Line data Source code
1 : /*
2 : * Resizable virtual memory filesystem for Linux.
3 : *
4 : * Copyright (C) 2000 Linus Torvalds.
5 : * 2000 Transmeta Corp.
6 : * 2000-2001 Christoph Rohland
7 : * 2000-2001 SAP AG
8 : * 2002 Red Hat Inc.
9 : * Copyright (C) 2002-2011 Hugh Dickins.
10 : * Copyright (C) 2011 Google Inc.
11 : * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 : * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 : *
14 : * Extended attribute support for tmpfs:
15 : * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 : * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 : *
18 : * tiny-shmem:
19 : * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 : *
21 : * This file is released under the GPL.
22 : */
23 :
24 : #include <linux/fs.h>
25 : #include <linux/init.h>
26 : #include <linux/vfs.h>
27 : #include <linux/mount.h>
28 : #include <linux/ramfs.h>
29 : #include <linux/pagemap.h>
30 : #include <linux/file.h>
31 : #include <linux/fileattr.h>
32 : #include <linux/mm.h>
33 : #include <linux/random.h>
34 : #include <linux/sched/signal.h>
35 : #include <linux/export.h>
36 : #include <linux/shmem_fs.h>
37 : #include <linux/swap.h>
38 : #include <linux/uio.h>
39 : #include <linux/hugetlb.h>
40 : #include <linux/fs_parser.h>
41 : #include <linux/swapfile.h>
42 : #include <linux/iversion.h>
43 : #include "swap.h"
44 :
45 : static struct vfsmount *shm_mnt;
46 :
47 : #ifdef CONFIG_SHMEM
48 : /*
49 : * This virtual memory filesystem is heavily based on the ramfs. It
50 : * extends ramfs by the ability to use swap and honor resource limits
51 : * which makes it a completely usable filesystem.
52 : */
53 :
54 : #include <linux/xattr.h>
55 : #include <linux/exportfs.h>
56 : #include <linux/posix_acl.h>
57 : #include <linux/posix_acl_xattr.h>
58 : #include <linux/mman.h>
59 : #include <linux/string.h>
60 : #include <linux/slab.h>
61 : #include <linux/backing-dev.h>
62 : #include <linux/writeback.h>
63 : #include <linux/pagevec.h>
64 : #include <linux/percpu_counter.h>
65 : #include <linux/falloc.h>
66 : #include <linux/splice.h>
67 : #include <linux/security.h>
68 : #include <linux/swapops.h>
69 : #include <linux/mempolicy.h>
70 : #include <linux/namei.h>
71 : #include <linux/ctype.h>
72 : #include <linux/migrate.h>
73 : #include <linux/highmem.h>
74 : #include <linux/seq_file.h>
75 : #include <linux/magic.h>
76 : #include <linux/syscalls.h>
77 : #include <linux/fcntl.h>
78 : #include <uapi/linux/memfd.h>
79 : #include <linux/userfaultfd_k.h>
80 : #include <linux/rmap.h>
81 : #include <linux/uuid.h>
82 :
83 : #include <linux/uaccess.h>
84 :
85 : #include "internal.h"
86 :
87 : #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 : #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 :
90 : /* Pretend that each entry is of this size in directory's i_size */
91 : #define BOGO_DIRENT_SIZE 20
92 :
93 : /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 : #define SHORT_SYMLINK_LEN 128
95 :
96 : /*
97 : * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 : * inode->i_private (with i_rwsem making sure that it has only one user at
99 : * a time): we would prefer not to enlarge the shmem inode just for that.
100 : */
101 : struct shmem_falloc {
102 : wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 : pgoff_t start; /* start of range currently being fallocated */
104 : pgoff_t next; /* the next page offset to be fallocated */
105 : pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 : pgoff_t nr_unswapped; /* how often writepage refused to swap out */
107 : };
108 :
109 : struct shmem_options {
110 : unsigned long long blocks;
111 : unsigned long long inodes;
112 : struct mempolicy *mpol;
113 : kuid_t uid;
114 : kgid_t gid;
115 : umode_t mode;
116 : bool full_inums;
117 : int huge;
118 : int seen;
119 : #define SHMEM_SEEN_BLOCKS 1
120 : #define SHMEM_SEEN_INODES 2
121 : #define SHMEM_SEEN_HUGE 4
122 : #define SHMEM_SEEN_INUMS 8
123 : };
124 :
125 : #ifdef CONFIG_TMPFS
126 : static unsigned long shmem_default_max_blocks(void)
127 : {
128 : return totalram_pages() / 2;
129 : }
130 :
131 : static unsigned long shmem_default_max_inodes(void)
132 : {
133 : unsigned long nr_pages = totalram_pages();
134 :
135 : return min(nr_pages - totalhigh_pages(), nr_pages / 2);
136 : }
137 : #endif
138 :
139 : static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
140 : struct folio **foliop, enum sgp_type sgp,
141 : gfp_t gfp, struct vm_area_struct *vma,
142 : vm_fault_t *fault_type);
143 :
144 : static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
145 : {
146 : return sb->s_fs_info;
147 : }
148 :
149 : /*
150 : * shmem_file_setup pre-accounts the whole fixed size of a VM object,
151 : * for shared memory and for shared anonymous (/dev/zero) mappings
152 : * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
153 : * consistent with the pre-accounting of private mappings ...
154 : */
155 0 : static inline int shmem_acct_size(unsigned long flags, loff_t size)
156 : {
157 0 : return (flags & VM_NORESERVE) ?
158 0 : 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
159 : }
160 :
161 : static inline void shmem_unacct_size(unsigned long flags, loff_t size)
162 : {
163 0 : if (!(flags & VM_NORESERVE))
164 0 : vm_unacct_memory(VM_ACCT(size));
165 : }
166 :
167 0 : static inline int shmem_reacct_size(unsigned long flags,
168 : loff_t oldsize, loff_t newsize)
169 : {
170 0 : if (!(flags & VM_NORESERVE)) {
171 0 : if (VM_ACCT(newsize) > VM_ACCT(oldsize))
172 0 : return security_vm_enough_memory_mm(current->mm,
173 0 : VM_ACCT(newsize) - VM_ACCT(oldsize));
174 0 : else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
175 0 : vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
176 : }
177 : return 0;
178 : }
179 :
180 : /*
181 : * ... whereas tmpfs objects are accounted incrementally as
182 : * pages are allocated, in order to allow large sparse files.
183 : * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
184 : * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
185 : */
186 : static inline int shmem_acct_block(unsigned long flags, long pages)
187 : {
188 0 : if (!(flags & VM_NORESERVE))
189 : return 0;
190 :
191 0 : return security_vm_enough_memory_mm(current->mm,
192 : pages * VM_ACCT(PAGE_SIZE));
193 : }
194 :
195 : static inline void shmem_unacct_blocks(unsigned long flags, long pages)
196 : {
197 0 : if (flags & VM_NORESERVE)
198 : vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
199 : }
200 :
201 0 : static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
202 : {
203 0 : struct shmem_inode_info *info = SHMEM_I(inode);
204 0 : struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
205 :
206 0 : if (shmem_acct_block(info->flags, pages))
207 : return false;
208 :
209 0 : if (sbinfo->max_blocks) {
210 0 : if (percpu_counter_compare(&sbinfo->used_blocks,
211 0 : sbinfo->max_blocks - pages) > 0)
212 : goto unacct;
213 0 : percpu_counter_add(&sbinfo->used_blocks, pages);
214 : }
215 :
216 : return true;
217 :
218 : unacct:
219 0 : shmem_unacct_blocks(info->flags, pages);
220 : return false;
221 : }
222 :
223 0 : static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
224 : {
225 0 : struct shmem_inode_info *info = SHMEM_I(inode);
226 0 : struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
227 :
228 0 : if (sbinfo->max_blocks)
229 0 : percpu_counter_sub(&sbinfo->used_blocks, pages);
230 0 : shmem_unacct_blocks(info->flags, pages);
231 0 : }
232 :
233 : static const struct super_operations shmem_ops;
234 : const struct address_space_operations shmem_aops;
235 : static const struct file_operations shmem_file_operations;
236 : static const struct inode_operations shmem_inode_operations;
237 : static const struct inode_operations shmem_dir_inode_operations;
238 : static const struct inode_operations shmem_special_inode_operations;
239 : static const struct vm_operations_struct shmem_vm_ops;
240 : static const struct vm_operations_struct shmem_anon_vm_ops;
241 : static struct file_system_type shmem_fs_type;
242 :
243 0 : bool vma_is_anon_shmem(struct vm_area_struct *vma)
244 : {
245 0 : return vma->vm_ops == &shmem_anon_vm_ops;
246 : }
247 :
248 0 : bool vma_is_shmem(struct vm_area_struct *vma)
249 : {
250 0 : return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
251 : }
252 :
253 : static LIST_HEAD(shmem_swaplist);
254 : static DEFINE_MUTEX(shmem_swaplist_mutex);
255 :
256 : /*
257 : * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
258 : * produces a novel ino for the newly allocated inode.
259 : *
260 : * It may also be called when making a hard link to permit the space needed by
261 : * each dentry. However, in that case, no new inode number is needed since that
262 : * internally draws from another pool of inode numbers (currently global
263 : * get_next_ino()). This case is indicated by passing NULL as inop.
264 : */
265 : #define SHMEM_INO_BATCH 1024
266 1 : static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
267 : {
268 1 : struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
269 : ino_t ino;
270 :
271 1 : if (!(sb->s_flags & SB_KERNMOUNT)) {
272 0 : raw_spin_lock(&sbinfo->stat_lock);
273 0 : if (sbinfo->max_inodes) {
274 0 : if (!sbinfo->free_inodes) {
275 0 : raw_spin_unlock(&sbinfo->stat_lock);
276 0 : return -ENOSPC;
277 : }
278 0 : sbinfo->free_inodes--;
279 : }
280 0 : if (inop) {
281 0 : ino = sbinfo->next_ino++;
282 0 : if (unlikely(is_zero_ino(ino)))
283 0 : ino = sbinfo->next_ino++;
284 0 : if (unlikely(!sbinfo->full_inums &&
285 : ino > UINT_MAX)) {
286 : /*
287 : * Emulate get_next_ino uint wraparound for
288 : * compatibility
289 : */
290 : if (IS_ENABLED(CONFIG_64BIT))
291 0 : pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
292 : __func__, MINOR(sb->s_dev));
293 : sbinfo->next_ino = 1;
294 0 : ino = sbinfo->next_ino++;
295 : }
296 0 : *inop = ino;
297 : }
298 0 : raw_spin_unlock(&sbinfo->stat_lock);
299 1 : } else if (inop) {
300 : /*
301 : * __shmem_file_setup, one of our callers, is lock-free: it
302 : * doesn't hold stat_lock in shmem_reserve_inode since
303 : * max_inodes is always 0, and is called from potentially
304 : * unknown contexts. As such, use a per-cpu batched allocator
305 : * which doesn't require the per-sb stat_lock unless we are at
306 : * the batch boundary.
307 : *
308 : * We don't need to worry about inode{32,64} since SB_KERNMOUNT
309 : * shmem mounts are not exposed to userspace, so we don't need
310 : * to worry about things like glibc compatibility.
311 : */
312 : ino_t *next_ino;
313 :
314 1 : next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
315 1 : ino = *next_ino;
316 1 : if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
317 1 : raw_spin_lock(&sbinfo->stat_lock);
318 1 : ino = sbinfo->next_ino;
319 1 : sbinfo->next_ino += SHMEM_INO_BATCH;
320 1 : raw_spin_unlock(&sbinfo->stat_lock);
321 1 : if (unlikely(is_zero_ino(ino)))
322 1 : ino++;
323 : }
324 1 : *inop = ino;
325 1 : *next_ino = ++ino;
326 1 : put_cpu();
327 : }
328 :
329 : return 0;
330 : }
331 :
332 : static void shmem_free_inode(struct super_block *sb)
333 : {
334 0 : struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
335 0 : if (sbinfo->max_inodes) {
336 0 : raw_spin_lock(&sbinfo->stat_lock);
337 0 : sbinfo->free_inodes++;
338 0 : raw_spin_unlock(&sbinfo->stat_lock);
339 : }
340 : }
341 :
342 : /**
343 : * shmem_recalc_inode - recalculate the block usage of an inode
344 : * @inode: inode to recalc
345 : *
346 : * We have to calculate the free blocks since the mm can drop
347 : * undirtied hole pages behind our back.
348 : *
349 : * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
350 : * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
351 : *
352 : * It has to be called with the spinlock held.
353 : */
354 0 : static void shmem_recalc_inode(struct inode *inode)
355 : {
356 0 : struct shmem_inode_info *info = SHMEM_I(inode);
357 : long freed;
358 :
359 0 : freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
360 0 : if (freed > 0) {
361 0 : info->alloced -= freed;
362 0 : inode->i_blocks -= freed * BLOCKS_PER_PAGE;
363 0 : shmem_inode_unacct_blocks(inode, freed);
364 : }
365 0 : }
366 :
367 0 : bool shmem_charge(struct inode *inode, long pages)
368 : {
369 0 : struct shmem_inode_info *info = SHMEM_I(inode);
370 : unsigned long flags;
371 :
372 0 : if (!shmem_inode_acct_block(inode, pages))
373 : return false;
374 :
375 : /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
376 0 : inode->i_mapping->nrpages += pages;
377 :
378 0 : spin_lock_irqsave(&info->lock, flags);
379 0 : info->alloced += pages;
380 0 : inode->i_blocks += pages * BLOCKS_PER_PAGE;
381 0 : shmem_recalc_inode(inode);
382 0 : spin_unlock_irqrestore(&info->lock, flags);
383 :
384 0 : return true;
385 : }
386 :
387 0 : void shmem_uncharge(struct inode *inode, long pages)
388 : {
389 0 : struct shmem_inode_info *info = SHMEM_I(inode);
390 : unsigned long flags;
391 :
392 : /* nrpages adjustment done by __filemap_remove_folio() or caller */
393 :
394 0 : spin_lock_irqsave(&info->lock, flags);
395 0 : info->alloced -= pages;
396 0 : inode->i_blocks -= pages * BLOCKS_PER_PAGE;
397 0 : shmem_recalc_inode(inode);
398 0 : spin_unlock_irqrestore(&info->lock, flags);
399 :
400 0 : shmem_inode_unacct_blocks(inode, pages);
401 0 : }
402 :
403 : /*
404 : * Replace item expected in xarray by a new item, while holding xa_lock.
405 : */
406 0 : static int shmem_replace_entry(struct address_space *mapping,
407 : pgoff_t index, void *expected, void *replacement)
408 : {
409 0 : XA_STATE(xas, &mapping->i_pages, index);
410 : void *item;
411 :
412 : VM_BUG_ON(!expected);
413 : VM_BUG_ON(!replacement);
414 0 : item = xas_load(&xas);
415 0 : if (item != expected)
416 : return -ENOENT;
417 0 : xas_store(&xas, replacement);
418 0 : return 0;
419 : }
420 :
421 : /*
422 : * Sometimes, before we decide whether to proceed or to fail, we must check
423 : * that an entry was not already brought back from swap by a racing thread.
424 : *
425 : * Checking page is not enough: by the time a SwapCache page is locked, it
426 : * might be reused, and again be SwapCache, using the same swap as before.
427 : */
428 0 : static bool shmem_confirm_swap(struct address_space *mapping,
429 : pgoff_t index, swp_entry_t swap)
430 : {
431 0 : return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
432 : }
433 :
434 : /*
435 : * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
436 : *
437 : * SHMEM_HUGE_NEVER:
438 : * disables huge pages for the mount;
439 : * SHMEM_HUGE_ALWAYS:
440 : * enables huge pages for the mount;
441 : * SHMEM_HUGE_WITHIN_SIZE:
442 : * only allocate huge pages if the page will be fully within i_size,
443 : * also respect fadvise()/madvise() hints;
444 : * SHMEM_HUGE_ADVISE:
445 : * only allocate huge pages if requested with fadvise()/madvise();
446 : */
447 :
448 : #define SHMEM_HUGE_NEVER 0
449 : #define SHMEM_HUGE_ALWAYS 1
450 : #define SHMEM_HUGE_WITHIN_SIZE 2
451 : #define SHMEM_HUGE_ADVISE 3
452 :
453 : /*
454 : * Special values.
455 : * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
456 : *
457 : * SHMEM_HUGE_DENY:
458 : * disables huge on shm_mnt and all mounts, for emergency use;
459 : * SHMEM_HUGE_FORCE:
460 : * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
461 : *
462 : */
463 : #define SHMEM_HUGE_DENY (-1)
464 : #define SHMEM_HUGE_FORCE (-2)
465 :
466 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
467 : /* ifdef here to avoid bloating shmem.o when not necessary */
468 :
469 : static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
470 :
471 : bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
472 : struct mm_struct *mm, unsigned long vm_flags)
473 : {
474 : loff_t i_size;
475 :
476 : if (!S_ISREG(inode->i_mode))
477 : return false;
478 : if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
479 : return false;
480 : if (shmem_huge == SHMEM_HUGE_DENY)
481 : return false;
482 : if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
483 : return true;
484 :
485 : switch (SHMEM_SB(inode->i_sb)->huge) {
486 : case SHMEM_HUGE_ALWAYS:
487 : return true;
488 : case SHMEM_HUGE_WITHIN_SIZE:
489 : index = round_up(index + 1, HPAGE_PMD_NR);
490 : i_size = round_up(i_size_read(inode), PAGE_SIZE);
491 : if (i_size >> PAGE_SHIFT >= index)
492 : return true;
493 : fallthrough;
494 : case SHMEM_HUGE_ADVISE:
495 : if (mm && (vm_flags & VM_HUGEPAGE))
496 : return true;
497 : fallthrough;
498 : default:
499 : return false;
500 : }
501 : }
502 :
503 : #if defined(CONFIG_SYSFS)
504 : static int shmem_parse_huge(const char *str)
505 : {
506 : if (!strcmp(str, "never"))
507 : return SHMEM_HUGE_NEVER;
508 : if (!strcmp(str, "always"))
509 : return SHMEM_HUGE_ALWAYS;
510 : if (!strcmp(str, "within_size"))
511 : return SHMEM_HUGE_WITHIN_SIZE;
512 : if (!strcmp(str, "advise"))
513 : return SHMEM_HUGE_ADVISE;
514 : if (!strcmp(str, "deny"))
515 : return SHMEM_HUGE_DENY;
516 : if (!strcmp(str, "force"))
517 : return SHMEM_HUGE_FORCE;
518 : return -EINVAL;
519 : }
520 : #endif
521 :
522 : #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
523 : static const char *shmem_format_huge(int huge)
524 : {
525 : switch (huge) {
526 : case SHMEM_HUGE_NEVER:
527 : return "never";
528 : case SHMEM_HUGE_ALWAYS:
529 : return "always";
530 : case SHMEM_HUGE_WITHIN_SIZE:
531 : return "within_size";
532 : case SHMEM_HUGE_ADVISE:
533 : return "advise";
534 : case SHMEM_HUGE_DENY:
535 : return "deny";
536 : case SHMEM_HUGE_FORCE:
537 : return "force";
538 : default:
539 : VM_BUG_ON(1);
540 : return "bad_val";
541 : }
542 : }
543 : #endif
544 :
545 : static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
546 : struct shrink_control *sc, unsigned long nr_to_split)
547 : {
548 : LIST_HEAD(list), *pos, *next;
549 : LIST_HEAD(to_remove);
550 : struct inode *inode;
551 : struct shmem_inode_info *info;
552 : struct folio *folio;
553 : unsigned long batch = sc ? sc->nr_to_scan : 128;
554 : int split = 0;
555 :
556 : if (list_empty(&sbinfo->shrinklist))
557 : return SHRINK_STOP;
558 :
559 : spin_lock(&sbinfo->shrinklist_lock);
560 : list_for_each_safe(pos, next, &sbinfo->shrinklist) {
561 : info = list_entry(pos, struct shmem_inode_info, shrinklist);
562 :
563 : /* pin the inode */
564 : inode = igrab(&info->vfs_inode);
565 :
566 : /* inode is about to be evicted */
567 : if (!inode) {
568 : list_del_init(&info->shrinklist);
569 : goto next;
570 : }
571 :
572 : /* Check if there's anything to gain */
573 : if (round_up(inode->i_size, PAGE_SIZE) ==
574 : round_up(inode->i_size, HPAGE_PMD_SIZE)) {
575 : list_move(&info->shrinklist, &to_remove);
576 : goto next;
577 : }
578 :
579 : list_move(&info->shrinklist, &list);
580 : next:
581 : sbinfo->shrinklist_len--;
582 : if (!--batch)
583 : break;
584 : }
585 : spin_unlock(&sbinfo->shrinklist_lock);
586 :
587 : list_for_each_safe(pos, next, &to_remove) {
588 : info = list_entry(pos, struct shmem_inode_info, shrinklist);
589 : inode = &info->vfs_inode;
590 : list_del_init(&info->shrinklist);
591 : iput(inode);
592 : }
593 :
594 : list_for_each_safe(pos, next, &list) {
595 : int ret;
596 : pgoff_t index;
597 :
598 : info = list_entry(pos, struct shmem_inode_info, shrinklist);
599 : inode = &info->vfs_inode;
600 :
601 : if (nr_to_split && split >= nr_to_split)
602 : goto move_back;
603 :
604 : index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
605 : folio = filemap_get_folio(inode->i_mapping, index);
606 : if (!folio)
607 : goto drop;
608 :
609 : /* No huge page at the end of the file: nothing to split */
610 : if (!folio_test_large(folio)) {
611 : folio_put(folio);
612 : goto drop;
613 : }
614 :
615 : /*
616 : * Move the inode on the list back to shrinklist if we failed
617 : * to lock the page at this time.
