusted_xattr_handler,
4072 &shmem_user_xattr_handler,
4073 NULL
4074 };
4075
shmem_listxattr(struct dentry * dentry,char * buffer,size_t size)4076 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
4077 {
4078 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
4079 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
4080 }
4081 #endif /* CONFIG_TMPFS_XATTR */
4082
4083 static const struct inode_operations shmem_short_symlink_operations = {
4084 .getattr = shmem_getattr,
4085 .setattr = shmem_setattr,
4086 .get_link = simple_get_link,
4087 #ifdef CONFIG_TMPFS_XATTR
4088 .listxattr = shmem_listxattr,
4089 #endif
4090 };
4091
4092 static const struct inode_operations shmem_symlink_inode_operations = {
4093 .getattr = shmem_getattr,
4094 .setattr = shmem_setattr,
4095 .get_link = shmem_get_link,
4096 #ifdef CONFIG_TMPFS_XATTR
4097 .listxattr = shmem_listxattr,
4098 #endif
4099 };
4100
shmem_get_parent(struct dentry * child)4101 static struct dentry *shmem_get_parent(struct dentry *child)
4102 {
4103 return ERR_PTR(-ESTALE);
4104 }
4105
shmem_match(struct inode * ino,void * vfh)4106 static int shmem_match(struct inode *ino, void *vfh)
4107 {
4108 __u32 *fh = vfh;
4109 __u64 inum = fh[2];
4110 inum = (inum << 32) | fh[1];
4111 return ino->i_ino == inum && fh[0] == ino->i_generation;
4112 }
4113
4114 /* Find any alias of inode, but prefer a hashed alias */
shmem_find_alias(struct inode * inode)4115 static struct dentry *shmem_find_alias(struct inode *inode)
4116 {
4117 struct dentry *alias = d_find_alias(inode);
4118
4119 return alias ?: d_find_any_alias(inode);
4120 }
4121
shmem_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)4122 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
4123 struct fid *fid, int fh_len, int fh_type)
4124 {
4125 struct inode *inode;
4126 struct dentry *dentry = NULL;
4127 u64 inum;
4128
4129 if (fh_len < 3)
4130 return NULL;
4131
4132 inum = fid->raw[2];
4133 inum = (inum << 32) | fid->raw[1];
4134
4135 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
4136 shmem_match, fid->raw);
4137 if (inode) {
4138 dentry = shmem_find_alias(inode);
4139 iput(inode);
4140 }
4141
4142 return dentry;
4143 }
4144
shmem_encode_fh(struct inode * inode,__u32 * fh,int * len,struct inode * parent)4145 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
4146 struct inode *parent)
4147 {
4148 if (*len < 3) {
4149 *len = 3;
4150 return FILEID_INVALID;
4151 }
4152
4153 if (inode_unhashed(inode)) {
4154 /* Unfortunately insert_inode_hash is not idempotent,
4155 * so as we hash inodes here rather than at creation
4156 * time, we need a lock to ensure we only try
4157 * to do it once
4158 */
4159 static DEFINE_SPINLOCK(lock);
4160 spin_lock(&lock);
4161 if (inode_unhashed(inode))
4162 __insert_inode_hash(inode,
4163 inode->i_ino + inode->i_generation);
4164 spin_unlock(&lock);
4165 }
4166
4167 fh[0] = inode->i_generation;
4168 fh[1] = inode->i_ino;
4169 fh[2] = ((__u64)inode->i_ino) >> 32;
4170
4171 *len = 3;
4172 return 1;
4173 }
4174
4175 static const struct export_operations shmem_export_ops = {
4176 .get_parent = shmem_get_parent,
4177 .encode_fh = shmem_encode_fh,
4178 .fh_to_dentry = shmem_fh_to_dentry,
4179 };
4180
4181 enum shmem_param {
4182 Opt_gid,
4183 Opt_huge,
4184 Opt_mode,
4185 Opt_mpol,
4186 Opt_nr_blocks,
4187 Opt_nr_inodes,
4188 Opt_size,
4189 Opt_uid,
4190 Opt_inode32,
4191 Opt_inode64,
4192 Opt_noswap,
4193 Opt_quota,
4194 Opt_usrquota,
4195 Opt_grpquota,
4196 Opt_usrquota_block_hardlimit,
4197 Opt_usrquota_inode_hardlimit,
4198 Opt_grpquota_block_hardlimit,
4199 Opt_grpquota_inode_hardlimit,
4200 };
4201
4202 static const struct constant_table shmem_param_enums_huge[] = {
4203 {"never", SHMEM_HUGE_NEVER },
4204 {"always", SHMEM_HUGE_ALWAYS },
4205 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
4206 {"advise", SHMEM_HUGE_ADVISE },
4207 {}
4208 };
4209
4210 const struct fs_parameter_spec shmem_fs_parameters[] = {
4211 fsparam_gid ("gid", Opt_gid),
4212 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
4213 fsparam_u32oct("mode", Opt_mode),
4214 fsparam_string("mpol", Opt_mpol),
4215 fsparam_string("nr_blocks", Opt_nr_blocks),
4216 fsparam_string("nr_inodes", Opt_nr_inodes),
4217 fsparam_string("size", Opt_size),
4218 fsparam_uid ("uid", Opt_uid),
4219 fsparam_flag ("inode32", Opt_inode32),
4220 fsparam_flag ("inode64", Opt_inode64),
4221 fsparam_flag ("noswap", Opt_noswap),
4222 #ifdef CONFIG_TMPFS_QUOTA
4223 fsparam_flag ("quota", Opt_quota),
4224 fsparam_flag ("usrquota", Opt_usrquota),
4225 fsparam_flag ("grpquota", Opt_grpquota),
4226 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
4227 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
4228 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
4229 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
4230 #endif
4231 {}
4232 };
4233
shmem_parse_one(struct fs_context * fc,struct fs_parameter * param)4234 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
4235 {
4236 struct shmem_options *ctx = fc->fs_private;
4237 struct fs_parse_result result;
4238 unsigned long long size;
4239 char *rest;
4240 int opt;
4241 kuid_t kuid;
4242 kgid_t kgid;
4243
4244 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
4245 if (opt < 0)
4246 return opt;
4247
4248 switch (opt) {
4249 case Opt_size:
4250 size = memparse(param->string, &rest);
4251 if (*rest == '%') {
4252 size <<= PAGE_SHIFT;
4253 size *= totalram_pages();
4254 do_div(size, 100);
4255 rest++;
4256 }
4257 if (*rest)
4258 goto bad_value;
4259 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
4260 ctx->seen |= SHMEM_SEEN_BLOCKS;
4261 break;
4262 case Opt_nr_blocks:
4263 ctx->blocks = memparse(param->string, &rest);
4264 if (*rest || ctx->blocks > LONG_MAX)
4265 goto bad_value;
4266 ctx->seen |= SHMEM_SEEN_BLOCKS;
4267 break;
4268 case Opt_nr_inodes:
4269 ctx->inodes = memparse(param->string, &rest);
4270 if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
4271 goto bad_value;
4272 ctx->seen |= SHMEM_SEEN_INODES;
4273 break;
4274 case Opt_mode:
4275 ctx->mode = result.uint_32 & 07777;
4276 break;
4277 case Opt_uid:
4278 kuid = result.uid;
4279
4280 /*
4281 * The requested uid must be representable in the
4282 * filesystem's idmapping.