618 : *
619 : * Waiting for the lock may lead to deadlock in the
620 : * reclaim path.
621 : */
622 : if (!folio_trylock(folio)) {
623 : folio_put(folio);
624 : goto move_back;
625 : }
626 :
627 : ret = split_folio(folio);
628 : folio_unlock(folio);
629 : folio_put(folio);
630 :
631 : /* If split failed move the inode on the list back to shrinklist */
632 : if (ret)
633 : goto move_back;
634 :
635 : split++;
636 : drop:
637 : list_del_init(&info->shrinklist);
638 : goto put;
639 : move_back:
640 : /*
641 : * Make sure the inode is either on the global list or deleted
642 : * from any local list before iput() since it could be deleted
643 : * in another thread once we put the inode (then the local list
644 : * is corrupted).
645 : */
646 : spin_lock(&sbinfo->shrinklist_lock);
647 : list_move(&info->shrinklist, &sbinfo->shrinklist);
648 : sbinfo->shrinklist_len++;
649 : spin_unlock(&sbinfo->shrinklist_lock);
650 : put:
651 : iput(inode);
652 : }
653 :
654 : return split;
655 : }
656 :
657 : static long shmem_unused_huge_scan(struct super_block *sb,
658 : struct shrink_control *sc)
659 : {
660 : struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
661 :
662 : if (!READ_ONCE(sbinfo->shrinklist_len))
663 : return SHRINK_STOP;
664 :
665 : return shmem_unused_huge_shrink(sbinfo, sc, 0);
666 : }
667 :
668 : static long shmem_unused_huge_count(struct super_block *sb,
669 : struct shrink_control *sc)
670 : {
671 : struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
672 : return READ_ONCE(sbinfo->shrinklist_len);
673 : }
674 : #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
675 :
676 : #define shmem_huge SHMEM_HUGE_DENY
677 :
678 0 : bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
679 : struct mm_struct *mm, unsigned long vm_flags)
680 : {
681 0 : return false;
682 : }
683 :
684 : static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
685 : struct shrink_control *sc, unsigned long nr_to_split)
686 : {
687 : return 0;
688 : }
689 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
690 :
691 : /*
692 : * Like filemap_add_folio, but error if expected item has gone.
693 : */
694 0 : static int shmem_add_to_page_cache(struct folio *folio,
695 : struct address_space *mapping,
696 : pgoff_t index, void *expected, gfp_t gfp,
697 : struct mm_struct *charge_mm)
698 : {
699 0 : XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
700 0 : long nr = folio_nr_pages(folio);
701 : int error;
702 :
703 : VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
704 : VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
705 : VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
706 : VM_BUG_ON(expected && folio_test_large(folio));
707 :
708 0 : folio_ref_add(folio, nr);
709 0 : folio->mapping = mapping;
710 0 : folio->index = index;
711 :
712 : if (!folio_test_swapcache(folio)) {
713 : error = mem_cgroup_charge(folio, charge_mm, gfp);
714 : if (error) {
715 : if (folio_test_pmd_mappable(folio)) {
716 : count_vm_event(THP_FILE_FALLBACK);
717 : count_vm_event(THP_FILE_FALLBACK_CHARGE);
718 : }
719 : goto error;
720 : }
721 : }
722 : folio_throttle_swaprate(folio, gfp);
723 :
724 : do {
725 0 : xas_lock_irq(&xas);
726 0 : if (expected != xas_find_conflict(&xas)) {
727 0 : xas_set_err(&xas, -EEXIST);
728 : goto unlock;
729 : }
730 0 : if (expected && xas_find_conflict(&xas)) {
731 0 : xas_set_err(&xas, -EEXIST);
732 : goto unlock;
733 : }
734 0 : xas_store(&xas, folio);
735 0 : if (xas_error(&xas))
736 : goto unlock;
737 0 : if (folio_test_pmd_mappable(folio)) {
738 : count_vm_event(THP_FILE_ALLOC);
739 : __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
740 : }
741 0 : mapping->nrpages += nr;
742 0 : __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
743 0 : __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
744 : unlock:
745 0 : xas_unlock_irq(&xas);
746 0 : } while (xas_nomem(&xas, gfp));
747 :
748 0 : if (xas_error(&xas)) {
749 0 : error = xas_error(&xas);
750 : goto error;
751 : }
752 :
753 : return 0;
754 : error:
755 0 : folio->mapping = NULL;
756 0 : folio_ref_sub(folio, nr);
757 : return error;
758 : }
759 :
760 : /*
761 : * Like delete_from_page_cache, but substitutes swap for @folio.
762 : */
763 0 : static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
764 : {
765 0 : struct address_space *mapping = folio->mapping;
766 0 : long nr = folio_nr_pages(folio);
767 : int error;
768 :
769 0 : xa_lock_irq(&mapping->i_pages);
770 0 : error = shmem_replace_entry(mapping, folio->index, folio, radswap);
771 0 : folio->mapping = NULL;
772 0 : mapping->nrpages -= nr;
773 0 : __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
774 0 : __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
775 0 : xa_unlock_irq(&mapping->i_pages);
776 0 : folio_put(folio);
777 0 : BUG_ON(error);
778 0 : }
779 :
780 : /*
781 : * Remove swap entry from page cache, free the swap and its page cache.
782 : */
783 0 : static int shmem_free_swap(struct address_space *mapping,
784 : pgoff_t index, void *radswap)
785 : {
786 : void *old;
787 :
788 0 : old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
789 0 : if (old != radswap)
790 : return -ENOENT;
791 0 : free_swap_and_cache(radix_to_swp_entry(radswap));
792 0 : return 0;
793 : }
794 :
795 : /*
796 : * Determine (in bytes) how many of the shmem object's pages mapped by the
797 : * given offsets are swapped out.
798 : *
799 : * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
800 : * as long as the inode doesn't go away and racy results are not a problem.
801 : */
802 0 : unsigned long shmem_partial_swap_usage(struct address_space *mapping,
803 : pgoff_t start, pgoff_t end)
804 : {
805 0 : XA_STATE(xas, &mapping->i_pages, start);
806 : struct page *page;
807 0 : unsigned long swapped = 0;
808 :
809 : rcu_read_lock();
810 0 : xas_for_each(&xas, page, end - 1) {
811 0 : if (xas_retry(&xas, page))
812 0 : continue;
813 0 : if (xa_is_value(page))
814 0 : swapped++;
815 :
816 0 : if (need_resched()) {
817 0 : xas_pause(&xas);
818 : cond_resched_rcu();
819 : }
820 : }
821 :
822 : rcu_read_unlock();
823 :
824 0 : return swapped << PAGE_SHIFT;
825 : }
826 :
827 : /*
828 : * Determine (in bytes) how many of the shmem object's pages mapped by the
829 : * given vma is swapped out.
830 : *
831 : * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
832 : * as long as the inode doesn't go away and racy results are not a problem.
833 : */
834 0 : unsigned long shmem_swap_usage(struct vm_area_struct *vma)
835 : {
836 0 : struct inode *inode = file_inode(vma->vm_file);
837 0 : struct shmem_inode_info *info = SHMEM_I(inode);
838 0 : struct address_space *mapping = inode->i_mapping;
839 : unsigned long swapped;
840 :
841 : /* Be careful as we don't hold info->lock */
842 0 : swapped = READ_ONCE(info->swapped);
843 :
844 : /*
845 : * The easier cases are when the shmem object has nothing in swap, or
846 : * the vma maps it whole. Then we can simply use the stats that we
847 : * already track.
848 : */
849 0 : if (!swapped)
850 : return 0;
851 :
852 0 : if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
853 0 : return swapped << PAGE_SHIFT;
854 :
855 : /* Here comes the more involved part */
856 0 : return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
857 0 : vma->vm_pgoff + vma_pages(vma));
858 : }
859 :
860 : /*
861 : * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
862 : */
863 0 : void shmem_unlock_mapping(struct address_space *mapping)
864 : {
865 : struct folio_batch fbatch;
866 0 : pgoff_t index = 0;
867 :
868 0 : folio_batch_init(&fbatch);
869 : /*
870 : * Minor point, but we might as well stop if someone else SHM_LOCKs it.
871 : */
872 0 : while (!mapping_unevictable(mapping) &&
873 0 : filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
874 0 : check_move_unevictable_folios(&fbatch);
875 0 : folio_batch_release(&fbatch);
876 0 : cond_resched();
877 : }
878 0 : }
879 :
880 0 : static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
881 : {
882 : struct folio *folio;
883 :
884 : /*
885 : * At first avoid shmem_get_folio(,,,SGP_READ): that fails
886 : * beyond i_size, and reports fallocated pages as holes.
887 : */
888 0 : folio = __filemap_get_folio(inode->i_mapping, index,
889 : FGP_ENTRY | FGP_LOCK, 0);
890 0 : if (!xa_is_value(folio))
891 : return folio;
892 : /*
893 : * But read a page back from swap if any of it is within i_size
894 : * (although in some cases this is just a waste of time).
895 : */
896 0 : folio = NULL;
897 0 : shmem_get_folio(inode, index, &folio, SGP_READ);
898 0 : return folio;
899 : }
900 :
901 : /*
902 : * Remove range of pages and swap entries from page cache, and free them.
903 : * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
904 : */
905 0 : static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
906 : bool unfalloc)
907 : {
908 0 : struct address_space *mapping = inode->i_mapping;
909 0 : struct shmem_inode_info *info = SHMEM_I(inode);
910 0 : pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
911 0 : pgoff_t end = (lend + 1) >> PAGE_SHIFT;
912 : struct folio_batch fbatch;
913 : pgoff_t indices[PAGEVEC_SIZE];
914 : struct folio *folio;
915 : bool same_folio;
916 0 : long nr_swaps_freed = 0;
917 : pgoff_t index;
918 : int i;
919 :
920 0 : if (lend == -1)
921 0 : end = -1; /* unsigned, so actually very big */
922 :
923 0 : if (info->fallocend > start && info->fallocend <= end && !unfalloc)
924 0 : info->fallocend = start;
925 :
926 0 : folio_batch_init(&fbatch);
927 0 : index = start;
928 0 : while (index < end && find_lock_entries(mapping, &index, end - 1,
929 : &fbatch, indices)) {
930 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
931 0 : folio = fbatch.folios[i];
932 :
933 0 : if (xa_is_value(folio)) {
934 0 : if (unfalloc)
935 0 : continue;
936 0 : nr_swaps_freed += !shmem_free_swap(mapping,
937 : indices[i], folio);
938 0 : continue;
939 : }
940 :
941 0 : if (!unfalloc || !folio_test_uptodate(folio))
942 0 : truncate_inode_folio(mapping, folio);
943 0 : folio_unlock(folio);
944 : }
945 0 : folio_batch_remove_exceptionals(&fbatch);
946 0 : folio_batch_release(&fbatch);
947 0 : cond_resched();
948 : }
949 :
950 : /*
951 : * When undoing a failed fallocate, we want none of the partial folio
952 : * zeroing and splitting below, but shall want to truncate the whole
953 : * folio when !uptodate indicates that it was added by this fallocate,
954 : * even when [lstart, lend] covers only a part of the folio.
955 : */
956 0 : if (unfalloc)
957 : goto whole_folios;
958 :
959 0 : same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
960 0 : folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
961 0 : if (folio) {
962 0 : same_folio = lend < folio_pos(folio) + folio_size(folio);
963 0 : folio_mark_dirty(folio);
964 0 : if (!truncate_inode_partial_folio(folio, lstart, lend)) {
965 0 : start = folio->index + folio_nr_pages(folio);
966 0 : if (same_folio)
967 0 : end = folio->index;
968 : }
969 0 : folio_unlock(folio);
970 : folio_put(folio);
971 : folio = NULL;
972 : }
973 :
974 0 : if (!same_folio)
975 0 : folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
976 0 : if (folio) {
977 0 : folio_mark_dirty(folio);
978 0 : if (!truncate_inode_partial_folio(folio, lstart, lend))
979 0 : end = folio->index;
980 0 : folio_unlock(folio);
981 : folio_put(folio);
982 : }
983 :
984 : whole_folios:
985 :
986 0 : index = start;
987 0 : while (index < end) {
988 0 : cond_resched();
989 :
990 0 : if (!find_get_entries(mapping, &index, end - 1, &fbatch,
991 : indices)) {
992 : /* If all gone or hole-punch or unfalloc, we're done */
993 0 : if (index == start || end != -1)
994 : break;
995 : /* But if truncating, restart to make sure all gone */
996 0 : index = start;
997 0 : continue;
998 : }
999 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
1000 0 : folio = fbatch.folios[i];
1001 :
1002 0 : if (xa_is_value(folio)) {
1003 0 : if (unfalloc)
1004 0 : continue;
1005 0 : if (shmem_free_swap(mapping, indices[i], folio)) {
1006 : /* Swap was replaced by page: retry */
1007 0 : index = indices[i];
1008 0 : break;
1009 : }
1010 0 : nr_swaps_freed++;
1011 0 : continue;
1012 : }
1013 :
1014 0 : folio_lock(folio);
1015 :
1016 0 : if (!unfalloc || !folio_test_uptodate(folio)) {
1017 0 : if (folio_mapping(folio) != mapping) {
1018 : /* Page was replaced by swap: retry */
1019 0 : folio_unlock(folio);
1020 0 : index = indices[i];
1021 0 : break;
1022 : }
1023 : VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1024 : folio);
1025 0 : truncate_inode_folio(mapping, folio);
1026 : }
1027 0 : folio_unlock(folio);
1028 : }
1029 0 : folio_batch_remove_exceptionals(&fbatch);
1030 : folio_batch_release(&fbatch);
1031 : }
1032 :
1033 0 : spin_lock_irq(&info->lock);
1034 0 : info->swapped -= nr_swaps_freed;
1035 0 : shmem_recalc_inode(inode);
1036 0 : spin_unlock_irq(&info->lock);
1037 0 : }
1038 :
1039 0 : void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1040 : {
1041 0 : shmem_undo_range(inode, lstart, lend, false);
1042 0 : inode->i_ctime = inode->i_mtime = current_time(inode);
1043 0 : inode_inc_iversion(inode);
1044 0 : }
1045 : EXPORT_SYMBOL_GPL(shmem_truncate_range);
1046 :
1047 0 : static int shmem_getattr(struct mnt_idmap *idmap,
1048 : const struct path *path, struct kstat *stat,
1049 : u32 request_mask, unsigned int query_flags)
1050 : {
1051 0 : struct inode *inode = path->dentry->d_inode;
1052 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1053 :
1054 0 : if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1055 0 : spin_lock_irq(&info->lock);
1056 0 : shmem_recalc_inode(inode);
1057 0 : spin_unlock_irq(&info->lock);
1058 : }
1059 0 : if (info->fsflags & FS_APPEND_FL)
1060 0 : stat->attributes |= STATX_ATTR_APPEND;
1061 0 : if (info->fsflags & FS_IMMUTABLE_FL)
1062 0 : stat->attributes |= STATX_ATTR_IMMUTABLE;
1063 0 : if (info->fsflags & FS_NODUMP_FL)
1064 0 : stat->attributes |= STATX_ATTR_NODUMP;
1065 0 : stat->attributes_mask |= (STATX_ATTR_APPEND |
1066 : STATX_ATTR_IMMUTABLE |
1067 : STATX_ATTR_NODUMP);
1068 0 : generic_fillattr(idmap, inode, stat);
1069 :
1070 0 : if (shmem_is_huge(inode, 0, false, NULL, 0))
1071 : stat->blksize = HPAGE_PMD_SIZE;
1072 :
1073 0 : if (request_mask & STATX_BTIME) {
1074 0 : stat->result_mask |= STATX_BTIME;
1075 0 : stat->btime.tv_sec = info->i_crtime.tv_sec;
1076 0 : stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1077 : }
1078 :
1079 0 : return 0;
1080 : }
1081 :
1082 0 : static int shmem_setattr(struct mnt_idmap *idmap,
1083 : struct dentry *dentry, struct iattr *attr)
1084 : {
1085 0 : struct inode *inode = d_inode(dentry);
1086 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1087 : int error;
1088 0 : bool update_mtime = false;
1089 0 : bool update_ctime = true;
1090 :
1091 0 : error = setattr_prepare(idmap, dentry, attr);
1092 0 : if (error)
1093 : return error;
1094 :
1095 0 : if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1096 0 : if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1097 : return -EPERM;
1098 : }
1099 : }
1100 :
1101 0 : if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1102 0 : loff_t oldsize = inode->i_size;
1103 0 : loff_t newsize = attr->ia_size;
1104 :
1105 : /* protected by i_rwsem */
1106 0 : if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1107 0 : (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1108 : return -EPERM;
1109 :
1110 0 : if (newsize != oldsize) {
1111 0 : error = shmem_reacct_size(SHMEM_I(inode)->flags,
1112 : oldsize, newsize);
1113 0 : if (error)
1114 : return error;
1115 0 : i_size_write(inode, newsize);
1116 0 : update_mtime = true;
1117 : } else {
1118 : update_ctime = false;
1119 : }
1120 0 : if (newsize <= oldsize) {
1121 0 : loff_t holebegin = round_up(newsize, PAGE_SIZE);
1122 0 : if (oldsize > holebegin)
1123 0 : unmap_mapping_range(inode->i_mapping,
1124 : holebegin, 0, 1);
1125 0 : if (info->alloced)
1126 0 : shmem_truncate_range(inode,
1127 : newsize, (loff_t)-1);
1128 : /* unmap again to remove racily COWed private pages */
1129 0 : if (oldsize > holebegin)
1130 0 : unmap_mapping_range(inode->i_mapping,
1131 : holebegin, 0, 1);
1132 : }
1133 : }
1134 :
1135 0 : setattr_copy(idmap, inode, attr);
1136 0 : if (attr->ia_valid & ATTR_MODE)
1137 0 : error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1138 0 : if (!error && update_ctime) {
1139 0 : inode->i_ctime = current_time(inode);
1140 0 : if (update_mtime)
1141 0 : inode->i_mtime = inode->i_ctime;
1142 : inode_inc_iversion(inode);
1143 : }
1144 : return error;
1145 : }
1146 :
1147 0 : static void shmem_evict_inode(struct inode *inode)
1148 : {
1149 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1150 0 : struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1151 :
1152 0 : if (shmem_mapping(inode->i_mapping)) {
1153 0 : shmem_unacct_size(info->flags, inode->i_size);
1154 0 : inode->i_size = 0;
1155 0 : mapping_set_exiting(inode->i_mapping);
1156 0 : shmem_truncate_range(inode, 0, (loff_t)-1);
1157 0 : if (!list_empty(&info->shrinklist)) {
1158 0 : spin_lock(&sbinfo->shrinklist_lock);
1159 0 : if (!list_empty(&info->shrinklist)) {
1160 0 : list_del_init(&info->shrinklist);
1161 0 : sbinfo->shrinklist_len--;
1162 : }
1163 0 : spin_unlock(&sbinfo->shrinklist_lock);
1164 : }
1165 0 : while (!list_empty(&info->swaplist)) {
1166 : /* Wait while shmem_unuse() is scanning this inode... */
1167 0 : wait_var_event(&info->stop_eviction,
1168 : !atomic_read(&info->stop_eviction));
1169 0 : mutex_lock(&shmem_swaplist_mutex);
1170 : /* ...but beware of the race if we peeked too early */
1171 0 : if (!atomic_read(&info->stop_eviction))
1172 0 : list_del_init(&info->swaplist);
1173 0 : mutex_unlock(&shmem_swaplist_mutex);
1174 : }
1175 : }
1176 :
1177 0 : simple_xattrs_free(&info->xattrs);
1178 0 : WARN_ON(inode->i_blocks);
1179 0 : shmem_free_inode(inode->i_sb);
1180 0 : clear_inode(inode);
1181 0 : }
1182 :
1183 0 : static int shmem_find_swap_entries(struct address_space *mapping,
1184 : pgoff_t start, struct folio_batch *fbatch,
1185 : pgoff_t *indices, unsigned int type)
1186 : {
1187 0 : XA_STATE(xas, &mapping->i_pages, start);
1188 : struct folio *folio;
1189 : swp_entry_t entry;
1190 :
1191 : rcu_read_lock();
1192 0 : xas_for_each(&xas, folio, ULONG_MAX) {
1193 0 : if (xas_retry(&xas, folio))
1194 0 : continue;
1195 :
1196 0 : if (!xa_is_value(folio))
1197 0 : continue;
1198 :
1199 0 : entry = radix_to_swp_entry(folio);
1200 : /*
1201 : * swapin error entries can be found in the mapping. But they're
1202 : * deliberately ignored here as we've done everything we can do.