4283 */
4284 if (!kuid_has_mapping(fc->user_ns, kuid))
4285 goto bad_value;
4286
4287 ctx->uid = kuid;
4288 break;
4289 case Opt_gid:
4290 kgid = result.gid;
4291
4292 /*
4293 * The requested gid must be representable in the
4294 * filesystem's idmapping.
4295 */
4296 if (!kgid_has_mapping(fc->user_ns, kgid))
4297 goto bad_value;
4298
4299 ctx->gid = kgid;
4300 break;
4301 case Opt_huge:
4302 ctx->huge = result.uint_32;
4303 if (ctx->huge != SHMEM_HUGE_NEVER &&
4304 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4305 has_transparent_hugepage()))
4306 goto unsupported_parameter;
4307 ctx->seen |= SHMEM_SEEN_HUGE;
4308 break;
4309 case Opt_mpol:
4310 if (IS_ENABLED(CONFIG_NUMA)) {
4311 mpol_put(ctx->mpol);
4312 ctx->mpol = NULL;
4313 if (mpol_parse_str(param->string, &ctx->mpol))
4314 goto bad_value;
4315 break;
4316 }
4317 goto unsupported_parameter;
4318 case Opt_inode32:
4319 ctx->full_inums = false;
4320 ctx->seen |= SHMEM_SEEN_INUMS;
4321 break;
4322 case Opt_inode64:
4323 if (sizeof(ino_t) < 8) {
4324 return invalfc(fc,
4325 "Cannot use inode64 with <64bit inums in kernel\n");
4326 }
4327 ctx->full_inums = true;
4328 ctx->seen |= SHMEM_SEEN_INUMS;
4329 break;
4330 case Opt_noswap:
4331 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
4332 return invalfc(fc,
4333 "Turning off swap in unprivileged tmpfs mounts unsupported");
4334 }
4335 ctx->noswap = true;
4336 ctx->seen |= SHMEM_SEEN_NOSWAP;
4337 break;
4338 case Opt_quota:
4339 if (fc->user_ns != &init_user_ns)
4340 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4341 ctx->seen |= SHMEM_SEEN_QUOTA;
4342 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
4343 break;
4344 case Opt_usrquota:
4345 if (fc->user_ns != &init_user_ns)
4346 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4347 ctx->seen |= SHMEM_SEEN_QUOTA;
4348 ctx->quota_types |= QTYPE_MASK_USR;
4349 break;
4350 case Opt_grpquota:
4351 if (fc->user_ns != &init_user_ns)
4352 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4353 ctx->seen |= SHMEM_SEEN_QUOTA;
4354 ctx->quota_types |= QTYPE_MASK_GRP;
4355 break;
4356 case Opt_usrquota_block_hardlimit:
4357 size = memparse(param->string, &rest);
4358 if (*rest || !size)
4359 goto bad_value;
4360 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4361 return invalfc(fc,
4362 "User quota block hardlimit too large.");
4363 ctx->qlimits.usrquota_bhardlimit = size;
4364 break;
4365 case Opt_grpquota_block_hardlimit:
4366 size = memparse(param->string, &rest);
4367 if (*rest || !size)
4368 goto bad_value;
4369 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4370 return invalfc(fc,
4371 "Group quota block hardlimit too large.");
4372 ctx->qlimits.grpquota_bhardlimit = size;
4373 break;
4374 case Opt_usrquota_inode_hardlimit:
4375 size = memparse(param->string, &rest);
4376 if (*rest || !size)
4377 goto bad_value;
4378 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4379 return invalfc(fc,
4380 "User quota inode hardlimit too large.");
4381 ctx->qlimits.usrquota_ihardlimit = size;
4382 break;
4383 case Opt_grpquota_inode_hardlimit:
4384 size = memparse(param->string, &rest);
4385 if (*rest || !size)
4386 goto bad_value;
4387 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4388 return invalfc(fc,
4389 "Group quota inode hardlimit too large.");
4390 ctx->qlimits.grpquota_ihardlimit = size;
4391 break;
4392 }
4393 return 0;
4394
4395 unsupported_parameter:
4396 return invalfc(fc, "Unsupported parameter '%s'", param->key);
4397 bad_value:
4398 return invalfc(fc, "Bad value for '%s'", param->key);
4399 }
4400
shmem_parse_options(struct fs_context * fc,void * data)4401 static int shmem_parse_options(struct fs_context *fc, void *data)
4402 {
4403 char *options = data;
4404
4405 if (options) {
4406 int err = security_sb_eat_lsm_opts(options, &fc->security);
4407 if (err)
4408 return err;
4409 }
4410
4411 while (options != NULL) {
4412 char *this_char = options;
4413 for (;;) {
4414 /*
4415 * NUL-terminate this option: unfortunately,
4416 * mount options form a comma-separated list,
4417 * but mpol's nodelist may also contain commas.
4418 */
4419 options = strchr(options, ',');
4420 if (options == NULL)
4421 break;
4422 options++;
4423 if (!isdigit(*options)) {
4424 options[-1] = '\0';
4425 break;
4426 }
4427 }
4428 if (*this_char) {
4429 char *value = strchr(this_char, '=');
4430 size_t len = 0;
4431 int err;
4432
4433 if (value) {
4434 *value++ = '\0';
4435 len = strlen(value);
4436 }
4437 err = vfs_parse_fs_string(fc, this_char, value, len);
4438 if (err < 0)
4439 return err;
4440 }
4441 }
4442 return 0;
4443 }
4444
4445 /*
4446 * Reconfigure a shmem filesystem.
4447 */
shmem_reconfigure(struct fs_context * fc)4448 static int shmem_reconfigure(struct fs_context *fc)
4449 {
4450 struct shmem_options *ctx = fc->fs_private;
4451 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
4452 unsigned long used_isp;
4453 struct mempolicy *mpol = NULL;
4454 const char *err;
4455
4456 raw_spin_lock(&sbinfo->stat_lock);
4457 used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
4458
4459 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4460 if (!sbinfo->max_blocks) {
4461 err = "Cannot retroactively limit size";
4462 goto out;
4463 }
4464 if (percpu_counter_compare(&sbinfo->used_blocks,
4465 ctx->blocks) > 0) {
4466 err = "Too small a size for current use";
4467 goto out;
4468 }
4469 }
4470 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4471 if (!sbinfo->max_inodes) {
4472 err = "Cannot retroactively limit inodes";
4473 goto out;
4474 }
4475 if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
4476 err = "Too few inodes for current use";
4477 goto out;
4478 }
4479 }
4480
4481 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4482 sbinfo->next_ino > UINT_MAX) {
4483 err = "Current inum too high to switch to 32-bit inums";
4484 goto out;
4485 }
4486 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4487 err = "Cannot disable swap on remount";
4488 goto out;
4489 }
4490 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4491 err = "Cannot enable swap on remount if it was disabled on first mount";
4492 goto out;
4493 }
4494
4495 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4496 !sb_any_quota_loaded(fc->root->d_sb)) {
4497 err = "Cannot enable quota on remount";
4498 goto out;
4499 }
4500
4501 #ifdef CONFIG_TMPFS_QUOTA
4502 #define CHANGED_LIMIT(name) \
4503 (ctx->qlimits.name## hardlimit && \
4504 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4505
4506 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4507 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4508 err = "Cannot change global quota limit on remount";
4509 goto out;
4510 }
4511 #endif /* CONFIG_TMPFS_QUOTA */
4512
4513 if (ctx->seen & SHMEM_SEEN_HUGE)
4514 sbinfo->huge = ctx->huge;
4515 if (ctx->seen & SHMEM_SEEN_INUMS)
4516 sbinfo->full_inums = ctx->full_inums;
4517 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4518 sbinfo->max_blocks = ctx->blocks;
4519 if (ctx->seen & SHMEM_SEEN_INODES) {
4520 sbinfo->max_inodes = ctx->inodes;
4521 sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
4522 }
4523
4524 /*
4525 * Preserve previous mempolicy unless mpol remount option was specified.
4526 */
4527 if (ctx->mpol) {
4528 mpol = sbinfo->mpol;
4529 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4530 ctx->mpol = NULL;
4531 }
4532
4533 if (ctx->noswap)
4534 sbinfo->noswap = true;
4535
4536 raw_spin_unlock(&sbinfo->stat_lock);
4537 mpol_put(mpol);
4538 return 0;
4539 out:
4540 raw_spin_unlock(&sbinfo->stat_lock);
4541 return invalfc(fc, "%s", err);
4542 }
4543
shmem_show_options(struct seq_file * seq,struct dentry * root)4544 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
4545 {
4546 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
4547 struct mempolicy *mpol;
4548
4549 if (sbinfo->max_blocks != shmem_default_max_blocks())
4550 seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks));
4551 if (sbinfo->max_inodes != shmem_default_max_inodes())
4552 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
4553 if (sbinfo->mode != (0777 | S_ISVTX))
4554 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
4555 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4556 seq_printf(seq, ",uid=%u",
4557 from_kuid_munged(&init_user_ns, sbinfo->uid));
4558 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4559 seq_printf(seq, ",gid=%u",
4560 from_kgid_munged(&init_user_ns, sbinfo->gid));
4561
4562 /*
4563 * Showing inode{64,32} might be useful even if it's the system default,
4564 * since then people don't have to resort to checking both here and
4565 * /proc/config.gz to confirm 64-bit inums were successfully applied
4566 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4567 *
4568 * We hide it when inode64 isn't the default and we are using 32-bit
4569 * inodes, since that probably just means the feature isn't even under
4570 * consideration.