1203 : */
1204 0 : if (swp_type(entry) != type)
1205 0 : continue;
1206 :
1207 0 : indices[folio_batch_count(fbatch)] = xas.xa_index;
1208 0 : if (!folio_batch_add(fbatch, folio))
1209 : break;
1210 :
1211 0 : if (need_resched()) {
1212 0 : xas_pause(&xas);
1213 : cond_resched_rcu();
1214 : }
1215 : }
1216 : rcu_read_unlock();
1217 :
1218 0 : return xas.xa_index;
1219 : }
1220 :
1221 : /*
1222 : * Move the swapped pages for an inode to page cache. Returns the count
1223 : * of pages swapped in, or the error in case of failure.
1224 : */
1225 0 : static int shmem_unuse_swap_entries(struct inode *inode,
1226 : struct folio_batch *fbatch, pgoff_t *indices)
1227 : {
1228 0 : int i = 0;
1229 0 : int ret = 0;
1230 0 : int error = 0;
1231 0 : struct address_space *mapping = inode->i_mapping;
1232 :
1233 0 : for (i = 0; i < folio_batch_count(fbatch); i++) {
1234 0 : struct folio *folio = fbatch->folios[i];
1235 :
1236 0 : if (!xa_is_value(folio))
1237 0 : continue;
1238 0 : error = shmem_swapin_folio(inode, indices[i],
1239 : &folio, SGP_CACHE,
1240 : mapping_gfp_mask(mapping),
1241 : NULL, NULL);
1242 0 : if (error == 0) {
1243 0 : folio_unlock(folio);
1244 0 : folio_put(folio);
1245 0 : ret++;
1246 : }
1247 0 : if (error == -ENOMEM)
1248 : break;
1249 0 : error = 0;
1250 : }
1251 0 : return error ? error : ret;
1252 : }
1253 :
1254 : /*
1255 : * If swap found in inode, free it and move page from swapcache to filecache.
1256 : */
1257 0 : static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1258 : {
1259 0 : struct address_space *mapping = inode->i_mapping;
1260 0 : pgoff_t start = 0;
1261 : struct folio_batch fbatch;
1262 : pgoff_t indices[PAGEVEC_SIZE];
1263 0 : int ret = 0;
1264 :
1265 : do {
1266 0 : folio_batch_init(&fbatch);
1267 0 : shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1268 0 : if (folio_batch_count(&fbatch) == 0) {
1269 : ret = 0;
1270 : break;
1271 : }
1272 :
1273 0 : ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1274 0 : if (ret < 0)
1275 : break;
1276 :
1277 0 : start = indices[folio_batch_count(&fbatch) - 1];
1278 : } while (true);
1279 :
1280 0 : return ret;
1281 : }
1282 :
1283 : /*
1284 : * Read all the shared memory data that resides in the swap
1285 : * device 'type' back into memory, so the swap device can be
1286 : * unused.
1287 : */
1288 0 : int shmem_unuse(unsigned int type)
1289 : {
1290 : struct shmem_inode_info *info, *next;
1291 0 : int error = 0;
1292 :
1293 0 : if (list_empty(&shmem_swaplist))
1294 : return 0;
1295 :
1296 0 : mutex_lock(&shmem_swaplist_mutex);
1297 0 : list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1298 0 : if (!info->swapped) {
1299 0 : list_del_init(&info->swaplist);
1300 0 : continue;
1301 : }
1302 : /*
1303 : * Drop the swaplist mutex while searching the inode for swap;
1304 : * but before doing so, make sure shmem_evict_inode() will not
1305 : * remove placeholder inode from swaplist, nor let it be freed
1306 : * (igrab() would protect from unlink, but not from unmount).
1307 : */
1308 0 : atomic_inc(&info->stop_eviction);
1309 0 : mutex_unlock(&shmem_swaplist_mutex);
1310 :
1311 0 : error = shmem_unuse_inode(&info->vfs_inode, type);
1312 0 : cond_resched();
1313 :
1314 0 : mutex_lock(&shmem_swaplist_mutex);
1315 0 : next = list_next_entry(info, swaplist);
1316 0 : if (!info->swapped)
1317 0 : list_del_init(&info->swaplist);
1318 0 : if (atomic_dec_and_test(&info->stop_eviction))
1319 0 : wake_up_var(&info->stop_eviction);
1320 0 : if (error)
1321 : break;
1322 : }
1323 0 : mutex_unlock(&shmem_swaplist_mutex);
1324 :
1325 0 : return error;
1326 : }
1327 :
1328 : /*
1329 : * Move the page from the page cache to the swap cache.
1330 : */
1331 0 : static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1332 : {
1333 0 : struct folio *folio = page_folio(page);
1334 : struct shmem_inode_info *info;
1335 : struct address_space *mapping;
1336 : struct inode *inode;
1337 : swp_entry_t swap;
1338 : pgoff_t index;
1339 :
1340 : /*
1341 : * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1342 : * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1343 : * and its shmem_writeback() needs them to be split when swapping.
1344 : */
1345 0 : if (folio_test_large(folio)) {
1346 : /* Ensure the subpages are still dirty */
1347 0 : folio_test_set_dirty(folio);
1348 0 : if (split_huge_page(page) < 0)
1349 : goto redirty;
1350 0 : folio = page_folio(page);
1351 : folio_clear_dirty(folio);
1352 : }
1353 :
1354 0 : BUG_ON(!folio_test_locked(folio));
1355 0 : mapping = folio->mapping;
1356 0 : index = folio->index;
1357 0 : inode = mapping->host;
1358 0 : info = SHMEM_I(inode);
1359 0 : if (info->flags & VM_LOCKED)
1360 : goto redirty;
1361 0 : if (!total_swap_pages)
1362 : goto redirty;
1363 :
1364 : /*
1365 : * Our capabilities prevent regular writeback or sync from ever calling
1366 : * shmem_writepage; but a stacking filesystem might use ->writepage of
1367 : * its underlying filesystem, in which case tmpfs should write out to
1368 : * swap only in response to memory pressure, and not for the writeback
1369 : * threads or sync.
1370 : */
1371 0 : if (!wbc->for_reclaim) {
1372 0 : WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1373 : goto redirty;
1374 : }
1375 :
1376 : /*
1377 : * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1378 : * value into swapfile.c, the only way we can correctly account for a
1379 : * fallocated folio arriving here is now to initialize it and write it.
1380 : *
1381 : * That's okay for a folio already fallocated earlier, but if we have
1382 : * not yet completed the fallocation, then (a) we want to keep track
1383 : * of this folio in case we have to undo it, and (b) it may not be a
1384 : * good idea to continue anyway, once we're pushing into swap. So
1385 : * reactivate the folio, and let shmem_fallocate() quit when too many.
1386 : */
1387 0 : if (!folio_test_uptodate(folio)) {
1388 0 : if (inode->i_private) {
1389 : struct shmem_falloc *shmem_falloc;
1390 0 : spin_lock(&inode->i_lock);
1391 0 : shmem_falloc = inode->i_private;
1392 0 : if (shmem_falloc &&
1393 0 : !shmem_falloc->waitq &&
1394 0 : index >= shmem_falloc->start &&
1395 0 : index < shmem_falloc->next)
1396 0 : shmem_falloc->nr_unswapped++;
1397 : else
1398 : shmem_falloc = NULL;
1399 0 : spin_unlock(&inode->i_lock);
1400 0 : if (shmem_falloc)
1401 : goto redirty;
1402 : }
1403 0 : folio_zero_range(folio, 0, folio_size(folio));
1404 0 : flush_dcache_folio(folio);
1405 : folio_mark_uptodate(folio);
1406 : }
1407 :
1408 0 : swap = folio_alloc_swap(folio);
1409 0 : if (!swap.val)
1410 : goto redirty;
1411 :
1412 : /*
1413 : * Add inode to shmem_unuse()'s list of swapped-out inodes,
1414 : * if it's not already there. Do it now before the folio is
1415 : * moved to swap cache, when its pagelock no longer protects
1416 : * the inode from eviction. But don't unlock the mutex until
1417 : * we've incremented swapped, because shmem_unuse_inode() will
1418 : * prune a !swapped inode from the swaplist under this mutex.
1419 : */
1420 0 : mutex_lock(&shmem_swaplist_mutex);
1421 0 : if (list_empty(&info->swaplist))
1422 0 : list_add(&info->swaplist, &shmem_swaplist);
1423 :
1424 0 : if (add_to_swap_cache(folio, swap,
1425 : __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1426 : NULL) == 0) {
1427 0 : spin_lock_irq(&info->lock);
1428 0 : shmem_recalc_inode(inode);
1429 0 : info->swapped++;
1430 0 : spin_unlock_irq(&info->lock);
1431 :
1432 0 : swap_shmem_alloc(swap);
1433 0 : shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1434 :
1435 0 : mutex_unlock(&shmem_swaplist_mutex);
1436 0 : BUG_ON(folio_mapped(folio));
1437 0 : swap_writepage(&folio->page, wbc);
1438 0 : return 0;
1439 : }
1440 :
1441 0 : mutex_unlock(&shmem_swaplist_mutex);
1442 0 : put_swap_folio(folio, swap);
1443 : redirty:
1444 0 : folio_mark_dirty(folio);
1445 0 : if (wbc->for_reclaim)
1446 : return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1447 0 : folio_unlock(folio);
1448 0 : return 0;
1449 : }
1450 :
1451 : #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1452 : static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1453 : {
1454 : char buffer[64];
1455 :
1456 : if (!mpol || mpol->mode == MPOL_DEFAULT)
1457 : return; /* show nothing */
1458 :
1459 : mpol_to_str(buffer, sizeof(buffer), mpol);
1460 :
1461 : seq_printf(seq, ",mpol=%s", buffer);
1462 : }
1463 :
1464 : static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1465 : {
1466 : struct mempolicy *mpol = NULL;
1467 : if (sbinfo->mpol) {
1468 : raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1469 : mpol = sbinfo->mpol;
1470 : mpol_get(mpol);
1471 : raw_spin_unlock(&sbinfo->stat_lock);
1472 : }
1473 : return mpol;
1474 : }
1475 : #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1476 : static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1477 : {
1478 : }
1479 : static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1480 : {
1481 : return NULL;
1482 : }
1483 : #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1484 : #ifndef CONFIG_NUMA
1485 : #define vm_policy vm_private_data
1486 : #endif
1487 :
1488 : static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1489 : struct shmem_inode_info *info, pgoff_t index)
1490 : {
1491 : /* Create a pseudo vma that just contains the policy */
1492 0 : vma_init(vma, NULL);
1493 : /* Bias interleave by inode number to distribute better across nodes */
1494 0 : vma->vm_pgoff = index + info->vfs_inode.i_ino;
1495 0 : vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1496 : }
1497 :
1498 : static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1499 : {
1500 : /* Drop reference taken by mpol_shared_policy_lookup() */
1501 0 : mpol_cond_put(vma->vm_policy);
1502 : }
1503 :
1504 0 : static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1505 : struct shmem_inode_info *info, pgoff_t index)
1506 : {
1507 : struct vm_area_struct pvma;
1508 : struct page *page;
1509 0 : struct vm_fault vmf = {
1510 : .vma = &pvma,
1511 : };
1512 :
1513 0 : shmem_pseudo_vma_init(&pvma, info, index);
1514 0 : page = swap_cluster_readahead(swap, gfp, &vmf);
1515 0 : shmem_pseudo_vma_destroy(&pvma);
1516 :
1517 0 : if (!page)
1518 : return NULL;
1519 0 : return page_folio(page);
1520 : }
1521 :
1522 : /*
1523 : * Make sure huge_gfp is always more limited than limit_gfp.
1524 : * Some of the flags set permissions, while others set limitations.
1525 : */
1526 : static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1527 : {
1528 : gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1529 : gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1530 : gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1531 : gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1532 :
1533 : /* Allow allocations only from the originally specified zones. */
1534 : result |= zoneflags;
1535 :
1536 : /*
1537 : * Minimize the result gfp by taking the union with the deny flags,
1538 : * and the intersection of the allow flags.
1539 : */
1540 : result |= (limit_gfp & denyflags);
1541 : result |= (huge_gfp & limit_gfp) & allowflags;
1542 :
1543 : return result;
1544 : }
1545 :
1546 : static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1547 : struct shmem_inode_info *info, pgoff_t index)
1548 : {
1549 : struct vm_area_struct pvma;
1550 : struct address_space *mapping = info->vfs_inode.i_mapping;
1551 : pgoff_t hindex;
1552 : struct folio *folio;
1553 :
1554 : hindex = round_down(index, HPAGE_PMD_NR);
1555 : if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1556 : XA_PRESENT))
1557 : return NULL;
1558 :
1559 : shmem_pseudo_vma_init(&pvma, info, hindex);
1560 : folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1561 : shmem_pseudo_vma_destroy(&pvma);
1562 : if (!folio)
1563 : count_vm_event(THP_FILE_FALLBACK);
1564 : return folio;
1565 : }
1566 :
1567 0 : static struct folio *shmem_alloc_folio(gfp_t gfp,
1568 : struct shmem_inode_info *info, pgoff_t index)
1569 : {
1570 : struct vm_area_struct pvma;
1571 : struct folio *folio;
1572 :
1573 0 : shmem_pseudo_vma_init(&pvma, info, index);
1574 0 : folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1575 0 : shmem_pseudo_vma_destroy(&pvma);
1576 :
1577 0 : return folio;
1578 : }
1579 :
1580 0 : static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1581 : pgoff_t index, bool huge)
1582 : {
1583 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1584 : struct folio *folio;
1585 : int nr;
1586 0 : int err = -ENOSPC;
1587 :
1588 : if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1589 0 : huge = false;
1590 0 : nr = huge ? HPAGE_PMD_NR : 1;
1591 :
1592 0 : if (!shmem_inode_acct_block(inode, nr))
1593 : goto failed;
1594 :
1595 : if (huge)
1596 : folio = shmem_alloc_hugefolio(gfp, info, index);
1597 : else
1598 0 : folio = shmem_alloc_folio(gfp, info, index);
1599 0 : if (folio) {
1600 0 : __folio_set_locked(folio);
1601 0 : __folio_set_swapbacked(folio);
1602 0 : return folio;
1603 : }
1604 :
1605 0 : err = -ENOMEM;
1606 0 : shmem_inode_unacct_blocks(inode, nr);
1607 : failed:
1608 0 : return ERR_PTR(err);
1609 : }
1610 :
1611 : /*
1612 : * When a page is moved from swapcache to shmem filecache (either by the
1613 : * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1614 : * shmem_unuse_inode()), it may have been read in earlier from swap, in
1615 : * ignorance of the mapping it belongs to. If that mapping has special
1616 : * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1617 : * we may need to copy to a suitable page before moving to filecache.
1618 : *
1619 : * In a future release, this may well be extended to respect cpuset and
1620 : * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1621 : * but for now it is a simple matter of zone.
1622 : */
1623 : static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1624 : {
1625 0 : return folio_zonenum(folio) > gfp_zone(gfp);
1626 : }
1627 :
1628 0 : static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1629 : struct shmem_inode_info *info, pgoff_t index)
1630 : {
1631 : struct folio *old, *new;
1632 : struct address_space *swap_mapping;
1633 : swp_entry_t entry;
1634 : pgoff_t swap_index;
1635 : int error;
1636 :
1637 0 : old = *foliop;
1638 0 : entry = folio_swap_entry(old);
1639 0 : swap_index = swp_offset(entry);
1640 0 : swap_mapping = swap_address_space(entry);
1641 :
1642 : /*
1643 : * We have arrived here because our zones are constrained, so don't
1644 : * limit chance of success by further cpuset and node constraints.