4571 *
4572 * As such:
4573 *
4574 * +-----------------+-----------------+
4575 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4576 * +------------------+-----------------+-----------------+
4577 * | full_inums=true | show | show |
4578 * | full_inums=false | show | hide |
4579 * +------------------+-----------------+-----------------+
4580 *
4581 */
4582 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4583 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
4584 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4585 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4586 if (sbinfo->huge)
4587 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4588 #endif
4589 mpol = shmem_get_sbmpol(sbinfo);
4590 shmem_show_mpol(seq, mpol);
4591 mpol_put(mpol);
4592 if (sbinfo->noswap)
4593 seq_printf(seq, ",noswap");
4594 #ifdef CONFIG_TMPFS_QUOTA
4595 if (sb_has_quota_active(root->d_sb, USRQUOTA))
4596 seq_printf(seq, ",usrquota");
4597 if (sb_has_quota_active(root->d_sb, GRPQUOTA))
4598 seq_printf(seq, ",grpquota");
4599 if (sbinfo->qlimits.usrquota_bhardlimit)
4600 seq_printf(seq, ",usrquota_block_hardlimit=%lld",
4601 sbinfo->qlimits.usrquota_bhardlimit);
4602 if (sbinfo->qlimits.grpquota_bhardlimit)
4603 seq_printf(seq, ",grpquota_block_hardlimit=%lld",
4604 sbinfo->qlimits.grpquota_bhardlimit);
4605 if (sbinfo->qlimits.usrquota_ihardlimit)
4606 seq_printf(seq, ",usrquota_inode_hardlimit=%lld",
4607 sbinfo->qlimits.usrquota_ihardlimit);
4608 if (sbinfo->qlimits.grpquota_ihardlimit)
4609 seq_printf(seq, ",grpquota_inode_hardlimit=%lld",
4610 sbinfo->qlimits.grpquota_ihardlimit);
4611 #endif
4612 return 0;
4613 }
4614
4615 #endif /* CONFIG_TMPFS */
4616
shmem_put_super(struct super_block * sb)4617 static void shmem_put_super(struct super_block *sb)
4618 {
4619 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4620
4621 #ifdef CONFIG_TMPFS_QUOTA
4622 shmem_disable_quotas(sb);
4623 #endif
4624 free_percpu(sbinfo->ino_batch);
4625 percpu_counter_destroy(&sbinfo->used_blocks);
4626 mpol_put(sbinfo->mpol);
4627 kfree(sbinfo);
4628 sb->s_fs_info = NULL;
4629 }
4630
shmem_fill_super(struct super_block * sb,struct fs_context * fc)4631 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
4632 {
4633 struct shmem_options *ctx = fc->fs_private;
4634 struct inode *inode;
4635 struct shmem_sb_info *sbinfo;
4636 int error = -ENOMEM;
4637
4638 /* Round up to L1_CACHE_BYTES to resist false sharing */
4639 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
4640 L1_CACHE_BYTES), GFP_KERNEL);
4641 if (!sbinfo)
4642 return error;
4643
4644 sb->s_fs_info = sbinfo;
4645
4646 #ifdef CONFIG_TMPFS
4647 /*
4648 * Per default we only allow half of the physical ram per
4649 * tmpfs instance, limiting inodes to one per page of lowmem;
4650 * but the internal instance is left unlimited.
4651 */
4652 if (!(sb->s_flags & SB_KERNMOUNT)) {
4653 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4654 ctx->blocks = shmem_default_max_blocks();
4655 if (!(ctx->seen & SHMEM_SEEN_INODES))
4656 ctx->inodes = shmem_default_max_inodes();
4657 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4658 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
4659 sbinfo->noswap = ctx->noswap;
4660 } else {
4661 sb->s_flags |= SB_NOUSER;
4662 }
4663 sb->s_export_op = &shmem_export_ops;
4664 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
4665 #else
4666 sb->s_flags |= SB_NOUSER;
4667 #endif
4668 sbinfo->max_blocks = ctx->blocks;
4669 sbinfo->max_inodes = ctx->inodes;
4670 sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
4671 if (sb->s_flags & SB_KERNMOUNT) {
4672 sbinfo->ino_batch = alloc_percpu(ino_t);
4673 if (!sbinfo->ino_batch)
4674 goto failed;
4675 }
4676 sbinfo->uid = ctx->uid;
4677 sbinfo->gid = ctx->gid;
4678 sbinfo->full_inums = ctx->full_inums;
4679 sbinfo->mode = ctx->mode;
4680 sbinfo->huge = ctx->huge;
4681 sbinfo->mpol = ctx->mpol;
4682 ctx->mpol = NULL;
4683
4684 raw_spin_lock_init(&sbinfo->stat_lock);
4685 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
4686 goto failed;
4687 spin_lock_init(&sbinfo->shrinklist_lock);
4688 INIT_LIST_HEAD(&sbinfo->shrinklist);
4689
4690 sb->s_maxbytes = MAX_LFS_FILESIZE;
4691 sb->s_blocksize = PAGE_SIZE;
4692 sb->s_blocksize_bits = PAGE_SHIFT;
4693 sb->s_magic = TMPFS_MAGIC;
4694 sb->s_op = &shmem_ops;
4695 sb->s_time_gran = 1;
4696 #ifdef CONFIG_TMPFS_XATTR
4697 sb->s_xattr = shmem_xattr_handlers;
4698 #endif
4699 #ifdef CONFIG_TMPFS_POSIX_ACL
4700 sb->s_flags |= SB_POSIXACL;
4701 #endif
4702 uuid_t uuid;
4703 uuid_gen(&uuid);
4704 super_set_uuid(sb, uuid.b, sizeof(uuid));
4705
4706 #ifdef CONFIG_TMPFS_QUOTA
4707 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4708 sb->dq_op = &shmem_quota_operations;
4709 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4710 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4711
4712 /* Copy the default limits from ctx into sbinfo */
4713 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4714 sizeof(struct shmem_quota_limits));
4715
4716 if (shmem_enable_quotas(sb, ctx->quota_types))
4717 goto failed;
4718 }
4719 #endif /* CONFIG_TMPFS_QUOTA */
4720
4721 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL,
4722 S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
4723 if (IS_ERR(inode)) {
4724 error = PTR_ERR(inode);
4725 goto failed;
4726 }
4727 inode->i_uid = sbinfo->uid;
4728 inode->i_gid = sbinfo->gid;
4729 sb->s_root = d_make_root(inode);
4730 if (!sb->s_root)
4731 goto failed;
4732 return 0;
4733
4734 failed:
4735 shmem_put_super(sb);
4736 return error;
4737 }
4738
shmem_get_tree(struct fs_context * fc)4739 static int shmem_get_tree(struct fs_context *fc)
4740 {
4741 return get_tree_nodev(fc, shmem_fill_super);
4742 }
4743
shmem_free_fc(struct fs_context * fc)4744 static void shmem_free_fc(struct fs_context *fc)
4745 {
4746 struct shmem_options *ctx = fc->fs_private;
4747
4748 if (ctx) {
4749 mpol_put(ctx->mpol);
4750 kfree(ctx);
4751 }
4752 }
4753
4754 static const struct fs_context_operations shmem_fs_context_ops = {
4755 .free = shmem_free_fc,
4756 .get_tree = shmem_get_tree,
4757 #ifdef CONFIG_TMPFS
4758 .parse_monolithic = shmem_parse_options,
4759 .parse_param = shmem_parse_one,
4760 .reconfigure = shmem_reconfigure,
4761 #endif
4762 };
4763
4764 static struct kmem_cache *shmem_inode_cachep __ro_after_init;
4765
shmem_alloc_inode(struct super_block * sb)4766 static struct inode *shmem_alloc_inode(struct super_block *sb)
4767 {
4768 struct shmem_inode_info *info;
4769 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4770 if (!info)
4771 return NULL;
4772 return &info->vfs_inode;
4773 }
4774
shmem_free_in_core_inode(struct inode * inode)4775 static void shmem_free_in_core_inode(struct inode *inode)
4776 {
4777 if (S_ISLNK(inode->i_mode))
4778 kfree(inode->i_link);
4779 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4780 }
4781
shmem_destroy_inode(struct inode * inode)4782 static void shmem_destroy_inode(struct inode *inode)
4783 {
4784 if (S_ISREG(inode->i_mode))
4785 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4786 if (S_ISDIR(inode->i_mode))
4787 simple_offset_destroy(shmem_get_offset_ctx(inode));
4788 }
4789
shmem_init_inode(void * foo)4790 static void shmem_init_inode(void *foo)
4791 {
4792 struct shmem_inode_info *info = foo;
4793 inode_init_once(&info->vfs_inode);
4794 }
4795
shmem_init_inodecache(void)4796 static void __init shmem_init_inodecache(void)
4797 {
4798 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4799 sizeof(struct shmem_inode_info),
4800 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4801 }
4802
shmem_destroy_inodecache(void)4803 static void __init shmem_destroy_inodecache(void)
4804 {
4805 kmem_cache_destroy(shmem_inode_cachep);
4806 }
4807
4808 /* Keep the page in page cache instead of truncating it */
shmem_error_remove_folio(struct address_space * mapping,struct folio * folio)4809 static int shmem_error_remove_folio(struct address_space *mapping,
4810 struct folio *folio)
4811 {
4812 return 0;
4813 }
4814
4815 static const struct address_space_operations shmem_aops = {
4816 .writepage = shmem_writepage,
4817 .dirty_folio = noop_dirty_folio,
4818 #ifdef CONFIG_TMPFS
4819 .write_begin = shmem_write_begin,
4820 .write_end = shmem_write_end,
4821 #endif
4822 #ifdef CONFIG_MIGRATION
4823 .migrate_folio = migrate_folio,
4824 #endif
4825 .error_remove_folio = shmem_error_remove_folio,
4826 };
4827
4828 static const struct file_operations shmem_file_operations = {
4829 .mmap = shmem_mmap,
4830 .open = shmem_file_open,
4831 .get_unmapped_area = shmem_get_unmapped_area,
4832 #ifdef CONFIG_TMPFS
4833 .llseek = shmem_file_llseek,
4834 .read_iter = shmem_file_read_iter,
4835 .write_iter = shmem_file_write_iter,
4836 .fsync = noop_fsync,
4837 .splice_read = shmem_file_splice_read,
4838 .splice_write = iter_file_splice_write,
4839 .fallocate = shmem_fallocate,
4840 #endif
4841 };
4842
4843 static const struct inode_operations shmem_inode_operations = {
4844 .getattr = shmem_getattr,
4845 .setattr = shmem_setattr,
4846 #ifdef CONFIG_TMPFS_XATTR
4847 .listxattr = shmem_listxattr,
4848 .set_acl = simple_set_acl,
4849 .fileattr_get = shmem_fileattr_get,
4850 .fileattr_set = shmem_fileattr_set,
4851 #endif
4852 };
4853
4854 static const struct inode_operations shmem_dir_inode_operations = {
4855 #ifdef CONFIG_TMPFS
4856 .getattr = shmem_getattr,
4857 .create = shmem_create,
4858 .lookup = simple_lookup,
4859 .link = shmem_link,
4860 .unlink = shmem_unlink,
4861 .symlink = shmem_symlink,
4862 .mkdir = shmem_mkdir,
4863 .rmdir = shmem_rmdir,
4864 .mknod = shmem_mknod,
4865 .rename = shmem_rename2,
4866 .tmpfile = shmem_tmpfile,
4867 .get_offset_ctx = shmem_get_offset_ctx,
4868 #endif
4869 #ifdef CONFIG_TMPFS_XATTR
4870 .listxattr = shmem_listxattr,
4871 .fileattr_get = shmem_fileattr_get,
4872 .fileattr_set = shmem_fileattr_set,
4873 #endif
4874 #ifdef CONFIG_TMPFS_POSIX_ACL
4875 .setattr = shmem_setattr,
4876 .set_acl = simple_set_acl,
4877 #endif
4878 };
4879
4880 static const struct inode_operations shmem_special_inode_operations = {
4881 .getattr = shmem_getattr,
4882 #ifdef CONFIG_TMPFS_XATTR
4883 .listxattr = shmem_listxattr,
4884 #endif
4885 #ifdef CONFIG_TMPFS_POSIX_ACL
4886 .setattr = shmem_setattr,
4887 .set_acl = simple_set_acl,
4888 #endif
4889 };
4890
4891 static const struct super_operations shmem_ops = {
4892 .alloc_inode = shmem_alloc_inode,
4893 .free_inode = shmem_free_in_core_inode,
4894 .destroy_inode = shmem_destroy_inode,
4895 #ifdef CONFIG_TMPFS
4896 .statfs = shmem_statfs,
4897 .show_options = shmem_show_options,
4898 #endif
4899 #ifdef CONFIG_TMPFS_QUOTA
4900 .get_dquots = shmem_get_dquots,
4901 #endif
4902 .evict_inode = shmem_evict_inode,
4903 .drop_inode = generic_delete_inode,
4904 .put_super = shmem_put_super,