1645 : */
1646 0 : gfp &= ~GFP_CONSTRAINT_MASK;
1647 : VM_BUG_ON_FOLIO(folio_test_large(old), old);
1648 0 : new = shmem_alloc_folio(gfp, info, index);
1649 0 : if (!new)
1650 : return -ENOMEM;
1651 :
1652 0 : folio_get(new);
1653 0 : folio_copy(new, old);
1654 0 : flush_dcache_folio(new);
1655 :
1656 0 : __folio_set_locked(new);
1657 0 : __folio_set_swapbacked(new);
1658 0 : folio_mark_uptodate(new);
1659 0 : folio_set_swap_entry(new, entry);
1660 0 : folio_set_swapcache(new);
1661 :
1662 : /*
1663 : * Our caller will very soon move newpage out of swapcache, but it's
1664 : * a nice clean interface for us to replace oldpage by newpage there.
1665 : */
1666 0 : xa_lock_irq(&swap_mapping->i_pages);
1667 0 : error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1668 0 : if (!error) {
1669 0 : mem_cgroup_migrate(old, new);
1670 0 : __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1671 0 : __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1672 0 : __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1673 0 : __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1674 : }
1675 0 : xa_unlock_irq(&swap_mapping->i_pages);
1676 :
1677 0 : if (unlikely(error)) {
1678 : /*
1679 : * Is this possible? I think not, now that our callers check
1680 : * both PageSwapCache and page_private after getting page lock;
1681 : * but be defensive. Reverse old to newpage for clear and free.
1682 : */
1683 : old = new;
1684 : } else {
1685 0 : folio_add_lru(new);
1686 0 : *foliop = new;
1687 : }
1688 :
1689 0 : folio_clear_swapcache(old);
1690 0 : old->private = NULL;
1691 :
1692 0 : folio_unlock(old);
1693 : folio_put_refs(old, 2);
1694 : return error;
1695 : }
1696 :
1697 0 : static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1698 : struct folio *folio, swp_entry_t swap)
1699 : {
1700 0 : struct address_space *mapping = inode->i_mapping;
1701 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1702 : swp_entry_t swapin_error;
1703 : void *old;
1704 :
1705 : swapin_error = make_swapin_error_entry();
1706 0 : old = xa_cmpxchg_irq(&mapping->i_pages, index,
1707 : swp_to_radix_entry(swap),
1708 : swp_to_radix_entry(swapin_error), 0);
1709 0 : if (old != swp_to_radix_entry(swap))
1710 : return;
1711 :
1712 0 : folio_wait_writeback(folio);
1713 0 : delete_from_swap_cache(folio);
1714 0 : spin_lock_irq(&info->lock);
1715 : /*
1716 : * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1717 : * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1718 : * shmem_evict_inode.
1719 : */
1720 0 : info->alloced--;
1721 0 : info->swapped--;
1722 0 : shmem_recalc_inode(inode);
1723 0 : spin_unlock_irq(&info->lock);
1724 0 : swap_free(swap);
1725 : }
1726 :
1727 : /*
1728 : * Swap in the folio pointed to by *foliop.
1729 : * Caller has to make sure that *foliop contains a valid swapped folio.
1730 : * Returns 0 and the folio in foliop if success. On failure, returns the
1731 : * error code and NULL in *foliop.
1732 : */
1733 0 : static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1734 : struct folio **foliop, enum sgp_type sgp,
1735 : gfp_t gfp, struct vm_area_struct *vma,
1736 : vm_fault_t *fault_type)
1737 : {
1738 0 : struct address_space *mapping = inode->i_mapping;
1739 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1740 0 : struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1741 : struct swap_info_struct *si;
1742 0 : struct folio *folio = NULL;
1743 : swp_entry_t swap;
1744 : int error;
1745 :
1746 : VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1747 0 : swap = radix_to_swp_entry(*foliop);
1748 0 : *foliop = NULL;
1749 :
1750 0 : if (is_swapin_error_entry(swap))
1751 : return -EIO;
1752 :
1753 0 : si = get_swap_device(swap);
1754 0 : if (!si) {
1755 0 : if (!shmem_confirm_swap(mapping, index, swap))
1756 : return -EEXIST;
1757 : else
1758 : return -EINVAL;
1759 : }
1760 :
1761 : /* Look it up and read it in.. */
1762 0 : folio = swap_cache_get_folio(swap, NULL, 0);
1763 0 : if (!folio) {
1764 : /* Or update major stats only when swapin succeeds?? */
1765 0 : if (fault_type) {
1766 0 : *fault_type |= VM_FAULT_MAJOR;
1767 0 : count_vm_event(PGMAJFAULT);
1768 0 : count_memcg_event_mm(charge_mm, PGMAJFAULT);
1769 : }
1770 : /* Here we actually start the io */
1771 0 : folio = shmem_swapin(swap, gfp, info, index);
1772 0 : if (!folio) {
1773 : error = -ENOMEM;
1774 : goto failed;
1775 : }
1776 : }
1777 :
1778 : /* We have to do this with folio locked to prevent races */
1779 0 : folio_lock(folio);
1780 0 : if (!folio_test_swapcache(folio) ||
1781 0 : folio_swap_entry(folio).val != swap.val ||
1782 0 : !shmem_confirm_swap(mapping, index, swap)) {
1783 : error = -EEXIST;
1784 : goto unlock;
1785 : }
1786 0 : if (!folio_test_uptodate(folio)) {
1787 : error = -EIO;
1788 : goto failed;
1789 : }
1790 0 : folio_wait_writeback(folio);
1791 :
1792 : /*
1793 : * Some architectures may have to restore extra metadata to the
1794 : * folio after reading from swap.
1795 : */
1796 0 : arch_swap_restore(swap, folio);
1797 :
1798 0 : if (shmem_should_replace_folio(folio, gfp)) {
1799 0 : error = shmem_replace_folio(&folio, gfp, info, index);
1800 0 : if (error)
1801 : goto failed;
1802 : }
1803 :
1804 0 : error = shmem_add_to_page_cache(folio, mapping, index,
1805 : swp_to_radix_entry(swap), gfp,
1806 : charge_mm);
1807 0 : if (error)
1808 : goto failed;
1809 :
1810 0 : spin_lock_irq(&info->lock);
1811 0 : info->swapped--;
1812 0 : shmem_recalc_inode(inode);
1813 0 : spin_unlock_irq(&info->lock);
1814 :
1815 0 : if (sgp == SGP_WRITE)
1816 0 : folio_mark_accessed(folio);
1817 :
1818 0 : delete_from_swap_cache(folio);
1819 0 : folio_mark_dirty(folio);
1820 0 : swap_free(swap);
1821 0 : put_swap_device(si);
1822 :
1823 0 : *foliop = folio;
1824 : return 0;
1825 : failed:
1826 0 : if (!shmem_confirm_swap(mapping, index, swap))
1827 0 : error = -EEXIST;
1828 0 : if (error == -EIO)
1829 0 : shmem_set_folio_swapin_error(inode, index, folio, swap);
1830 : unlock:
1831 0 : if (folio) {
1832 0 : folio_unlock(folio);
1833 0 : folio_put(folio);
1834 : }
1835 0 : put_swap_device(si);
1836 :
1837 : return error;
1838 : }
1839 :
1840 : /*
1841 : * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1842 : *
1843 : * If we allocate a new one we do not mark it dirty. That's up to the
1844 : * vm. If we swap it in we mark it dirty since we also free the swap
1845 : * entry since a page cannot live in both the swap and page cache.
1846 : *
1847 : * vma, vmf, and fault_type are only supplied by shmem_fault:
1848 : * otherwise they are NULL.
1849 : */
1850 0 : static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1851 : struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1852 : struct vm_area_struct *vma, struct vm_fault *vmf,
1853 : vm_fault_t *fault_type)
1854 : {
1855 0 : struct address_space *mapping = inode->i_mapping;
1856 0 : struct shmem_inode_info *info = SHMEM_I(inode);
1857 : struct shmem_sb_info *sbinfo;
1858 : struct mm_struct *charge_mm;
1859 : struct folio *folio;
1860 : pgoff_t hindex;
1861 : gfp_t huge_gfp;
1862 : int error;
1863 0 : int once = 0;
1864 0 : int alloced = 0;
1865 :
1866 0 : if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1867 : return -EFBIG;
1868 : repeat:
1869 0 : if (sgp <= SGP_CACHE &&
1870 0 : ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1871 : return -EINVAL;
1872 : }
1873 :
1874 0 : sbinfo = SHMEM_SB(inode->i_sb);
1875 0 : charge_mm = vma ? vma->vm_mm : NULL;
1876 :
1877 0 : folio = __filemap_get_folio(mapping, index, FGP_ENTRY | FGP_LOCK, 0);
1878 : if (folio && vma && userfaultfd_minor(vma)) {
1879 : if (!xa_is_value(folio)) {
1880 : folio_unlock(folio);
1881 : folio_put(folio);
1882 : }
1883 : *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1884 : return 0;
1885 : }
1886 :
1887 0 : if (xa_is_value(folio)) {
1888 0 : error = shmem_swapin_folio(inode, index, &folio,
1889 : sgp, gfp, vma, fault_type);
1890 0 : if (error == -EEXIST)
1891 : goto repeat;
1892 :
1893 0 : *foliop = folio;
1894 : return error;
1895 : }
1896 :
1897 0 : if (folio) {
1898 0 : if (sgp == SGP_WRITE)
1899 0 : folio_mark_accessed(folio);
1900 0 : if (folio_test_uptodate(folio))
1901 : goto out;
1902 : /* fallocated folio */
1903 0 : if (sgp != SGP_READ)
1904 : goto clear;
1905 0 : folio_unlock(folio);
1906 0 : folio_put(folio);
1907 : }
1908 :
1909 : /*
1910 : * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1911 : * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1912 : */
1913 0 : *foliop = NULL;
1914 0 : if (sgp == SGP_READ)
1915 : return 0;
1916 0 : if (sgp == SGP_NOALLOC)
1917 : return -ENOENT;
1918 :
1919 : /*
1920 : * Fast cache lookup and swap lookup did not find it: allocate.
1921 : */
1922 :
1923 : if (vma && userfaultfd_missing(vma)) {
1924 : *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1925 : return 0;
1926 : }
1927 :
1928 : if (!shmem_is_huge(inode, index, false,
1929 : vma ? vma->vm_mm : NULL, vma ? vma->vm_flags : 0))
1930 : goto alloc_nohuge;
1931 :
1932 : huge_gfp = vma_thp_gfp_mask(vma);
1933 : huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1934 : folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
1935 : if (IS_ERR(folio)) {
1936 : alloc_nohuge:
1937 0 : folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
1938 : }
1939 0 : if (IS_ERR(folio)) {
1940 0 : int retry = 5;
1941 :
1942 0 : error = PTR_ERR(folio);
1943 0 : folio = NULL;
1944 : if (error != -ENOSPC)
1945 : goto unlock;
1946 : /*
1947 : * Try to reclaim some space by splitting a large folio
1948 : * beyond i_size on the filesystem.
1949 : */
1950 : while (retry--) {
1951 : int ret;
1952 :
1953 : ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1954 : if (ret == SHRINK_STOP)
1955 : break;
1956 : if (ret)
1957 : goto alloc_nohuge;
1958 : }
1959 : goto unlock;
1960 : }
1961 :
1962 0 : hindex = round_down(index, folio_nr_pages(folio));
1963 :
1964 0 : if (sgp == SGP_WRITE)
1965 0 : __folio_set_referenced(folio);
1966 :
1967 0 : error = shmem_add_to_page_cache(folio, mapping, hindex,
1968 : NULL, gfp & GFP_RECLAIM_MASK,
1969 : charge_mm);
1970 0 : if (error)
1971 : goto unacct;
1972 0 : folio_add_lru(folio);
1973 :
1974 0 : spin_lock_irq(&info->lock);
1975 0 : info->alloced += folio_nr_pages(folio);
1976 0 : inode->i_blocks += (blkcnt_t)BLOCKS_PER_PAGE << folio_order(folio);
1977 0 : shmem_recalc_inode(inode);
1978 0 : spin_unlock_irq(&info->lock);
1979 0 : alloced = true;
1980 :
1981 0 : if (folio_test_pmd_mappable(folio) &&
1982 : DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1983 : folio_next_index(folio) - 1) {
1984 : /*
1985 : * Part of the large folio is beyond i_size: subject
1986 : * to shrink under memory pressure.
1987 : */
1988 : spin_lock(&sbinfo->shrinklist_lock);
1989 : /*
1990 : * _careful to defend against unlocked access to
1991 : * ->shrink_list in shmem_unused_huge_shrink()
1992 : */
1993 : if (list_empty_careful(&info->shrinklist)) {
1994 : list_add_tail(&info->shrinklist,
1995 : &sbinfo->shrinklist);
1996 : sbinfo->shrinklist_len++;
1997 : }
1998 : spin_unlock(&sbinfo->shrinklist_lock);
1999 : }
2000 :
2001 : /*
2002 : * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2003 : */
2004 0 : if (sgp == SGP_FALLOC)
2005 0 : sgp = SGP_WRITE;
2006 : clear:
2007 : /*
2008 : * Let SGP_WRITE caller clear ends if write does not fill folio;
2009 : * but SGP_FALLOC on a folio fallocated earlier must initialize
2010 : * it now, lest undo on failure cancel our earlier guarantee.
2011 : */
2012 0 : if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2013 0 : long i, n = folio_nr_pages(folio);
2014 :
2015 0 : for (i = 0; i < n; i++)
2016 0 : clear_highpage(folio_page(folio, i));
2017 0 : flush_dcache_folio(folio);
2018 0 : folio_mark_uptodate(folio);
2019 : }
2020 :
2021 : /* Perhaps the file has been truncated since we checked */
2022 0 : if (sgp <= SGP_CACHE &&
2023 0 : ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2024 0 : if (alloced) {
2025 0 : folio_clear_dirty(folio);
2026 0 : filemap_remove_folio(folio);
2027 0 : spin_lock_irq(&info->lock);
2028 0 : shmem_recalc_inode(inode);
2029 0 : spin_unlock_irq(&info->lock);
2030 : }
2031 : error = -EINVAL;
2032 : goto unlock;
2033 : }
2034 : out:
2035 0 : *foliop = folio;
2036 : return 0;
2037 :
2038 : /*
2039 : * Error recovery.
2040 : */
2041 : unacct:
2042 0 : shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2043 :
2044 0 : if (folio_test_large(folio)) {
2045 0 : folio_unlock(folio);
2046 0 : folio_put(folio);
2047 : goto alloc_nohuge;
2048 : }
2049 : unlock:
2050 0 : if (folio) {
2051 0 : folio_unlock(folio);
2052 0 : folio_put(folio);
2053 : }
2054 0 : if (error == -ENOSPC && !once++) {
2055 0 : spin_lock_irq(&info->lock);
2056 0 : shmem_recalc_inode(inode);
2057 0 : spin_unlock_irq(&info->lock);
2058 : goto repeat;
2059 : }
2060 0 : if (error == -EEXIST)
2061 : goto repeat;
2062 : return error;
2063 : }
2064 :
2065 0 : int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2066 : enum sgp_type sgp)
2067 : {
2068 0 : return shmem_get_folio_gfp(inode, index, foliop, sgp,
2069 : mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2070 : }
2071 :
2072 : /*
2073 : * This is like autoremove_wake_function, but it removes the wait queue
2074 : * entry unconditionally - even if something else had already woken the
2075 : * target.
2076 : */
2077 0 : static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2078 : {
2079 0 : int ret = default_wake_function(wait, mode, sync, key);
2080 0 : list_del_init(&wait->entry);
2081 0 : return ret;
2082 : }
2083 :
2084 0 : static vm_fault_t shmem_fault(struct vm_fault *vmf)
2085 : {
2086 0 : struct vm_area_struct *vma = vmf->vma;
2087 0 : struct inode *inode = file_inode(vma->vm_file);
2088 0 : gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2089 0 : struct folio *folio = NULL;
2090 : int err;
2091 0 : vm_fault_t ret = VM_FAULT_LOCKED;
2092 :
2093 : /*
2094 : * Trinity finds that probing a hole which tmpfs is punching can
2095 : * prevent the hole-punch from ever completing: which in turn
2096 : * locks writers out with its hold on i_rwsem. So refrain from
2097 : * faulting pages into the hole while it's being punched. Although
2098 : * shmem_undo_range() does remove the additions, it may be unable to
2099 : * keep up, as each new page needs its own unmap_mapping_range() call,
2100 : * and the i_mmap tree grows ever slower to scan if new vmas are added.
2101 : *
2102 : * It does not matter if we sometimes reach this check just before the
2103 : * hole-punch begins, so that one fault then races with the punch:
2104 : * we just need to make racing faults a rare case.
2105 : *
2106 : * The implementation below would be much simpler if we just used a
2107 : * standard mutex or completion: but we cannot take i_rwsem in fault,
2108 : * and bloating every shmem inode for this unlikely case would be sad.
2109 : */
2110 0 : if (unlikely(inode->i_private)) {
2111 : struct shmem_falloc *shmem_falloc;
2112 :
2113 0 : spin_lock(&inode->i_lock);
2114 0 : shmem_falloc = inode->i_private;
2115 0 : if (shmem_falloc &&
2116 0 : shmem_falloc->waitq &&
2117 0 : vmf->pgoff >= shmem_falloc->start &&
2118 0 : vmf->pgoff < shmem_falloc->next) {
2119 : struct file *fpin;
2120 : wait_queue_head_t *shmem_falloc_waitq;
2121 0 : DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2122 :
2123 0 : ret = VM_FAULT_NOPAGE;
2124 0 : fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2125 0 : if (fpin)
2126 0 : ret = VM_FAULT_RETRY;
2127 :
2128 0 : shmem_falloc_waitq = shmem_falloc->waitq;
2129 0 : prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2130 : TASK_UNINTERRUPTIBLE);
2131 0 : spin_unlock(&inode->i_lock);
2132 0 : schedule();
2133 :
2134 : /*
2135 : * shmem_falloc_waitq points into the shmem_fallocate()
2136 : * stack of the hole-punching task: shmem_falloc_waitq
2137 : * is usually invalid by the time we reach here, but
2138 : * finish_wait() does not dereference it in that case;
2139 : * though i_lock needed lest racing with wake_up_all().
2140 : */
2141 0 : spin_lock(&inode->i_lock);
2142 0 : finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2143 0 : spin_unlock(&inode->i_lock);
2144 :
2145 0 : if (fpin)
2146 0 : fput(fpin);
2147 0 : return ret;
2148 : }
2149 0 : spin_unlock(&inode->i_lock);
2150 : }
2151 :
2152 0 : err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2153 : gfp, vma, vmf, &ret);
2154 0 : if (err)
2155 : return vmf_error(err);
2156 0 : if (folio)
2157 0 : vmf->page = folio_file_page(folio, vmf->pgoff);
2158 0 : return ret;
2159 : }
2160 :
2161 0 : unsigned long shmem_get_unmapped_area(struct file *file,
2162 : unsigned long uaddr, unsigned long len,
2163 : unsigned long pgoff, unsigned long flags)
2164 : {
2165 : unsigned long (*get_area)(struct file *,
2166 : unsigned long, unsigned long, unsigned long, unsigned long);
2167 : unsigned long addr;
2168 : unsigned long offset;
2169 : unsigned long inflated_len;
2170 : unsigned long inflated_addr;
2171 : unsigned long inflated_offset;
2172 :
2173 0 : if (len > TASK_SIZE)
2174 : return -ENOMEM;
2175 :
2176 0 : get_area = current->mm->get_unmapped_area;
2177 0 : addr = get_area(file, uaddr, len, pgoff, flags);
2178 :
2179 : if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2180 0 : return addr;
2181 : if (IS_ERR_VALUE(addr))
2182 : return addr;
2183 : if (addr & ~PAGE_MASK)
2184 : return addr;
2185 : if (addr > TASK_SIZE - len)
2186 : return addr;
2187 :
2188 : if (shmem_huge == SHMEM_HUGE_DENY)
2189 : return addr;
2190 : if (len < HPAGE_PMD_SIZE)
2191 : return addr;
2192 : if (flags & MAP_FIXED)
2193 : return addr;
2194 : /*
2195 : * Our priority is to support MAP_SHARED mapped hugely;
2196 : * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2197 : * But if caller specified an address hint and we allocated area there
2198 : * successfully, respect that as before.