4905 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4906 .nr_cached_objects = shmem_unused_huge_count,
4907 .free_cached_objects = shmem_unused_huge_scan,
4908 #endif
4909 };
4910
4911 static const struct vm_operations_struct shmem_vm_ops = {
4912 .fault = shmem_fault,
4913 .map_pages = filemap_map_pages,
4914 #ifdef CONFIG_NUMA
4915 .set_policy = shmem_set_policy,
4916 .get_policy = shmem_get_policy,
4917 #endif
4918 };
4919
4920 static const struct vm_operations_struct shmem_anon_vm_ops = {
4921 .fault = shmem_fault,
4922 .map_pages = filemap_map_pages,
4923 #ifdef CONFIG_NUMA
4924 .set_policy = shmem_set_policy,
4925 .get_policy = shmem_get_policy,
4926 #endif
4927 };
4928
shmem_init_fs_context(struct fs_context * fc)4929 int shmem_init_fs_context(struct fs_context *fc)
4930 {
4931 struct shmem_options *ctx;
4932
4933 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4934 if (!ctx)
4935 return -ENOMEM;
4936
4937 ctx->mode = 0777 | S_ISVTX;
4938 ctx->uid = current_fsuid();
4939 ctx->gid = current_fsgid();
4940
4941 fc->fs_private = ctx;
4942 fc->ops = &shmem_fs_context_ops;
4943 return 0;
4944 }
4945
4946 static struct file_system_type shmem_fs_type = {
4947 .owner = THIS_MODULE,
4948 .name = "tmpfs",
4949 .init_fs_context = shmem_init_fs_context,
4950 #ifdef CONFIG_TMPFS
4951 .parameters = shmem_fs_parameters,
4952 #endif
4953 .kill_sb = kill_litter_super,
4954 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4955 };
4956
shmem_init(void)4957 void __init shmem_init(void)
4958 {
4959 int error;
4960
4961 shmem_init_inodecache();
4962
4963 #ifdef CONFIG_TMPFS_QUOTA
4964 register_quota_format(&shmem_quota_format);
4965 #endif
4966
4967 error = register_filesystem(&shmem_fs_type);
4968 if (error) {
4969 pr_err("Could not register tmpfs\n");
4970 goto out2;
4971 }
4972
4973 shm_mnt = kern_mount(&shmem_fs_type);
4974 if (IS_ERR(shm_mnt)) {
4975 error = PTR_ERR(shm_mnt);
4976 pr_err("Could not kern_mount tmpfs\n");
4977 goto out1;
4978 }
4979
4980 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4981 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4982 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4983 else
4984 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4985
4986 /*
4987 * Default to setting PMD-sized THP to inherit the global setting and
4988 * disable all other multi-size THPs.
4989 */
4990 huge_shmem_orders_inherit = BIT(HPAGE_PMD_ORDER);
4991 #endif
4992 return;
4993
4994 out1:
4995 unregister_filesystem(&shmem_fs_type);
4996 out2:
4997 #ifdef CONFIG_TMPFS_QUOTA
4998 unregister_quota_format(&shmem_quota_format);
4999 #endif
5000 shmem_destroy_inodecache();
5001 shm_mnt = ERR_PTR(error);
5002 }
5003
5004 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
shmem_enabled_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)5005 static ssize_t shmem_enabled_show(struct kobject *kobj,
5006 struct kobj_attribute *attr, char *buf)
5007 {
5008 static const int values[] = {
5009 SHMEM_HUGE_ALWAYS,
5010 SHMEM_HUGE_WITHIN_SIZE,
5011 SHMEM_HUGE_ADVISE,
5012 SHMEM_HUGE_NEVER,
5013 SHMEM_HUGE_DENY,
5014 SHMEM_HUGE_FORCE,
5015 };
5016 int len = 0;
5017 int i;
5018
5019 for (i = 0; i < ARRAY_SIZE(values); i++) {
5020 len += sysfs_emit_at(buf, len,
5021 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
5022 i ? " " : "", shmem_format_huge(values[i]));
5023 }
5024 len += sysfs_emit_at(buf, len, "\n");
5025
5026 return len;
5027 }
5028
shmem_enabled_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)5029 static ssize_t shmem_enabled_store(struct kobject *kobj,
5030 struct kobj_attribute *attr, const char *buf, size_t count)
5031 {
5032 char tmp[16];
5033 int huge;
5034
5035 if (count + 1 > sizeof(tmp))
5036 return -EINVAL;
5037 memcpy(tmp, buf, count);
5038 tmp[count] = '\0';
5039 if (count && tmp[count - 1] == '\n')
5040 tmp[count - 1] = '\0';
5041
5042 huge = shmem_parse_huge(tmp);
5043 if (huge == -EINVAL)
5044 return -EINVAL;
5045 if (!has_transparent_hugepage() &&
5046 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
5047 return -EINVAL;
5048
5049 /* Do not override huge allocation policy with non-PMD sized mTHP */
5050 if (huge == SHMEM_HUGE_FORCE &&
5051 huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
5052 return -EINVAL;
5053
5054 shmem_huge = huge;
5055 if (shmem_huge > SHMEM_HUGE_DENY)
5056 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
5057 return count;
5058 }
5059
5060 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
5061 static DEFINE_SPINLOCK(huge_shmem_orders_lock);
5062
thpsize_shmem_enabled_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)5063 static ssize_t thpsize_shmem_enabled_show(struct kobject *kobj,
5064 struct kobj_attribute *attr, char *buf)
5065 {
5066 int order = to_thpsize(kobj)->order;
5067 const char *output;
5068
5069 if (test_bit(order, &huge_shmem_orders_always))
5070 output = "[always] inherit within_size advise never";
5071 else if (test_bit(order, &huge_shmem_orders_inherit))
5072 output = "always [inherit] within_size advise never";
5073 else if (test_bit(order, &huge_shmem_orders_within_size))
5074 output = "always inherit [within_size] advise never";
5075 else if (test_bit(order, &huge_shmem_orders_madvise))
5076 output = "always inherit within_size [advise] never";
5077 else
5078 output = "always inherit within_size advise [never]";
5079
5080 return sysfs_emit(buf, "%s\n", output);
5081 }
5082