2199 : */
2200 : if (uaddr == addr)
2201 : return addr;
2202 :
2203 : if (shmem_huge != SHMEM_HUGE_FORCE) {
2204 : struct super_block *sb;
2205 :
2206 : if (file) {
2207 : VM_BUG_ON(file->f_op != &shmem_file_operations);
2208 : sb = file_inode(file)->i_sb;
2209 : } else {
2210 : /*
2211 : * Called directly from mm/mmap.c, or drivers/char/mem.c
2212 : * for "/dev/zero", to create a shared anonymous object.
2213 : */
2214 : if (IS_ERR(shm_mnt))
2215 : return addr;
2216 : sb = shm_mnt->mnt_sb;
2217 : }
2218 : if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2219 : return addr;
2220 : }
2221 :
2222 : offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2223 : if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2224 : return addr;
2225 : if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2226 : return addr;
2227 :
2228 : inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2229 : if (inflated_len > TASK_SIZE)
2230 : return addr;
2231 : if (inflated_len < len)
2232 : return addr;
2233 :
2234 : inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2235 : if (IS_ERR_VALUE(inflated_addr))
2236 : return addr;
2237 : if (inflated_addr & ~PAGE_MASK)
2238 : return addr;
2239 :
2240 : inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2241 : inflated_addr += offset - inflated_offset;
2242 : if (inflated_offset > offset)
2243 : inflated_addr += HPAGE_PMD_SIZE;
2244 :
2245 : if (inflated_addr > TASK_SIZE - len)
2246 : return addr;
2247 : return inflated_addr;
2248 : }
2249 :
2250 : #ifdef CONFIG_NUMA
2251 : static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2252 : {
2253 : struct inode *inode = file_inode(vma->vm_file);
2254 : return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2255 : }
2256 :
2257 : static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2258 : unsigned long addr)
2259 : {
2260 : struct inode *inode = file_inode(vma->vm_file);
2261 : pgoff_t index;
2262 :
2263 : index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2264 : return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2265 : }
2266 : #endif
2267 :
2268 0 : int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2269 : {
2270 0 : struct inode *inode = file_inode(file);
2271 0 : struct shmem_inode_info *info = SHMEM_I(inode);
2272 0 : int retval = -ENOMEM;
2273 :
2274 : /*
2275 : * What serializes the accesses to info->flags?
2276 : * ipc_lock_object() when called from shmctl_do_lock(),
2277 : * no serialization needed when called from shm_destroy().
2278 : */
2279 0 : if (lock && !(info->flags & VM_LOCKED)) {
2280 0 : if (!user_shm_lock(inode->i_size, ucounts))
2281 : goto out_nomem;
2282 0 : info->flags |= VM_LOCKED;
2283 0 : mapping_set_unevictable(file->f_mapping);
2284 : }
2285 0 : if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2286 0 : user_shm_unlock(inode->i_size, ucounts);
2287 0 : info->flags &= ~VM_LOCKED;
2288 0 : mapping_clear_unevictable(file->f_mapping);
2289 : }
2290 : retval = 0;
2291 :
2292 : out_nomem:
2293 0 : return retval;
2294 : }
2295 :
2296 0 : static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2297 : {
2298 0 : struct inode *inode = file_inode(file);
2299 0 : struct shmem_inode_info *info = SHMEM_I(inode);
2300 : int ret;
2301 :
2302 0 : ret = seal_check_future_write(info->seals, vma);
2303 0 : if (ret)
2304 : return ret;
2305 :
2306 : /* arm64 - allow memory tagging on RAM-based files */
2307 0 : vm_flags_set(vma, VM_MTE_ALLOWED);
2308 :
2309 0 : file_accessed(file);
2310 : /* This is anonymous shared memory if it is unlinked at the time of mmap */
2311 0 : if (inode->i_nlink)
2312 0 : vma->vm_ops = &shmem_vm_ops;
2313 : else
2314 0 : vma->vm_ops = &shmem_anon_vm_ops;
2315 : return 0;
2316 : }
2317 :
2318 : #ifdef CONFIG_TMPFS_XATTR
2319 : static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2320 :
2321 : /*
2322 : * chattr's fsflags are unrelated to extended attributes,
2323 : * but tmpfs has chosen to enable them under the same config option.
2324 : */
2325 : static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2326 : {
2327 : unsigned int i_flags = 0;
2328 :
2329 : if (fsflags & FS_NOATIME_FL)
2330 : i_flags |= S_NOATIME;
2331 : if (fsflags & FS_APPEND_FL)
2332 : i_flags |= S_APPEND;
2333 : if (fsflags & FS_IMMUTABLE_FL)
2334 : i_flags |= S_IMMUTABLE;
2335 : /*
2336 : * But FS_NODUMP_FL does not require any action in i_flags.
2337 : */
2338 : inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2339 : }
2340 : #else
2341 : static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2342 : {
2343 : }
2344 : #define shmem_initxattrs NULL
2345 : #endif
2346 :
2347 1 : static struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb,
2348 : struct inode *dir, umode_t mode, dev_t dev,
2349 : unsigned long flags)
2350 : {
2351 : struct inode *inode;
2352 : struct shmem_inode_info *info;
2353 1 : struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2354 : ino_t ino;
2355 :
2356 1 : if (shmem_reserve_inode(sb, &ino))
2357 : return NULL;
2358 :
2359 1 : inode = new_inode(sb);
2360 1 : if (inode) {
2361 1 : inode->i_ino = ino;
2362 1 : inode_init_owner(idmap, inode, dir, mode);
2363 1 : inode->i_blocks = 0;
2364 1 : inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2365 1 : inode->i_generation = get_random_u32();
2366 1 : info = SHMEM_I(inode);
2367 1 : memset(info, 0, (char *)inode - (char *)info);
2368 1 : spin_lock_init(&info->lock);
2369 2 : atomic_set(&info->stop_eviction, 0);
2370 1 : info->seals = F_SEAL_SEAL;
2371 1 : info->flags = flags & VM_NORESERVE;
2372 1 : info->i_crtime = inode->i_mtime;
2373 1 : info->fsflags = (dir == NULL) ? 0 :
2374 0 : SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2375 : if (info->fsflags)
2376 : shmem_set_inode_flags(inode, info->fsflags);
2377 2 : INIT_LIST_HEAD(&info->shrinklist);
2378 2 : INIT_LIST_HEAD(&info->swaplist);
2379 1 : simple_xattrs_init(&info->xattrs);
2380 1 : cache_no_acl(inode);
2381 2 : mapping_set_large_folios(inode->i_mapping);
2382 :
2383 1 : switch (mode & S_IFMT) {
2384 : default:
2385 0 : inode->i_op = &shmem_special_inode_operations;
2386 0 : init_special_inode(inode, mode, dev);
2387 0 : break;
2388 : case S_IFREG:
2389 0 : inode->i_mapping->a_ops = &shmem_aops;
2390 0 : inode->i_op = &shmem_inode_operations;
2391 0 : inode->i_fop = &shmem_file_operations;
2392 0 : mpol_shared_policy_init(&info->policy,
2393 : shmem_get_sbmpol(sbinfo));
2394 : break;
2395 : case S_IFDIR:
2396 1 : inc_nlink(inode);
2397 : /* Some things misbehave if size == 0 on a directory */
2398 1 : inode->i_size = 2 * BOGO_DIRENT_SIZE;
2399 1 : inode->i_op = &shmem_dir_inode_operations;
2400 1 : inode->i_fop = &simple_dir_operations;
2401 1 : break;
2402 : case S_IFLNK:
2403 : /*
2404 : * Must not load anything in the rbtree,
2405 : * mpol_free_shared_policy will not be called.
2406 : */
2407 : mpol_shared_policy_init(&info->policy, NULL);
2408 : break;
2409 : }
2410 :
2411 : lockdep_annotate_inode_mutex_key(inode);
2412 : } else
2413 : shmem_free_inode(sb);
2414 : return inode;
2415 : }
2416 :
2417 : #ifdef CONFIG_USERFAULTFD
2418 : int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2419 : pmd_t *dst_pmd,
2420 : struct vm_area_struct *dst_vma,
2421 : unsigned long dst_addr,
2422 : unsigned long src_addr,
2423 : bool zeropage, bool wp_copy,
2424 : struct page **pagep)
2425 : {
2426 : struct inode *inode = file_inode(dst_vma->vm_file);
2427 : struct shmem_inode_info *info = SHMEM_I(inode);
2428 : struct address_space *mapping = inode->i_mapping;
2429 : gfp_t gfp = mapping_gfp_mask(mapping);
2430 : pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2431 : void *page_kaddr;
2432 : struct folio *folio;
2433 : int ret;
2434 : pgoff_t max_off;
2435 :
2436 : if (!shmem_inode_acct_block(inode, 1)) {
2437 : /*
2438 : * We may have got a page, returned -ENOENT triggering a retry,
2439 : * and now we find ourselves with -ENOMEM. Release the page, to
2440 : * avoid a BUG_ON in our caller.
2441 : */
2442 : if (unlikely(*pagep)) {
2443 : put_page(*pagep);
2444 : *pagep = NULL;
2445 : }
2446 : return -ENOMEM;
2447 : }
2448 :
2449 : if (!*pagep) {
2450 : ret = -ENOMEM;
2451 : folio = shmem_alloc_folio(gfp, info, pgoff);
2452 : if (!folio)
2453 : goto out_unacct_blocks;
2454 :
2455 : if (!zeropage) { /* COPY */
2456 : page_kaddr = kmap_local_folio(folio, 0);
2457 : /*
2458 : * The read mmap_lock is held here. Despite the
2459 : * mmap_lock being read recursive a deadlock is still
2460 : * possible if a writer has taken a lock. For example:
2461 : *
2462 : * process A thread 1 takes read lock on own mmap_lock
2463 : * process A thread 2 calls mmap, blocks taking write lock
2464 : * process B thread 1 takes page fault, read lock on own mmap lock
2465 : * process B thread 2 calls mmap, blocks taking write lock
2466 : * process A thread 1 blocks taking read lock on process B
2467 : * process B thread 1 blocks taking read lock on process A
2468 : *
2469 : * Disable page faults to prevent potential deadlock
2470 : * and retry the copy outside the mmap_lock.
2471 : */
2472 : pagefault_disable();
2473 : ret = copy_from_user(page_kaddr,
2474 : (const void __user *)src_addr,
2475 : PAGE_SIZE);
2476 : pagefault_enable();
2477 : kunmap_local(page_kaddr);
2478 :
2479 : /* fallback to copy_from_user outside mmap_lock */
2480 : if (unlikely(ret)) {
2481 : *pagep = &folio->page;
2482 : ret = -ENOENT;
2483 : /* don't free the page */
2484 : goto out_unacct_blocks;
2485 : }
2486 :
2487 : flush_dcache_folio(folio);
2488 : } else { /* ZEROPAGE */
2489 : clear_user_highpage(&folio->page, dst_addr);
2490 : }
2491 : } else {
2492 : folio = page_folio(*pagep);
2493 : VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2494 : *pagep = NULL;
2495 : }
2496 :
2497 : VM_BUG_ON(folio_test_locked(folio));
2498 : VM_BUG_ON(folio_test_swapbacked(folio));
2499 : __folio_set_locked(folio);
2500 : __folio_set_swapbacked(folio);
2501 : __folio_mark_uptodate(folio);
2502 :
2503 : ret = -EFAULT;
2504 : max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2505 : if (unlikely(pgoff >= max_off))
2506 : goto out_release;
2507 :
2508 : ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2509 : gfp & GFP_RECLAIM_MASK, dst_mm);
2510 : if (ret)
2511 : goto out_release;
2512 :
2513 : ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2514 : &folio->page, true, wp_copy);
2515 : if (ret)
2516 : goto out_delete_from_cache;
2517 :
2518 : spin_lock_irq(&info->lock);
2519 : info->alloced++;
2520 : inode->i_blocks += BLOCKS_PER_PAGE;
2521 : shmem_recalc_inode(inode);
2522 : spin_unlock_irq(&info->lock);
2523 :
2524 : folio_unlock(folio);
2525 : return 0;
2526 : out_delete_from_cache:
2527 : filemap_remove_folio(folio);
2528 : out_release:
2529 : folio_unlock(folio);
2530 : folio_put(folio);
2531 : out_unacct_blocks:
2532 : shmem_inode_unacct_blocks(inode, 1);
2533 : return ret;
2534 : }
2535 : #endif /* CONFIG_USERFAULTFD */
2536 :
2537 : #ifdef CONFIG_TMPFS
2538 : static const struct inode_operations shmem_symlink_inode_operations;
2539 : static const struct inode_operations shmem_short_symlink_operations;
2540 :
2541 : static int
2542 : shmem_write_begin(struct file *file, struct address_space *mapping,
2543 : loff_t pos, unsigned len,
2544 : struct page **pagep, void **fsdata)
2545 : {
2546 : struct inode *inode = mapping->host;
2547 : struct shmem_inode_info *info = SHMEM_I(inode);
2548 : pgoff_t index = pos >> PAGE_SHIFT;
2549 : struct folio *folio;
2550 : int ret = 0;
2551 :
2552 : /* i_rwsem is held by caller */
2553 : if (unlikely(info->seals & (F_SEAL_GROW |
2554 : F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2555 : if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2556 : return -EPERM;
2557 : if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2558 : return -EPERM;
2559 : }
2560 :
2561 : ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2562 :
2563 : if (ret)
2564 : return ret;
2565 :
2566 : *pagep = folio_file_page(folio, index);
2567 : if (PageHWPoison(*pagep)) {
2568 : folio_unlock(folio);
2569 : folio_put(folio);
2570 : *pagep = NULL;
2571 : return -EIO;
2572 : }
2573 :
2574 : return 0;
2575 : }
2576 :
2577 : static int
2578 : shmem_write_end(struct file *file, struct address_space *mapping,
2579 : loff_t pos, unsigned len, unsigned copied,
2580 : struct page *page, void *fsdata)
2581 : {
2582 : struct folio *folio = page_folio(page);
2583 : struct inode *inode = mapping->host;
2584 :
2585 : if (pos + copied > inode->i_size)
2586 : i_size_write(inode, pos + copied);
2587 :
2588 : if (!folio_test_uptodate(folio)) {
2589 : if (copied < folio_size(folio)) {
2590 : size_t from = offset_in_folio(folio, pos);
2591 : folio_zero_segments(folio, 0, from,
2592 : from + copied, folio_size(folio));
2593 : }
2594 : folio_mark_uptodate(folio);
2595 : }
2596 : folio_mark_dirty(folio);
2597 : folio_unlock(folio);
2598 : folio_put(folio);
2599 :
2600 : return copied;
2601 : }
2602 :
2603 : static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2604 : {
2605 : struct file *file = iocb->ki_filp;
2606 : struct inode *inode = file_inode(file);
2607 : struct address_space *mapping = inode->i_mapping;
2608 : pgoff_t index;
2609 : unsigned long offset;
2610 : int error = 0;
2611 : ssize_t retval = 0;
2612 : loff_t *ppos = &iocb->ki_pos;
2613 :
2614 : index = *ppos >> PAGE_SHIFT;
2615 : offset = *ppos & ~PAGE_MASK;
2616 :
2617 : for (;;) {
2618 : struct folio *folio = NULL;
2619 : struct page *page = NULL;
2620 : pgoff_t end_index;
2621 : unsigned long nr, ret;
2622 : loff_t i_size = i_size_read(inode);
2623 :
2624 : end_index = i_size >> PAGE_SHIFT;
2625 : if (index > end_index)
2626 : break;
2627 : if (index == end_index) {
2628 : nr = i_size & ~PAGE_MASK;
2629 : if (nr <= offset)
2630 : break;
2631 : }
2632 :
2633 : error = shmem_get_folio(inode, index, &folio, SGP_READ);
2634 : if (error) {
2635 : if (error == -EINVAL)
2636 : error = 0;
2637 : break;
2638 : }
2639 : if (folio) {
2640 : folio_unlock(folio);
2641 :
2642 : page = folio_file_page(folio, index);
2643 : if (PageHWPoison(page)) {
2644 : folio_put(folio);
2645 : error = -EIO;
2646 : break;
2647 : }
2648 : }
2649 :
2650 : /*
2651 : * We must evaluate after, since reads (unlike writes)
2652 : * are called without i_rwsem protection against truncate
2653 : */
2654 : nr = PAGE_SIZE;
2655 : i_size = i_size_read(inode);
2656 : end_index = i_size >> PAGE_SHIFT;
2657 : if (index == end_index) {
2658 : nr = i_size & ~PAGE_MASK;
2659 : if (nr <= offset) {
2660 : if (folio)
2661 : folio_put(folio);
2662 : break;
2663 : }
2664 : }
2665 : nr -= offset;
2666 :
2667 : if (folio) {
2668 : /*
2669 : * If users can be writing to this page using arbitrary
2670 : * virtual addresses, take care about potential aliasing
2671 : * before reading the page on the kernel side.
2672 : */
2673 : if (mapping_writably_mapped(mapping))
2674 : flush_dcache_page(page);
2675 : /*
2676 : * Mark the page accessed if we read the beginning.
2677 : */
2678 : if (!offset)
2679 : folio_mark_accessed(folio);
2680 : /*
2681 : * Ok, we have the page, and it's up-to-date, so
2682 : * now we can copy it to user space...
2683 : */
2684 : ret = copy_page_to_iter(page, offset, nr, to);
2685 : folio_put(folio);
2686 :
2687 : } else if (user_backed_iter(to)) {
2688 : /*
2689 : * Copy to user tends to be so well optimized, but
2690 : * clear_user() not so much, that it is noticeably
2691 : * faster to copy the zero page instead of clearing.
2692 : */
2693 : ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2694 : } else {
2695 : /*
2696 : * But submitting the same page twice in a row to
2697 : * splice() - or others? - can result in confusion:
2698 : * so don't attempt that optimization on pipes etc.
2699 : */
2700 : ret = iov_iter_zero(nr, to);
2701 : }
2702 :
2703 : retval += ret;
2704 : offset += ret;
2705 : index += offset >> PAGE_SHIFT;
2706 : offset &= ~PAGE_MASK;
2707 :
2708 : if (!iov_iter_count(to))
2709 : break;
2710 : if (ret < nr) {
2711 : error = -EFAULT;
2712 : break;
2713 : }
2714 : cond_resched();
2715 : }
2716 :
2717 : *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2718 : file_accessed(file);
2719 : return retval ? retval : error;
2720 : }
2721 :
2722 : static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2723 : {
2724 : struct address_space *mapping = file->f_mapping;
2725 : struct inode *inode = mapping->host;
2726 :
2727 : if (whence != SEEK_DATA && whence != SEEK_HOLE)
2728 : return generic_file_llseek_size(file, offset, whence,
2729 : MAX_LFS_FILESIZE, i_size_read(inode));
2730 : if (offset < 0)
2731 : return -ENXIO;
2732 :
2733 : inode_lock(inode);
2734 : /* We're holding i_rwsem so we can access i_size directly */
2735 : offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2736 : if (offset >= 0)
2737 : offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2738 : inode_unlock(inode);
2739 : return offset;
2740 : }
2741 :
2742 : static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2743 : loff_t len)
2744 : {
2745 : struct inode *inode = file_inode(file);
2746 : struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2747 : struct shmem_inode_info *info = SHMEM_I(inode);
2748 : struct shmem_falloc shmem_falloc;
2749 : pgoff_t start, index, end, undo_fallocend;
2750 : int error;
2751 :
2752 : if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2753 : return -EOPNOTSUPP;
2754 :
2755 : inode_lock(inode);
2756 :
2757 : if (mode & FALLOC_FL_PUNCH_HOLE) {
2758 : struct address_space *mapping = file->f_mapping;
2759 : loff_t unmap_start = round_up(offset, PAGE_SIZE);
2760 : loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2761 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2762 :
2763 : /* protected by i_rwsem */
2764 : if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2765 : error = -EPERM;
2766 : goto out;
2767 : }
2768 :
2769 : shmem_falloc.waitq = &shmem_falloc_waitq;
2770 : shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2771 : shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2772 : spin_lock(&inode->i_lock);
2773 : inode->i_private = &shmem_falloc;
2774 : spin_unlock(&inode->i_lock);
2775 :
2776 : if ((u64)unmap_end > (u64)unmap_start)
2777 : unmap_mapping_range(mapping, unmap_start,
2778 : 1 + unmap_end - unmap_start, 0);
2779 : shmem_truncate_range(inode, offset, offset + len - 1);
2780 : /* No need to unmap again: hole-punching leaves COWed pages */
2781 :
2782 : spin_lock(&inode->i_lock);
2783 : inode->i_private = NULL;
2784 : wake_up_all(&shmem_falloc_waitq);
2785 : WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2786 : spin_unlock(&inode->i_lock);
2787 : error = 0;
2788 : goto out;
2789 : }
2790 :
2791 : /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2792 : error = inode_newsize_ok(inode, offset + len);
2793 : if (error)
2794 : goto out;
2795 :
2796 : if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2797 : error = -EPERM;
2798 : goto out;
2799 : }
2800 :
2801 : start = offset >> PAGE_SHIFT;
2802 : end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2803 : /* Try to avoid a swapstorm if len is impossible to satisfy */
2804 : if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2805 : error = -ENOSPC;
2806 : goto out;
2807 : }
2808 :
2809 : shmem_falloc.waitq = NULL;
2810 : shmem_falloc.start = start;
2811 : shmem_falloc.next = start;
2812 : shmem_falloc.nr_falloced = 0;
2813 : shmem_falloc.nr_unswapped = 0;
2814 : spin_lock(&inode->i_lock);
2815 : inode->i_private = &shmem_falloc;
2816 : spin_unlock(&inode->i_lock);
2817 :
2818 : /*
2819 : * info->fallocend is only relevant when huge pages might be
2820 : * involved: to prevent split_huge_page() freeing fallocated
2821 : * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2822 : */
2823 : undo_fallocend = info->fallocend;
2824 : if (info->fallocend < end)
2825 : info->fallocend = end;
2826 :
2827 : for (index = start; index < end; ) {
2828 : struct folio *folio;
2829 :
2830 : /*
2831 : * Good, the fallocate(2) manpage permits EINTR: we may have
2832 : * been interrupted because we are using up too much memory.
2833 : */
2834 : if (signal_pending(current))
2835 : error = -EINTR;
2836 : else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2837 : error = -ENOMEM;
2838 : else
2839 : error = shmem_get_folio(inode, index, &folio,
2840 : SGP_FALLOC);
2841 : if (error) {
2842 : info->fallocend = undo_fallocend;
2843 : /* Remove the !uptodate folios we added */
2844 : if (index > start) {
2845 : shmem_undo_range(inode,
2846 : (loff_t)start << PAGE_SHIFT,
2847 : ((loff_t)index << PAGE_SHIFT) - 1, true);
2848 : }
2849 : goto undone;
2850 : }
2851 :
2852 : /*
2853 : * Here is a more important optimization than it appears:
2854 : * a second SGP_FALLOC on the same large folio will clear it,
2855 : * making it uptodate and un-undoable if we fail later.
2856 : */
2857 : index = folio_next_index(folio);
2858 : /* Beware 32-bit wraparound */
2859 : if (!index)
2860 : index--;
2861 :
2862 : /*
2863 : * Inform shmem_writepage() how far we have reached.
2864 : * No need for lock or barrier: we have the page lock.
2865 : */
2866 : if (!folio_test_uptodate(folio))
2867 : shmem_falloc.nr_falloced += index - shmem_falloc.next;
2868 : shmem_falloc.next = index;
2869 :
2870 : /*
2871 : * If !uptodate, leave it that way so that freeable folios
2872 : * can be recognized if we need to rollback on error later.
2873 : * But mark it dirty so that memory pressure will swap rather
2874 : * than free the folios we are allocating (and SGP_CACHE folios
2875 : * might still be clean: we now need to mark those dirty too).
2876 : */
2877 : folio_mark_dirty(folio);
2878 : folio_unlock(folio);
2879 : folio_put(folio);
2880 : cond_resched();
2881 : }
2882 :
2883 : if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2884 : i_size_write(inode, offset + len);
2885 : undone:
2886 : spin_lock(&inode->i_lock);
2887 : inode->i_private = NULL;
2888 : spin_unlock(&inode->i_lock);
2889 : out:
2890 : if (!error)
2891 : file_modified(file);
2892 : inode_unlock(inode);
2893 : return error;
2894 : }
2895 :
2896 : static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2897 : {
2898 : struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2899 :
2900 : buf->f_type = TMPFS_MAGIC;
2901 : buf->f_bsize = PAGE_SIZE;
2902 : buf->f_namelen = NAME_MAX;
2903 : if (sbinfo->max_blocks) {
2904 : buf->f_blocks = sbinfo->max_blocks;
2905 : buf->f_bavail =
2906 : buf->f_bfree = sbinfo->max_blocks -
2907 : percpu_counter_sum(&sbinfo->used_blocks);
2908 : }
2909 : if (sbinfo->max_inodes) {
2910 : buf->f_files = sbinfo->max_inodes;
2911 : buf->f_ffree = sbinfo->free_inodes;
2912 : }
2913 : /* else leave those fields 0 like simple_statfs */
2914 :
2915 : buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2916 :
2917 : return 0;
2918 : }
2919 :
2920 : /*
2921 : * File creation. Allocate an inode, and we're done..
2922 : */
2923 : static int
2924 : shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
2925 : struct dentry *dentry, umode_t mode, dev_t dev)
2926 : {
2927 : struct inode *inode;
2928 : int error = -ENOSPC;
2929 :
2930 : inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
2931 : if (inode) {
2932 : error = simple_acl_create(dir, inode);
2933 : if (error)
2934 : goto out_iput;
2935 : error = security_inode_init_security(inode, dir,
2936 : &dentry->d_name,
2937 : shmem_initxattrs, NULL);
2938 : if (error && error != -EOPNOTSUPP)
2939 : goto out_iput;
2940 :
2941 : error = 0;
2942 : dir->i_size += BOGO_DIRENT_SIZE;
2943 : dir->i_ctime = dir->i_mtime = current_time(dir);
2944 : inode_inc_iversion(dir);
2945 : d_instantiate(dentry, inode);
2946 : dget(dentry); /* Extra count - pin the dentry in core */
2947 : }
2948 : return error;
2949 : out_iput:
2950 : iput(inode);
2951 : return error;
2952 : }
2953 :
2954 : static int
2955 : shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2956 : struct file *file, umode_t mode)
2957 : {
2958 : struct inode *inode;
2959 : int error = -ENOSPC;
2960 :
2961 : inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
2962 : if (inode) {
2963 : error = security_inode_init_security(inode, dir,
2964 : NULL,
2965 : shmem_initxattrs, NULL);
2966 : if (error && error != -EOPNOTSUPP)
2967 : goto out_iput;
2968 : error = simple_acl_create(dir, inode);
2969 : if (error)
2970 : goto out_iput;
2971 : d_tmpfile(file, inode);
2972 : }
2973 : return finish_open_simple(file, error);
2974 : out_iput:
2975 : iput(inode);
2976 : return error;
2977 : }
2978 :
2979 : static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2980 : struct dentry *dentry, umode_t mode)
2981 : {
2982 : int error;
2983 :
2984 : error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
2985 : if (error)
2986 : return error;
2987 : inc_nlink(dir);
2988 : return 0;
2989 : }
2990 :
2991 : static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
2992 : struct dentry *dentry, umode_t mode, bool excl)
2993 : {
2994 : return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
2995 : }
2996 :
2997 : /*
2998 : * Link a file..
2999 : */
3000 : static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3001 : {
3002 : struct inode *inode = d_inode(old_dentry);
3003 : int ret = 0;
3004 :
3005 : /*
3006 : * No ordinary (disk based) filesystem counts links as inodes;
3007 : * but each new link needs a new dentry, pinning lowmem, and
3008 : * tmpfs dentries cannot be pruned until they are unlinked.
3009 : * But if an O_TMPFILE file is linked into the tmpfs, the
3010 : * first link must skip that, to get the accounting right.
3011 : */
3012 : if (inode->i_nlink) {
3013 : ret = shmem_reserve_inode(inode->i_sb, NULL);
3014 : if (ret)
3015 : goto out;
3016 : }
3017 :
3018 : dir->i_size += BOGO_DIRENT_SIZE;
3019 : inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3020 : inode_inc_iversion(dir);
3021 : inc_nlink(inode);
3022 : ihold(inode); /* New dentry reference */
3023 : dget(dentry); /* Extra pinning count for the created dentry */
3024 : d_instantiate(dentry, inode);
3025 : out:
3026 : return ret;
3027 : }
3028 :
3029 : static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3030 : {
3031 : struct inode *inode = d_inode(dentry);
3032 :
3033 : if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3034 : shmem_free_inode(inode->i_sb);
3035 :
3036 : dir->i_size -= BOGO_DIRENT_SIZE;
3037 : inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3038 : inode_inc_iversion(dir);
3039 : drop_nlink(inode);
3040 : dput(dentry); /* Undo the count from "create" - this does all the work */
3041 : return 0;
3042 : }
3043 :
3044 : static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3045 : {
3046 : if (!simple_empty(dentry))
3047 : return -ENOTEMPTY;
3048 :
3049 : drop_nlink(d_inode(dentry));
3050 : drop_nlink(dir);
3051 : return shmem_unlink(dir, dentry);
3052 : }
3053 :
3054 : static int shmem_whiteout(struct mnt_idmap *idmap,
3055 : struct inode *old_dir, struct dentry *old_dentry)
3056 : {
3057 : struct dentry *whiteout;
3058 : int error;
3059 :
3060 : whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3061 : if (!whiteout)
3062 : return -ENOMEM;
3063 :
3064 : error = shmem_mknod(idmap, old_dir, whiteout,
3065 : S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3066 : dput(whiteout);
3067 : if (error)
3068 : return error;
3069 :
3070 : /*
3071 : * Cheat and hash the whiteout while the old dentry is still in
3072 : * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3073 : *
3074 : * d_lookup() will consistently find one of them at this point,
3075 : * not sure which one, but that isn't even important.
3076 : */
3077 : d_rehash(whiteout);
3078 : return 0;
3079 : }
3080 :
3081 : /*
3082 : * The VFS layer already does all the dentry stuff for rename,
3083 : * we just have to decrement the usage count for the target if
3084 : * it exists so that the VFS layer correctly free's it when it
3085 : * gets overwritten.
3086 : */
3087 : static int shmem_rename2(struct mnt_idmap *idmap,
3088 : struct inode *old_dir, struct dentry *old_dentry,
3089 : struct inode *new_dir, struct dentry *new_dentry,
3090 : unsigned int flags)
3091 : {
3092 : struct inode *inode = d_inode(old_dentry);
3093 : int they_are_dirs = S_ISDIR(inode->i_mode);
3094 :
3095 : if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3096 : return -EINVAL;
3097 :
3098 : if (flags & RENAME_EXCHANGE)
3099 : return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
3100 :
3101 : if (!simple_empty(new_dentry))
3102 : return -ENOTEMPTY;
3103 :
3104 : if (flags & RENAME_WHITEOUT) {
3105 : int error;
3106 :
3107 : error = shmem_whiteout(idmap, old_dir, old_dentry);
3108 : if (error)
3109 : return error;
3110 : }
3111 :
3112 : if (d_really_is_positive(new_dentry)) {
3113 : (void) shmem_unlink(new_dir, new_dentry);
3114 : if (they_are_dirs) {
3115 : drop_nlink(d_inode(new_dentry));
3116 : drop_nlink(old_dir);
3117 : }
3118 : } else if (they_are_dirs) {
3119 : drop_nlink(old_dir);
3120 : inc_nlink(new_dir);
3121 : }
3122 :
3123 : old_dir->i_size -= BOGO_DIRENT_SIZE;
3124 : new_dir->i_size += BOGO_DIRENT_SIZE;
3125 : old_dir->i_ctime = old_dir->i_mtime =
3126 : new_dir->i_ctime = new_dir->i_mtime =
3127 : inode->i_ctime = current_time(old_dir);
3128 : inode_inc_iversion(old_dir);
3129 : inode_inc_iversion(new_dir);
3130 : return 0;
3131 : }
3132 :
3133 : static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3134 : struct dentry *dentry, const char *symname)
3135 : {
3136 : int error;
3137 : int len;
3138 : struct inode *inode;
3139 : struct folio *folio;
3140 :
3141 : len = strlen(symname) + 1;
3142 : if (len > PAGE_SIZE)
3143 : return -ENAMETOOLONG;
3144 :
3145 : inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3146 : VM_NORESERVE);
3147 : if (!inode)
3148 : return -ENOSPC;
3149 :
3150 : error = security_inode_init_security(inode, dir, &dentry->d_name,
3151 : shmem_initxattrs, NULL);
3152 : if (error && error != -EOPNOTSUPP) {
3153 : iput(inode);
3154 : return error;
3155 : }
3156 :
3157 : inode->i_size = len-1;
3158 : if (len <= SHORT_SYMLINK_LEN) {
3159 : inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3160 : if (!inode->i_link) {
3161 : iput(inode);
3162 : return -ENOMEM;
3163 : }
3164 : inode->i_op = &shmem_short_symlink_operations;
3165 : } else {
3166 : inode_nohighmem(inode);
3167 : error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3168 : if (error) {
3169 : iput(inode);
3170 : return error;
3171 : }
3172 : inode->i_mapping->a_ops = &shmem_aops;
3173 : inode->i_op = &shmem_symlink_inode_operations;
3174 : memcpy(folio_address(folio), symname, len);
3175 : folio_mark_uptodate(folio);
3176 : folio_mark_dirty(folio);
3177 : folio_unlock(folio);
3178 : folio_put(folio);
3179 : }
3180 : dir->i_size += BOGO_DIRENT_SIZE;
3181 : dir->i_ctime = dir->i_mtime = current_time(dir);
3182 : inode_inc_iversion(dir);
3183 : d_instantiate(dentry, inode);
3184 : dget(dentry);
3185 : return 0;
3186 : }
3187 :
3188 : static void shmem_put_link(void *arg)
3189 : {
3190 : folio_mark_accessed(arg);
3191 : folio_put(arg);
3192 : }
3193 :
3194 : static const char *shmem_get_link(struct dentry *dentry,
3195 : struct inode *inode,
3196 : struct delayed_call *done)
3197 : {
3198 : struct folio *folio = NULL;
3199 : int error;
3200 :
3201 : if (!dentry) {
3202 : folio = filemap_get_folio(inode->i_mapping, 0);
3203 : if (!folio)
3204 : return ERR_PTR(-ECHILD);
3205 : if (PageHWPoison(folio_page(folio, 0)) ||
3206 : !folio_test_uptodate(folio)) {
3207 : folio_put(folio);
3208 : return ERR_PTR(-ECHILD);
3209 : }
3210 : } else {
3211 : error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3212 : if (error)
3213 : return ERR_PTR(error);
3214 : if (!folio)
3215 : return ERR_PTR(-ECHILD);
3216 : if (PageHWPoison(folio_page(folio, 0))) {
3217 : folio_unlock(folio);
3218 : folio_put(folio);
3219 : return ERR_PTR(-ECHILD);
3220 : }
3221 : folio_unlock(folio);
3222 : }
3223 : set_delayed_call(done, shmem_put_link, folio);
3224 : return folio_address(folio);
3225 : }
3226 :
3227 : #ifdef CONFIG_TMPFS_XATTR
3228 :
3229 : static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3230 : {
3231 : struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3232 :
3233 : fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3234 :
3235 : return 0;
3236 : }
3237 :
3238 : static int shmem_fileattr_set(struct mnt_idmap *idmap,
3239 : struct dentry *dentry, struct fileattr *fa)
3240 : {
3241 : struct inode *inode = d_inode(dentry);
3242 : struct shmem_inode_info *info = SHMEM_I(inode);
3243 :
3244 : if (fileattr_has_fsx(fa))
3245 : return -EOPNOTSUPP;
3246 : if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3247 : return -EOPNOTSUPP;
3248 :
3249 : info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3250 : (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3251 :
3252 : shmem_set_inode_flags(inode, info->fsflags);
3253 : inode->i_ctime = current_time(inode);
3254 : inode_inc_iversion(inode);
3255 : return 0;
3256 : }
3257 :
3258 : /*
3259 : * Superblocks without xattr inode operations may get some security.* xattr
3260 : * support from the LSM "for free". As soon as we have any other xattrs
3261 : * like ACLs, we also need to implement the security.* handlers at
3262 : * filesystem level, though.
3263 : */
3264 :
3265 : /*
3266 : * Callback for security_inode_init_security() for acquiring xattrs.