thpsize_shmem_enabled_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)5083 static ssize_t thpsize_shmem_enabled_store(struct kobject *kobj,
5084 struct kobj_attribute *attr,
5085 const char *buf, size_t count)
5086 {
5087 int order = to_thpsize(kobj)->order;
5088 ssize_t ret = count;
5089
5090 if (sysfs_streq(buf, "always")) {
5091 spin_lock(&huge_shmem_orders_lock);
5092 clear_bit(order, &huge_shmem_orders_inherit);
5093 clear_bit(order, &huge_shmem_orders_madvise);
5094 clear_bit(order, &huge_shmem_orders_within_size);
5095 set_bit(order, &huge_shmem_orders_always);
5096 spin_unlock(&huge_shmem_orders_lock);
5097 } else if (sysfs_streq(buf, "inherit")) {
5098 /* Do not override huge allocation policy with non-PMD sized mTHP */
5099 if (shmem_huge == SHMEM_HUGE_FORCE &&
5100 order != HPAGE_PMD_ORDER)
5101 return -EINVAL;
5102
5103 spin_lock(&huge_shmem_orders_lock);
5104 clear_bit(order, &huge_shmem_orders_always);
5105 clear_bit(order, &huge_shmem_orders_madvise);
5106 clear_bit(order, &huge_shmem_orders_within_size);
5107 set_bit(order, &huge_shmem_orders_inherit);
5108 spin_unlock(&huge_shmem_orders_lock);
5109 } else if (sysfs_streq(buf, "within_size")) {
5110 spin_lock(&huge_shmem_orders_lock);
5111 clear_bit(order, &huge_shmem_orders_always);
5112 clear_bit(order, &huge_shmem_orders_inherit);
5113 clear_bit(order, &huge_shmem_orders_madvise);
5114 set_bit(order, &huge_shmem_orders_within_size);
5115 spin_unlock(&huge_shmem_orders_lock);
5116 } else if (sysfs_streq(buf, "advise")) {
5117 spin_lock(&huge_shmem_orders_lock);
5118 clear_bit(order, &huge_shmem_orders_always);
5119 clear_bit(order, &huge_shmem_orders_inherit);
5120 clear_bit(order, &huge_shmem_orders_within_size);
5121 set_bit(order, &huge_shmem_orders_madvise);
5122 spin_unlock(&huge_shmem_orders_lock);
5123 } else if (sysfs_streq(buf, "never")) {
5124 spin_lock(&huge_shmem_orders_lock);
5125 clear_bit(order, &huge_shmem_orders_always);
5126 clear_bit(order, &huge_shmem_orders_inherit);
5127 clear_bit(order, &huge_shmem_orders_within_size);
5128 clear_bit(order, &huge_shmem_orders_madvise);
5129 spin_unlock(&huge_shmem_orders_lock);
5130 } else {
5131 ret = -EINVAL;
5132 }
5133
5134 return ret;
5135 }
5136
5137 struct kobj_attribute thpsize_shmem_enabled_attr =
5138 __ATTR(shmem_enabled, 0644, thpsize_shmem_enabled_show, thpsize_shmem_enabled_store);
5139 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
5140
5141 #else /* !CONFIG_SHMEM */
5142
5143 /*
5144 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
5145 *
5146 * This is intended for small system where the benefits of the full
5147 * shmem code (swap-backed and resource-limited) are outweighed by
5148 * their complexity. On systems without swap this code should be
5149 * effectively equivalent, but much lighter weight.
5150 */
5151
5152 static struct file_system_type shmem_fs_type = {
5153 .name = "tmpfs",
5154 .init_fs_context = ramfs_init_fs_context,
5155 .parameters = ramfs_fs_parameters,
5156 .kill_sb = ramfs_kill_sb,
5157 .fs_flags = FS_USERNS_MOUNT,
5158 };
5159
shmem_init(void)5160 void __init shmem_init(void)
5161 {
5162 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
5163
5164 shm_mnt = kern_mount(&shmem_fs_type);
5165 BUG_ON(IS_ERR(shm_mnt));
5166 }
5167
shmem_unuse(unsigned int type)5168 int shmem_unuse(unsigned int type)
5169 {
5170 return 0;
5171 }
5172
shmem_lock(struct file * file,int lock,struct ucounts * ucounts)5173 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
5174 {
5175 return 0;
5176 }
5177
shmem_unlock_mapping(struct address_space * mapping)5178 void shmem_unlock_mapping(struct address_space *mapping)
5179 {
5180 }
5181
5182 #ifdef CONFIG_MMU
shmem_get_unmapped_area(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)5183 unsigned long shmem_get_unmapped_area(struct file *file,
5184 unsigned long addr, unsigned long len,
5185 unsigned long pgoff, unsigned long flags)
5186 {
5187 return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
5188 }
5189 #endif
5190
shmem_truncate_range(struct inode * inode,loff_t lstart,loff_t lend)5191 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
5192 {
5193 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
5194 }
5195 EXPORT_SYMBOL_GPL(shmem_truncate_range);
5196
5197 #define shmem_vm_ops generic_file_vm_ops
5198 #define shmem_anon_vm_ops generic_file_vm_ops
5199 #define shmem_file_operations ramfs_file_operations
5200 #define shmem_acct_size(flags, size) 0
5201 #define shmem_unacct_size(flags, size) do {} while (0)
5202
shmem_get_inode(struct mnt_idmap * idmap,struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,unsigned long flags)5203 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
5204 struct super_block *sb, struct inode *dir,
5205 umode_t mode, dev_t dev, unsigned long flags)
5206 {
5207 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
5208 return inode ? inode : ERR_PTR(-ENOSPC);
5209 }
5210
5211 #endif /* CONFIG_SHMEM */
5212
5213 /* common code */
5214
__shmem_file_setup(struct vfsmount * mnt,const char * name,loff_t size,unsigned long flags,unsigned int i_flags)5215 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name,
5216 loff_t size, unsigned long flags, unsigned int i_flags)
5217 {
5218 struct inode *inode;
5219 struct file *res;
5220
5221 if (IS_ERR(mnt))
5222 return ERR_CAST(mnt);
5223
5224 if (size < 0 || size > MAX_LFS_FILESIZE)
5225 return ERR_PTR(-EINVAL);