3267 : */
3268 : static int shmem_initxattrs(struct inode *inode,
3269 : const struct xattr *xattr_array,
3270 : void *fs_info)
3271 : {
3272 : struct shmem_inode_info *info = SHMEM_I(inode);
3273 : const struct xattr *xattr;
3274 : struct simple_xattr *new_xattr;
3275 : size_t len;
3276 :
3277 : for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3278 : new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3279 : if (!new_xattr)
3280 : return -ENOMEM;
3281 :
3282 : len = strlen(xattr->name) + 1;
3283 : new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3284 : GFP_KERNEL);
3285 : if (!new_xattr->name) {
3286 : kvfree(new_xattr);
3287 : return -ENOMEM;
3288 : }
3289 :
3290 : memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3291 : XATTR_SECURITY_PREFIX_LEN);
3292 : memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3293 : xattr->name, len);
3294 :
3295 : simple_xattr_add(&info->xattrs, new_xattr);
3296 : }
3297 :
3298 : return 0;
3299 : }
3300 :
3301 : static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3302 : struct dentry *unused, struct inode *inode,
3303 : const char *name, void *buffer, size_t size)
3304 : {
3305 : struct shmem_inode_info *info = SHMEM_I(inode);
3306 :
3307 : name = xattr_full_name(handler, name);
3308 : return simple_xattr_get(&info->xattrs, name, buffer, size);
3309 : }
3310 :
3311 : static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3312 : struct mnt_idmap *idmap,
3313 : struct dentry *unused, struct inode *inode,
3314 : const char *name, const void *value,
3315 : size_t size, int flags)
3316 : {
3317 : struct shmem_inode_info *info = SHMEM_I(inode);
3318 : int err;
3319 :
3320 : name = xattr_full_name(handler, name);
3321 : err = simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3322 : if (!err) {
3323 : inode->i_ctime = current_time(inode);
3324 : inode_inc_iversion(inode);
3325 : }
3326 : return err;
3327 : }
3328 :
3329 : static const struct xattr_handler shmem_security_xattr_handler = {
3330 : .prefix = XATTR_SECURITY_PREFIX,
3331 : .get = shmem_xattr_handler_get,
3332 : .set = shmem_xattr_handler_set,
3333 : };
3334 :
3335 : static const struct xattr_handler shmem_trusted_xattr_handler = {
3336 : .prefix = XATTR_TRUSTED_PREFIX,
3337 : .get = shmem_xattr_handler_get,
3338 : .set = shmem_xattr_handler_set,
3339 : };
3340 :
3341 : static const struct xattr_handler *shmem_xattr_handlers[] = {
3342 : #ifdef CONFIG_TMPFS_POSIX_ACL
3343 : &posix_acl_access_xattr_handler,
3344 : &posix_acl_default_xattr_handler,
3345 : #endif
3346 : &shmem_security_xattr_handler,
3347 : &shmem_trusted_xattr_handler,
3348 : NULL
3349 : };
3350 :
3351 : static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3352 : {
3353 : struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3354 : return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3355 : }
3356 : #endif /* CONFIG_TMPFS_XATTR */
3357 :
3358 : static const struct inode_operations shmem_short_symlink_operations = {
3359 : .getattr = shmem_getattr,
3360 : .get_link = simple_get_link,
3361 : #ifdef CONFIG_TMPFS_XATTR
3362 : .listxattr = shmem_listxattr,
3363 : #endif
3364 : };
3365 :
3366 : static const struct inode_operations shmem_symlink_inode_operations = {
3367 : .getattr = shmem_getattr,
3368 : .get_link = shmem_get_link,
3369 : #ifdef CONFIG_TMPFS_XATTR
3370 : .listxattr = shmem_listxattr,
3371 : #endif
3372 : };
3373 :
3374 : static struct dentry *shmem_get_parent(struct dentry *child)
3375 : {
3376 : return ERR_PTR(-ESTALE);
3377 : }
3378 :
3379 : static int shmem_match(struct inode *ino, void *vfh)
3380 : {
3381 : __u32 *fh = vfh;
3382 : __u64 inum = fh[2];
3383 : inum = (inum << 32) | fh[1];
3384 : return ino->i_ino == inum && fh[0] == ino->i_generation;
3385 : }
3386 :
3387 : /* Find any alias of inode, but prefer a hashed alias */
3388 : static struct dentry *shmem_find_alias(struct inode *inode)
3389 : {
3390 : struct dentry *alias = d_find_alias(inode);
3391 :
3392 : return alias ?: d_find_any_alias(inode);
3393 : }
3394 :
3395 :
3396 : static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3397 : struct fid *fid, int fh_len, int fh_type)
3398 : {
3399 : struct inode *inode;
3400 : struct dentry *dentry = NULL;
3401 : u64 inum;
3402 :
3403 : if (fh_len < 3)
3404 : return NULL;
3405 :
3406 : inum = fid->raw[2];
3407 : inum = (inum << 32) | fid->raw[1];
3408 :
3409 : inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3410 : shmem_match, fid->raw);
3411 : if (inode) {
3412 : dentry = shmem_find_alias(inode);
3413 : iput(inode);
3414 : }
3415 :
3416 : return dentry;
3417 : }
3418 :
3419 : static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3420 : struct inode *parent)
3421 : {
3422 : if (*len < 3) {
3423 : *len = 3;
3424 : return FILEID_INVALID;
3425 : }
3426 :
3427 : if (inode_unhashed(inode)) {
3428 : /* Unfortunately insert_inode_hash is not idempotent,
3429 : * so as we hash inodes here rather than at creation
3430 : * time, we need a lock to ensure we only try
3431 : * to do it once
3432 : */
3433 : static DEFINE_SPINLOCK(lock);
3434 : spin_lock(&lock);
3435 : if (inode_unhashed(inode))
3436 : __insert_inode_hash(inode,
3437 : inode->i_ino + inode->i_generation);
3438 : spin_unlock(&lock);
3439 : }
3440 :
3441 : fh[0] = inode->i_generation;
3442 : fh[1] = inode->i_ino;
3443 : fh[2] = ((__u64)inode->i_ino) >> 32;
3444 :
3445 : *len = 3;
3446 : return 1;
3447 : }
3448 :
3449 : static const struct export_operations shmem_export_ops = {
3450 : .get_parent = shmem_get_parent,
3451 : .encode_fh = shmem_encode_fh,
3452 : .fh_to_dentry = shmem_fh_to_dentry,
3453 : };
3454 :
3455 : enum shmem_param {
3456 : Opt_gid,
3457 : Opt_huge,
3458 : Opt_mode,
3459 : Opt_mpol,
3460 : Opt_nr_blocks,
3461 : Opt_nr_inodes,
3462 : Opt_size,
3463 : Opt_uid,
3464 : Opt_inode32,
3465 : Opt_inode64,
3466 : };
3467 :
3468 : static const struct constant_table shmem_param_enums_huge[] = {
3469 : {"never", SHMEM_HUGE_NEVER },
3470 : {"always", SHMEM_HUGE_ALWAYS },
3471 : {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3472 : {"advise", SHMEM_HUGE_ADVISE },
3473 : {}
3474 : };
3475 :
3476 : const struct fs_parameter_spec shmem_fs_parameters[] = {
3477 : fsparam_u32 ("gid", Opt_gid),
3478 : fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3479 : fsparam_u32oct("mode", Opt_mode),
3480 : fsparam_string("mpol", Opt_mpol),
3481 : fsparam_string("nr_blocks", Opt_nr_blocks),
3482 : fsparam_string("nr_inodes", Opt_nr_inodes),
3483 : fsparam_string("size", Opt_size),
3484 : fsparam_u32 ("uid", Opt_uid),
3485 : fsparam_flag ("inode32", Opt_inode32),
3486 : fsparam_flag ("inode64", Opt_inode64),
3487 : {}
3488 : };
3489 :
3490 : static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3491 : {
3492 : struct shmem_options *ctx = fc->fs_private;
3493 : struct fs_parse_result result;
3494 : unsigned long long size;
3495 : char *rest;
3496 : int opt;
3497 :
3498 : opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3499 : if (opt < 0)
3500 : return opt;
3501 :
3502 : switch (opt) {
3503 : case Opt_size:
3504 : size = memparse(param->string, &rest);
3505 : if (*rest == '%') {
3506 : size <<= PAGE_SHIFT;
3507 : size *= totalram_pages();
3508 : do_div(size, 100);
3509 : rest++;
3510 : }
3511 : if (*rest)
3512 : goto bad_value;
3513 : ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3514 : ctx->seen |= SHMEM_SEEN_BLOCKS;
3515 : break;
3516 : case Opt_nr_blocks:
3517 : ctx->blocks = memparse(param->string, &rest);
3518 : if (*rest || ctx->blocks > S64_MAX)
3519 : goto bad_value;
3520 : ctx->seen |= SHMEM_SEEN_BLOCKS;
3521 : break;
3522 : case Opt_nr_inodes:
3523 : ctx->inodes = memparse(param->string, &rest);
3524 : if (*rest)
3525 : goto bad_value;
3526 : ctx->seen |= SHMEM_SEEN_INODES;
3527 : break;
3528 : case Opt_mode:
3529 : ctx->mode = result.uint_32 & 07777;
3530 : break;
3531 : case Opt_uid:
3532 : ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3533 : if (!uid_valid(ctx->uid))
3534 : goto bad_value;
3535 : break;
3536 : case Opt_gid:
3537 : ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3538 : if (!gid_valid(ctx->gid))
3539 : goto bad_value;
3540 : break;
3541 : case Opt_huge:
3542 : ctx->huge = result.uint_32;
3543 : if (ctx->huge != SHMEM_HUGE_NEVER &&
3544 : !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3545 : has_transparent_hugepage()))
3546 : goto unsupported_parameter;
3547 : ctx->seen |= SHMEM_SEEN_HUGE;
3548 : break;
3549 : case Opt_mpol:
3550 : if (IS_ENABLED(CONFIG_NUMA)) {
3551 : mpol_put(ctx->mpol);
3552 : ctx->mpol = NULL;
3553 : if (mpol_parse_str(param->string, &ctx->mpol))
3554 : goto bad_value;
3555 : break;
3556 : }
3557 : goto unsupported_parameter;
3558 : case Opt_inode32:
3559 : ctx->full_inums = false;
3560 : ctx->seen |= SHMEM_SEEN_INUMS;
3561 : break;
3562 : case Opt_inode64:
3563 : if (sizeof(ino_t) < 8) {
3564 : return invalfc(fc,
3565 : "Cannot use inode64 with <64bit inums in kernel\n");
3566 : }
3567 : ctx->full_inums = true;
3568 : ctx->seen |= SHMEM_SEEN_INUMS;
3569 : break;
3570 : }
3571 : return 0;
3572 :
3573 : unsupported_parameter:
3574 : return invalfc(fc, "Unsupported parameter '%s'", param->key);
3575 : bad_value:
3576 : return invalfc(fc, "Bad value for '%s'", param->key);
3577 : }
3578 :
3579 : static int shmem_parse_options(struct fs_context *fc, void *data)
3580 : {
3581 : char *options = data;
3582 :
3583 : if (options) {
3584 : int err = security_sb_eat_lsm_opts(options, &fc->security);
3585 : if (err)
3586 : return err;
3587 : }
3588 :
3589 : while (options != NULL) {
3590 : char *this_char = options;
3591 : for (;;) {
3592 : /*
3593 : * NUL-terminate this option: unfortunately,
3594 : * mount options form a comma-separated list,
3595 : * but mpol's nodelist may also contain commas.
3596 : */
3597 : options = strchr(options, ',');
3598 : if (options == NULL)
3599 : break;
3600 : options++;
3601 : if (!isdigit(*options)) {
3602 : options[-1] = '\0';
3603 : break;
3604 : }
3605 : }
3606 : if (*this_char) {
3607 : char *value = strchr(this_char, '=');
3608 : size_t len = 0;
3609 : int err;
3610 :
3611 : if (value) {
3612 : *value++ = '\0';
3613 : len = strlen(value);
3614 : }
3615 : err = vfs_parse_fs_string(fc, this_char, value, len);
3616 : if (err < 0)
3617 : return err;
3618 : }
3619 : }
3620 : return 0;
3621 : }
3622 :
3623 : /*
3624 : * Reconfigure a shmem filesystem.
3625 : *
3626 : * Note that we disallow change from limited->unlimited blocks/inodes while any
3627 : * are in use; but we must separately disallow unlimited->limited, because in
3628 : * that case we have no record of how much is already in use.
3629 : */
3630 : static int shmem_reconfigure(struct fs_context *fc)
3631 : {
3632 : struct shmem_options *ctx = fc->fs_private;
3633 : struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3634 : unsigned long inodes;
3635 : struct mempolicy *mpol = NULL;
3636 : const char *err;
3637 :
3638 : raw_spin_lock(&sbinfo->stat_lock);
3639 : inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3640 :
3641 : if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3642 : if (!sbinfo->max_blocks) {
3643 : err = "Cannot retroactively limit size";
3644 : goto out;
3645 : }
3646 : if (percpu_counter_compare(&sbinfo->used_blocks,
3647 : ctx->blocks) > 0) {
3648 : err = "Too small a size for current use";
3649 : goto out;
3650 : }
3651 : }
3652 : if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3653 : if (!sbinfo->max_inodes) {
3654 : err = "Cannot retroactively limit inodes";
3655 : goto out;
3656 : }
3657 : if (ctx->inodes < inodes) {
3658 : err = "Too few inodes for current use";
3659 : goto out;
3660 : }
3661 : }
3662 :
3663 : if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3664 : sbinfo->next_ino > UINT_MAX) {
3665 : err = "Current inum too high to switch to 32-bit inums";
3666 : goto out;
3667 : }
3668 :
3669 : if (ctx->seen & SHMEM_SEEN_HUGE)
3670 : sbinfo->huge = ctx->huge;
3671 : if (ctx->seen & SHMEM_SEEN_INUMS)
3672 : sbinfo->full_inums = ctx->full_inums;
3673 : if (ctx->seen & SHMEM_SEEN_BLOCKS)
3674 : sbinfo->max_blocks = ctx->blocks;
3675 : if (ctx->seen & SHMEM_SEEN_INODES) {
3676 : sbinfo->max_inodes = ctx->inodes;
3677 : sbinfo->free_inodes = ctx->inodes - inodes;
3678 : }
3679 :
3680 : /*
3681 : * Preserve previous mempolicy unless mpol remount option was specified.
3682 : */
3683 : if (ctx->mpol) {
3684 : mpol = sbinfo->mpol;
3685 : sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3686 : ctx->mpol = NULL;
3687 : }
3688 : raw_spin_unlock(&sbinfo->stat_lock);
3689 : mpol_put(mpol);
3690 : return 0;
3691 : out:
3692 : raw_spin_unlock(&sbinfo->stat_lock);
3693 : return invalfc(fc, "%s", err);
3694 : }
3695 :
3696 : static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3697 : {
3698 : struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3699 :
3700 : if (sbinfo->max_blocks != shmem_default_max_blocks())
3701 : seq_printf(seq, ",size=%luk",
3702 : sbinfo->max_blocks << (PAGE_SHIFT - 10));
3703 : if (sbinfo->max_inodes != shmem_default_max_inodes())
3704 : seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3705 : if (sbinfo->mode != (0777 | S_ISVTX))
3706 : seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3707 : if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3708 : seq_printf(seq, ",uid=%u",
3709 : from_kuid_munged(&init_user_ns, sbinfo->uid));
3710 : if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3711 : seq_printf(seq, ",gid=%u",
3712 : from_kgid_munged(&init_user_ns, sbinfo->gid));
3713 :
3714 : /*
3715 : * Showing inode{64,32} might be useful even if it's the system default,
3716 : * since then people don't have to resort to checking both here and
3717 : * /proc/config.gz to confirm 64-bit inums were successfully applied
3718 : * (which may not even exist if IKCONFIG_PROC isn't enabled).
3719 : *
3720 : * We hide it when inode64 isn't the default and we are using 32-bit
3721 : * inodes, since that probably just means the feature isn't even under
3722 : * consideration.
3723 : *
3724 : * As such:
3725 : *
3726 : * +-----------------+-----------------+
3727 : * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3728 : * +------------------+-----------------+-----------------+
3729 : * | full_inums=true | show | show |
3730 : * | full_inums=false | show | hide |
3731 : * +------------------+-----------------+-----------------+
3732 : *
3733 : */
3734 : if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3735 : seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3736 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3737 : /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3738 : if (sbinfo->huge)
3739 : seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3740 : #endif
3741 : shmem_show_mpol(seq, sbinfo->mpol);
3742 : return 0;
3743 : }
3744 :
3745 : #endif /* CONFIG_TMPFS */
3746 :
3747 0 : static void shmem_put_super(struct super_block *sb)
3748 : {
3749 0 : struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3750 :
3751 0 : free_percpu(sbinfo->ino_batch);
3752 0 : percpu_counter_destroy(&sbinfo->used_blocks);
3753 0 : mpol_put(sbinfo->mpol);
3754 0 : kfree(sbinfo);
3755 0 : sb->s_fs_info = NULL;
3756 0 : }
3757 :
3758 1 : static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3759 : {
3760 1 : struct shmem_options *ctx = fc->fs_private;
3761 : struct inode *inode;
3762 : struct shmem_sb_info *sbinfo;
3763 :
3764 : /* Round up to L1_CACHE_BYTES to resist false sharing */
3765 1 : sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3766 : L1_CACHE_BYTES), GFP_KERNEL);
3767 1 : if (!sbinfo)
3768 : return -ENOMEM;
3769 :
3770 1 : sb->s_fs_info = sbinfo;
3771 :
3772 : #ifdef CONFIG_TMPFS
3773 : /*
3774 : * Per default we only allow half of the physical ram per
3775 : * tmpfs instance, limiting inodes to one per page of lowmem;
3776 : * but the internal instance is left unlimited.