5226
5227 if (shmem_acct_size(flags, size))
5228 return ERR_PTR(-ENOMEM);
5229
5230 if (is_idmapped_mnt(mnt))
5231 return ERR_PTR(-EINVAL);
5232
5233 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
5234 S_IFREG | S_IRWXUGO, 0, flags);
5235 if (IS_ERR(inode)) {
5236 shmem_unacct_size(flags, size);
5237 return ERR_CAST(inode);
5238 }
5239 inode->i_flags |= i_flags;
5240 inode->i_size = size;
5241 clear_nlink(inode); /* It is unlinked */
5242 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
5243 if (!IS_ERR(res))
5244 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
5245 &shmem_file_operations);
5246 if (IS_ERR(res))
5247 iput(inode);
5248 return res;
5249 }
5250
5251 /**
5252 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
5253 * kernel internal. There will be NO LSM permission checks against the
5254 * underlying inode. So users of this interface must do LSM checks at a
5255 * higher layer. The users are the big_key and shm implementations. LSM
5256 * checks are provided at the key or shm level rather than the inode.
5257 * @name: name for dentry (to be seen in /proc/<pid>/maps
5258 * @size: size to be set for the file
5259 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5260 */
shmem_kernel_file_setup(const char * name,loff_t size,unsigned long flags)5261 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
5262 {
5263 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
5264 }
5265 EXPORT_SYMBOL_GPL(shmem_kernel_file_setup);
5266
5267 /**
5268 * shmem_file_setup - get an unlinked file living in tmpfs
5269 * @name: name for dentry (to be seen in /proc/<pid>/maps
5270 * @size: size to be set for the file
5271 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5272 */
shmem_file_setup(const char * name,loff_t size,unsigned long flags)5273 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
5274 {
5275 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
5276 }
5277 EXPORT_SYMBOL_GPL(shmem_file_setup);
5278
5279 /**
5280 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
5281 * @mnt: the tmpfs mount where the file will be created
5282 * @name: name for dentry (to be seen in /proc/<pid>/maps
5283 * @size: size to be set for the file
5284 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5285 */
shmem_file_setup_with_mnt(struct vfsmount * mnt,const char * name,loff_t size,unsigned long flags)5286 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
5287 loff_t size, unsigned long flags)
5288 {
5289 return __shmem_file_setup(mnt, name, size, flags, 0);
5290 }
5291 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
5292
5293 /**
5294 * shmem_zero_setup - setup a shared anonymous mapping
5295 * @vma: the vma to be mmapped is prepared by do_mmap
5296 */
shmem_zero_setup(struct vm_area_struct * vma)5297 int shmem_zero_setup(struct vm_area_struct *vma)
5298 {
5299 struct file *file;
5300 loff_t size = vma->vm_end - vma->vm_start;
5301
5302 /*
5303 * Cloning a new file under mmap_lock leads to a lock ordering conflict
5304 * between XFS directory reading and selinux: since this file is only
5305 * accessible to the user through its mapping, use S_PRIVATE flag to
5306 * bypass file security, in the same way as shmem_kernel_file_setup().
5307 */
5308 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
5309 if (IS_ERR(file))
5310 return PTR_ERR(file);
5311
5312 if (vma->vm_file)
5313 fput(vma->vm_file);
5314 vma->vm_file = file;
5315 vma->vm_ops = &shmem_anon_vm_ops;
5316
5317 return 0;
5318 }
5319
5320 /**
5321 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
5322 * @mapping: the folio's address_space
5323 * @index: the folio index
5324 * @gfp: the page allocator flags to use if allocating
5325 *
5326 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
5327 * with any new page allocations done using the specified allocation flags.
5328 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
5329 * suit tmpfs, since it may have pages in swapcache, and needs to find those
5330 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
5331 *
5332 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
5333 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
5334 */
shmem_read_folio_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)5335 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
5336 pgoff_t index, gfp_t gfp)
5337 {
5338 #ifdef CONFIG_SHMEM
5339 struct inode *inode = mapping->host;
5340 struct folio *folio;
5341 int error;
5342
5343 error = shmem_get_folio_gfp(inode, index, 0, &folio, SGP_CACHE,
5344 gfp, NULL, NULL);
5345 if (error)
5346 return ERR_PTR(error);
5347
5348 folio_unlock(folio);
5349 return folio;
5350 #else
5351 /*
5352 * The tiny !SHMEM case uses ramfs without swap
5353 */
5354 return mapping_read_folio_gfp(mapping, index, gfp);
5355 #endif
5356 }
5357 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
5358
shmem_read_mapping_page_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)5359 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
5360 pgoff_t index, gfp_t gfp)
5361 {
5362 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
5363 struct page *page;
5364
5365 if (IS_ERR(folio))
5366 return &folio->page;
5367
5368 page = folio_file_page(folio, index);
5369 if (PageHWPoison(page)) {
5370 folio_put(folio);
5371 return ERR_PTR(-EIO);
5372 }
5373
5374 return page;
5375 }
5376 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
5377