3777 : */
3778 : if (!(sb->s_flags & SB_KERNMOUNT)) {
3779 : if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3780 : ctx->blocks = shmem_default_max_blocks();
3781 : if (!(ctx->seen & SHMEM_SEEN_INODES))
3782 : ctx->inodes = shmem_default_max_inodes();
3783 : if (!(ctx->seen & SHMEM_SEEN_INUMS))
3784 : ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3785 : } else {
3786 : sb->s_flags |= SB_NOUSER;
3787 : }
3788 : sb->s_export_op = &shmem_export_ops;
3789 : sb->s_flags |= SB_NOSEC | SB_I_VERSION;
3790 : #else
3791 1 : sb->s_flags |= SB_NOUSER;
3792 : #endif
3793 1 : sbinfo->max_blocks = ctx->blocks;
3794 1 : sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3795 1 : if (sb->s_flags & SB_KERNMOUNT) {
3796 1 : sbinfo->ino_batch = alloc_percpu(ino_t);
3797 1 : if (!sbinfo->ino_batch)
3798 : goto failed;
3799 : }
3800 1 : sbinfo->uid = ctx->uid;
3801 1 : sbinfo->gid = ctx->gid;
3802 1 : sbinfo->full_inums = ctx->full_inums;
3803 1 : sbinfo->mode = ctx->mode;
3804 1 : sbinfo->huge = ctx->huge;
3805 1 : sbinfo->mpol = ctx->mpol;
3806 1 : ctx->mpol = NULL;
3807 :
3808 : raw_spin_lock_init(&sbinfo->stat_lock);
3809 2 : if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3810 : goto failed;
3811 1 : spin_lock_init(&sbinfo->shrinklist_lock);
3812 2 : INIT_LIST_HEAD(&sbinfo->shrinklist);
3813 :
3814 1 : sb->s_maxbytes = MAX_LFS_FILESIZE;
3815 1 : sb->s_blocksize = PAGE_SIZE;
3816 1 : sb->s_blocksize_bits = PAGE_SHIFT;
3817 1 : sb->s_magic = TMPFS_MAGIC;
3818 1 : sb->s_op = &shmem_ops;
3819 1 : sb->s_time_gran = 1;
3820 : #ifdef CONFIG_TMPFS_XATTR
3821 : sb->s_xattr = shmem_xattr_handlers;
3822 : #endif
3823 : #ifdef CONFIG_TMPFS_POSIX_ACL
3824 : sb->s_flags |= SB_POSIXACL;
3825 : #endif
3826 1 : uuid_gen(&sb->s_uuid);
3827 :
3828 1 : inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
3829 : VM_NORESERVE);
3830 1 : if (!inode)
3831 : goto failed;
3832 1 : inode->i_uid = sbinfo->uid;
3833 1 : inode->i_gid = sbinfo->gid;
3834 1 : sb->s_root = d_make_root(inode);
3835 1 : if (!sb->s_root)
3836 : goto failed;
3837 : return 0;
3838 :
3839 : failed:
3840 0 : shmem_put_super(sb);
3841 0 : return -ENOMEM;
3842 : }
3843 :
3844 1 : static int shmem_get_tree(struct fs_context *fc)
3845 : {
3846 1 : return get_tree_nodev(fc, shmem_fill_super);
3847 : }
3848 :
3849 1 : static void shmem_free_fc(struct fs_context *fc)
3850 : {
3851 1 : struct shmem_options *ctx = fc->fs_private;
3852 :
3853 1 : if (ctx) {
3854 1 : mpol_put(ctx->mpol);
3855 1 : kfree(ctx);
3856 : }
3857 1 : }
3858 :
3859 : static const struct fs_context_operations shmem_fs_context_ops = {
3860 : .free = shmem_free_fc,
3861 : .get_tree = shmem_get_tree,
3862 : #ifdef CONFIG_TMPFS
3863 : .parse_monolithic = shmem_parse_options,
3864 : .parse_param = shmem_parse_one,
3865 : .reconfigure = shmem_reconfigure,
3866 : #endif
3867 : };
3868 :
3869 : static struct kmem_cache *shmem_inode_cachep;
3870 :
3871 1 : static struct inode *shmem_alloc_inode(struct super_block *sb)
3872 : {
3873 : struct shmem_inode_info *info;
3874 2 : info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
3875 1 : if (!info)
3876 : return NULL;
3877 1 : return &info->vfs_inode;
3878 : }
3879 :
3880 0 : static void shmem_free_in_core_inode(struct inode *inode)
3881 : {
3882 0 : if (S_ISLNK(inode->i_mode))
3883 0 : kfree(inode->i_link);
3884 0 : kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3885 0 : }
3886 :
3887 0 : static void shmem_destroy_inode(struct inode *inode)
3888 : {
3889 : if (S_ISREG(inode->i_mode))
3890 : mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3891 0 : }
3892 :
3893 12 : static void shmem_init_inode(void *foo)
3894 : {
3895 12 : struct shmem_inode_info *info = foo;
3896 12 : inode_init_once(&info->vfs_inode);
3897 12 : }
3898 :
3899 : static void shmem_init_inodecache(void)
3900 : {
3901 1 : shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3902 : sizeof(struct shmem_inode_info),
3903 : 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3904 : }
3905 :
3906 : static void shmem_destroy_inodecache(void)
3907 : {
3908 0 : kmem_cache_destroy(shmem_inode_cachep);
3909 : }
3910 :
3911 : /* Keep the page in page cache instead of truncating it */
3912 0 : static int shmem_error_remove_page(struct address_space *mapping,
3913 : struct page *page)
3914 : {
3915 0 : return 0;
3916 : }
3917 :
3918 : const struct address_space_operations shmem_aops = {
3919 : .writepage = shmem_writepage,
3920 : .dirty_folio = noop_dirty_folio,
3921 : #ifdef CONFIG_TMPFS
3922 : .write_begin = shmem_write_begin,
3923 : .write_end = shmem_write_end,
3924 : #endif
3925 : #ifdef CONFIG_MIGRATION
3926 : .migrate_folio = migrate_folio,
3927 : #endif
3928 : .error_remove_page = shmem_error_remove_page,
3929 : };
3930 : EXPORT_SYMBOL(shmem_aops);
3931 :
3932 : static const struct file_operations shmem_file_operations = {
3933 : .mmap = shmem_mmap,
3934 : .open = generic_file_open,
3935 : .get_unmapped_area = shmem_get_unmapped_area,
3936 : #ifdef CONFIG_TMPFS
3937 : .llseek = shmem_file_llseek,
3938 : .read_iter = shmem_file_read_iter,
3939 : .write_iter = generic_file_write_iter,
3940 : .fsync = noop_fsync,
3941 : .splice_read = generic_file_splice_read,
3942 : .splice_write = iter_file_splice_write,
3943 : .fallocate = shmem_fallocate,
3944 : #endif
3945 : };
3946 :
3947 : static const struct inode_operations shmem_inode_operations = {
3948 : .getattr = shmem_getattr,
3949 : .setattr = shmem_setattr,
3950 : #ifdef CONFIG_TMPFS_XATTR
3951 : .listxattr = shmem_listxattr,
3952 : .set_acl = simple_set_acl,
3953 : .fileattr_get = shmem_fileattr_get,
3954 : .fileattr_set = shmem_fileattr_set,
3955 : #endif
3956 : };
3957 :
3958 : static const struct inode_operations shmem_dir_inode_operations = {
3959 : #ifdef CONFIG_TMPFS
3960 : .getattr = shmem_getattr,
3961 : .create = shmem_create,
3962 : .lookup = simple_lookup,
3963 : .link = shmem_link,
3964 : .unlink = shmem_unlink,
3965 : .symlink = shmem_symlink,
3966 : .mkdir = shmem_mkdir,
3967 : .rmdir = shmem_rmdir,
3968 : .mknod = shmem_mknod,
3969 : .rename = shmem_rename2,
3970 : .tmpfile = shmem_tmpfile,
3971 : #endif
3972 : #ifdef CONFIG_TMPFS_XATTR
3973 : .listxattr = shmem_listxattr,
3974 : .fileattr_get = shmem_fileattr_get,
3975 : .fileattr_set = shmem_fileattr_set,
3976 : #endif
3977 : #ifdef CONFIG_TMPFS_POSIX_ACL
3978 : .setattr = shmem_setattr,
3979 : .set_acl = simple_set_acl,
3980 : #endif
3981 : };
3982 :
3983 : static const struct inode_operations shmem_special_inode_operations = {
3984 : .getattr = shmem_getattr,
3985 : #ifdef CONFIG_TMPFS_XATTR
3986 : .listxattr = shmem_listxattr,
3987 : #endif
3988 : #ifdef CONFIG_TMPFS_POSIX_ACL
3989 : .setattr = shmem_setattr,
3990 : .set_acl = simple_set_acl,
3991 : #endif
3992 : };
3993 :
3994 : static const struct super_operations shmem_ops = {
3995 : .alloc_inode = shmem_alloc_inode,
3996 : .free_inode = shmem_free_in_core_inode,
3997 : .destroy_inode = shmem_destroy_inode,
3998 : #ifdef CONFIG_TMPFS
3999 : .statfs = shmem_statfs,
4000 : .show_options = shmem_show_options,
4001 : #endif
4002 : .evict_inode = shmem_evict_inode,
4003 : .drop_inode = generic_delete_inode,
4004 : .put_super = shmem_put_super,
4005 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4006 : .nr_cached_objects = shmem_unused_huge_count,
4007 : .free_cached_objects = shmem_unused_huge_scan,
4008 : #endif
4009 : };
4010 :
4011 : static const struct vm_operations_struct shmem_vm_ops = {
4012 : .fault = shmem_fault,
4013 : .map_pages = filemap_map_pages,
4014 : #ifdef CONFIG_NUMA
4015 : .set_policy = shmem_set_policy,
4016 : .get_policy = shmem_get_policy,
4017 : #endif
4018 : };
4019 :
4020 : static const struct vm_operations_struct shmem_anon_vm_ops = {
4021 : .fault = shmem_fault,
4022 : .map_pages = filemap_map_pages,
4023 : #ifdef CONFIG_NUMA
4024 : .set_policy = shmem_set_policy,
4025 : .get_policy = shmem_get_policy,
4026 : #endif
4027 : };
4028 :
4029 1 : int shmem_init_fs_context(struct fs_context *fc)
4030 : {
4031 : struct shmem_options *ctx;
4032 :
4033 1 : ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4034 1 : if (!ctx)
4035 : return -ENOMEM;
4036 :
4037 1 : ctx->mode = 0777 | S_ISVTX;
4038 1 : ctx->uid = current_fsuid();
4039 1 : ctx->gid = current_fsgid();
4040 :
4041 1 : fc->fs_private = ctx;
4042 1 : fc->ops = &shmem_fs_context_ops;
4043 1 : return 0;
4044 : }
4045 :
4046 : static struct file_system_type shmem_fs_type = {
4047 : .owner = THIS_MODULE,
4048 : .name = "tmpfs",
4049 : .init_fs_context = shmem_init_fs_context,
4050 : #ifdef CONFIG_TMPFS
4051 : .parameters = shmem_fs_parameters,
4052 : #endif
4053 : .kill_sb = kill_litter_super,
4054 : #ifdef CONFIG_SHMEM
4055 : .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4056 : #else
4057 : .fs_flags = FS_USERNS_MOUNT,
4058 : #endif
4059 : };
4060 :
4061 1 : void __init shmem_init(void)
4062 : {
4063 : int error;
4064 :
4065 : shmem_init_inodecache();
4066 :
4067 1 : error = register_filesystem(&shmem_fs_type);
4068 1 : if (error) {
4069 0 : pr_err("Could not register tmpfs\n");
4070 0 : goto out2;
4071 : }
4072 :
4073 1 : shm_mnt = kern_mount(&shmem_fs_type);
4074 2 : if (IS_ERR(shm_mnt)) {
4075 0 : error = PTR_ERR(shm_mnt);
4076 0 : pr_err("Could not kern_mount tmpfs\n");
4077 : goto out1;
4078 : }
4079 :
4080 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4081 : if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4082 : SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4083 : else
4084 : shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4085 : #endif
4086 : return;
4087 :
4088 : out1:
4089 0 : unregister_filesystem(&shmem_fs_type);
4090 : out2:
4091 : shmem_destroy_inodecache();
4092 0 : shm_mnt = ERR_PTR(error);
4093 : }
4094 :
4095 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4096 : static ssize_t shmem_enabled_show(struct kobject *kobj,
4097 : struct kobj_attribute *attr, char *buf)
4098 : {
4099 : static const int values[] = {
4100 : SHMEM_HUGE_ALWAYS,
4101 : SHMEM_HUGE_WITHIN_SIZE,
4102 : SHMEM_HUGE_ADVISE,
4103 : SHMEM_HUGE_NEVER,
4104 : SHMEM_HUGE_DENY,
4105 : SHMEM_HUGE_FORCE,
4106 : };
4107 : int len = 0;
4108 : int i;
4109 :
4110 : for (i = 0; i < ARRAY_SIZE(values); i++) {
4111 : len += sysfs_emit_at(buf, len,
4112 : shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4113 : i ? " " : "",
4114 : shmem_format_huge(values[i]));
4115 : }
4116 :
4117 : len += sysfs_emit_at(buf, len, "\n");
4118 :
4119 : return len;
4120 : }
4121 :
4122 : static ssize_t shmem_enabled_store(struct kobject *kobj,
4123 : struct kobj_attribute *attr, const char *buf, size_t count)
4124 : {
4125 : char tmp[16];
4126 : int huge;
4127 :
4128 : if (count + 1 > sizeof(tmp))
4129 : return -EINVAL;
4130 : memcpy(tmp, buf, count);
4131 : tmp[count] = '\0';
4132 : if (count && tmp[count - 1] == '\n')
4133 : tmp[count - 1] = '\0';
4134 :
4135 : huge = shmem_parse_huge(tmp);
4136 : if (huge == -EINVAL)
4137 : return -EINVAL;
4138 : if (!has_transparent_hugepage() &&
4139 : huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4140 : return -EINVAL;
4141 :
4142 : shmem_huge = huge;
4143 : if (shmem_huge > SHMEM_HUGE_DENY)
4144 : SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4145 : return count;
4146 : }
4147 :
4148 : struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4149 : #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4150 :
4151 : #else /* !CONFIG_SHMEM */
4152 :
4153 : /*
4154 : * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4155 : *
4156 : * This is intended for small system where the benefits of the full
4157 : * shmem code (swap-backed and resource-limited) are outweighed by
4158 : * their complexity. On systems without swap this code should be
4159 : * effectively equivalent, but much lighter weight.
4160 : */
4161 :
4162 : static struct file_system_type shmem_fs_type = {
4163 : .name = "tmpfs",
4164 : .init_fs_context = ramfs_init_fs_context,
4165 : .parameters = ramfs_fs_parameters,
4166 : .kill_sb = kill_litter_super,
4167 : .fs_flags = FS_USERNS_MOUNT,
4168 : };
4169 :
4170 : void __init shmem_init(void)
4171 : {
4172 : BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4173 :
4174 : shm_mnt = kern_mount(&shmem_fs_type);
4175 : BUG_ON(IS_ERR(shm_mnt));
4176 : }
4177 :
4178 : int shmem_unuse(unsigned int type)
4179 : {
4180 : return 0;
4181 : }
4182 :
4183 : int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4184 : {
4185 : return 0;
4186 : }
4187 :
4188 : void shmem_unlock_mapping(struct address_space *mapping)
4189 : {
4190 : }
4191 :
4192 : #ifdef CONFIG_MMU
4193 : unsigned long shmem_get_unmapped_area(struct file *file,
4194 : unsigned long addr, unsigned long len,
4195 : unsigned long pgoff, unsigned long flags)
4196 : {
4197 : return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4198 : }
4199 : #endif
4200 :
4201 : void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4202 : {
4203 : truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4204 : }
4205 : EXPORT_SYMBOL_GPL(shmem_truncate_range);
4206 :
4207 : #define shmem_vm_ops generic_file_vm_ops
4208 : #define shmem_anon_vm_ops generic_file_vm_ops
4209 : #define shmem_file_operations ramfs_file_operations
4210 : #define shmem_get_inode(idmap, sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4211 : #define shmem_acct_size(flags, size) 0
4212 : #define shmem_unacct_size(flags, size) do {} while (0)
4213 :
4214 : #endif /* CONFIG_SHMEM */
4215 :
4216 : /* common code */
4217 :
4218 0 : static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4219 : unsigned long flags, unsigned int i_flags)
4220 : {
4221 : struct inode *inode;
4222 : struct file *res;
4223 :
4224 0 : if (IS_ERR(mnt))
4225 : return ERR_CAST(mnt);
4226 :
4227 0 : if (size < 0 || size > MAX_LFS_FILESIZE)
4228 : return ERR_PTR(-EINVAL);
4229 :
4230 0 : if (shmem_acct_size(flags, size))
4231 : return ERR_PTR(-ENOMEM);
4232 :
4233 0 : if (is_idmapped_mnt(mnt))
4234 : return ERR_PTR(-EINVAL);
4235 :
4236 0 : inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4237 : S_IFREG | S_IRWXUGO, 0, flags);
4238 0 : if (unlikely(!inode)) {
4239 : shmem_unacct_size(flags, size);
4240 : return ERR_PTR(-ENOSPC);
4241 : }
4242 0 : inode->i_flags |= i_flags;
4243 0 : inode->i_size = size;
4244 0 : clear_nlink(inode); /* It is unlinked */
4245 0 : res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4246 0 : if (!IS_ERR(res))
4247 0 : res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4248 : &shmem_file_operations);
4249 0 : if (IS_ERR(res))
4250 0 : iput(inode);
4251 : return res;
4252 : }
4253 :
4254 : /**
4255 : * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4256 : * kernel internal. There will be NO LSM permission checks against the
4257 : * underlying inode. So users of this interface must do LSM checks at a
4258 : * higher layer. The users are the big_key and shm implementations. LSM
4259 : * checks are provided at the key or shm level rather than the inode.
4260 : * @name: name for dentry (to be seen in /proc/<pid>/maps
4261 : * @size: size to be set for the file
4262 : * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4263 : */
4264 0 : struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4265 : {
4266 0 : return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4267 : }
4268 :
4269 : /**
4270 : * shmem_file_setup - get an unlinked file living in tmpfs
4271 : * @name: name for dentry (to be seen in /proc/<pid>/maps
4272 : * @size: size to be set for the file
4273 : * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4274 : */
4275 0 : struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4276 : {
4277 0 : return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4278 : }
4279 : EXPORT_SYMBOL_GPL(shmem_file_setup);
4280 :
4281 : /**
4282 : * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4283 : * @mnt: the tmpfs mount where the file will be created
4284 : * @name: name for dentry (to be seen in /proc/<pid>/maps
4285 : * @size: size to be set for the file
4286 : * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4287 : */
4288 0 : struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4289 : loff_t size, unsigned long flags)
4290 : {
4291 0 : return __shmem_file_setup(mnt, name, size, flags, 0);
4292 : }
4293 : EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4294 :
4295 : /**
4296 : * shmem_zero_setup - setup a shared anonymous mapping
4297 : * @vma: the vma to be mmapped is prepared by do_mmap
4298 : */
4299 0 : int shmem_zero_setup(struct vm_area_struct *vma)
4300 : {
4301 : struct file *file;
4302 0 : loff_t size = vma->vm_end - vma->vm_start;
4303 :
4304 : /*
4305 : * Cloning a new file under mmap_lock leads to a lock ordering conflict
4306 : * between XFS directory reading and selinux: since this file is only
4307 : * accessible to the user through its mapping, use S_PRIVATE flag to
4308 : * bypass file security, in the same way as shmem_kernel_file_setup().
4309 : */
4310 0 : file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4311 0 : if (IS_ERR(file))
4312 0 : return PTR_ERR(file);
4313 :
4314 0 : if (vma->vm_file)
4315 0 : fput(vma->vm_file);
4316 0 : vma->vm_file = file;
4317 0 : vma->vm_ops = &shmem_anon_vm_ops;
4318 :
4319 0 : return 0;
4320 : }
4321 :
4322 : /**
4323 : * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4324 : * @mapping: the folio's address_space
4325 : * @index: the folio index
4326 : * @gfp: the page allocator flags to use if allocating
4327 : *
4328 : * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4329 : * with any new page allocations done using the specified allocation flags.
4330 : * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4331 : * suit tmpfs, since it may have pages in swapcache, and needs to find those
4332 : * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4333 : *
4334 : * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4335 : * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4336 : */
4337 0 : struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4338 : pgoff_t index, gfp_t gfp)
4339 : {
4340 : #ifdef CONFIG_SHMEM
4341 0 : struct inode *inode = mapping->host;
4342 : struct folio *folio;
4343 : int error;
4344 :
4345 0 : BUG_ON(!shmem_mapping(mapping));
4346 0 : error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4347 : gfp, NULL, NULL, NULL);
4348 0 : if (error)
4349 0 : return ERR_PTR(error);
4350 :
4351 0 : folio_unlock(folio);
4352 0 : return folio;
4353 : #else
4354 : /*
4355 : * The tiny !SHMEM case uses ramfs without swap
4356 : */
4357 : return mapping_read_folio_gfp(mapping, index, gfp);
4358 : #endif
4359 : }
4360 : EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4361 :
4362 0 : struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4363 : pgoff_t index, gfp_t gfp)
4364 : {
4365 0 : struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4366 : struct page *page;
4367 :
4368 0 : if (IS_ERR(folio))
4369 0 : return &folio->page;
4370 :
4371 0 : page = folio_file_page(folio, index);
4372 : if (PageHWPoison(page)) {
4373 : folio_put(folio);
4374 : return ERR_PTR(-EIO);
4375 : }
4376 :
4377 0 : return page;
4378 : }
4379 : EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
|
